Monday, July 17, 2006

AMD64 4x4 ready to frag Conroe XE by 60%

INQ reports that AMD 4x4 will boost performance by 80%. Since the 2.93 GHZ Conroe XE 6800 is about 10% faster than the 2.8GHZ Athlon 64 FX62, the 4x4 will outperform Con XE 6800+ by about 64%. In other words, Intel needs at least a Conroe at 2.93*1.6 = 4.8GHZ to compete against 4x4 at 2.8GHZ. As far I know, 4.8GHZ is unattainable with Con XE even if liquid helium is used for cooling.

As I have pointed out, 4x4 is a permanent solution to pin down Intel at 50% of AMD's performance or less by taking advantage of the Direct Connect Architecture.

119 Comments:

Anonymous Anonymous said...

Oh, man - what are you comparing? apple to oranges?
dual CPU with 2 * dual CPU?
are you mad? what is your point?
It is obvious that 2 dual CPUs will be faster than 1 dual CPU.....
don't forget to compare prices.... AMD solution will twice as expenisve as the Intel one....

2:03 PM, July 17, 2006  
Blogger Sharikou, Ph. D said...

dual CPU with 2 * dual CPU?
are you mad? what is your point?


Who cares? A computer user see a computer with specific performance and price point. Two X2 3800+ will frag a Con XE at 30% of the cost.

Conroe XE has about 300 million transistors, so do two AMD64 CPUs combined.

If Intel can bundle 10 Conroes into one package and get 10x performance, do it. But that's not possible, because each of the core will only get 1066/20 = 50MHZ of bandwidth.

2:09 PM, July 17, 2006  
Anonymous Anonymous said...

The Game Continues

The game of tit-for-tat continues. Intel has the buzz now, AMD will try to move the game back onto their field.

If they can successfully move dual-socket into the enthusiast/mainstream sector, (like Nvidia and SLI) they will shift the game.

And 4x4 looks like it should be able to play a strong game indeed.

A few million to encourage game and driver developers to do multithreading should help too.

2:11 PM, July 17, 2006  
Anonymous Anonymous said...

who cares?
A Ferrari needs 12 cylinders to be 50% faster than a Volkswagen 4 cylinder.
only thing that matters: Ferrari is faster.

2:12 PM, July 17, 2006  
Blogger Mad Mod Mike said...

"If Intel can bundle 10 Conroes into one package and get 10x performance, do it. But that's not possible, because each of the core will only get 1066/20 = 50MHZ of bandwidth."

Dual 1333MHz FSB's w/ Quad Channel FB-DIMM DDR2-667 only gets Woodcrest 5GB/s Memory Bandwidth, over a 20GB/s Bandwidth possible...and people say it isn't saturated? Right...

50MHz of bandwidth ya, and about 20MHz of Memory Bandwidth.

2:13 PM, July 17, 2006  
Anonymous jackall said...

http://www.dailytech.com/article.aspx?newsid=3361

It looks like AMD won't be making much profit for the next two quarters.

2:20 PM, July 17, 2006  
Blogger Eddie said...

Chicagraf0 here

Illustrious Sharikou, I am willing to give you a chance for your unrelenting optimism:

Could you explain for your whole audience how is it that applications are going to take advantage of the extra two cores in 4x4 to materialize the advantages of Hypertransport/DCA?

2:23 PM, July 17, 2006  
Anonymous Edward said...

4x4 is a cool concept, and it's a right direction to the future. But I believe it can give that 80% performance boost only on very few selected scenarios. For most today's applications, 4x4 perforance boost might be much less than 50%.

Anyway, I hope I am wrong... but unless we see more real results, your conclusion on 4x4 fragging Conroe is just too early.

2:28 PM, July 17, 2006  
Blogger Sharikou, Ph. D said...

Could you explain for your whole audience how is it that applications are going to take advantage of the extra two cores in 4x4 to materialize the advantages of Hypertransport/DCA?

Only a handful of people in the world can explain why and how DCA is better. The rest will just enjoy the fact that their 4x4 computer frags Conroe by 60% at lower cost.

2:38 PM, July 17, 2006  
Anonymous Anonymous said...

That is a paradigm shift. AMD shifts the rule of games from single core to dual core. Now, AMD plays the trick again, shifts the rule of game from single socket to dual-socket.

This approach has quite a few advantages. First, users can determine thier performance reuqirements by buying extra CPU. Second, since the CPU is basically the same one, the chance of having funny problems is a lot remote than a brand new CPU (see what Woodcrest has done). Third, since the cost vs performance is not linear, this approach tends to have better cost-effectiveness.

It is questions how soon AMD can pull this 4X4 out of its hat.

2:48 PM, July 17, 2006  
Anonymous Anonymous said...

You say that the the XE Conroe is about 10% faster than the FX, but you forget to say that's in your specialy picked out games. From what reviews do you get your data?

From what I've seen Conroe is a lot better in the benchmarks you do not mention, like video decoding/encoding, 3D animations rendering, photoshop. When multitasking the above Conroe is a lot faster than 10% clock per clock against the FX or any other A64 AM2.

Go pick out any review with the benchmarks mentioned above and give me a reason why 4x4 will be 60% faster than Conroe.

4x4 will make more noise and cost more and perform worse per core.

2:57 PM, July 17, 2006  
Blogger mrmaloon said...

In the benchmarks you are referring too (Cinebench and 3D Mark that I've seen 4x4 numbers quoted), Intel leads AMD (FX-62) by about 25-30% with the E6800 today.

Furthermore, if your application is not heavily multi-threaded and you aren't running vista you will not see an improvement evidently. This reeks of a platform designed to only win a couple benchmarks.

Given that Conroe has already been shown to OC to 3.4 to 3.5Ghz on air and is roughly 30% faster than FX-62 already, AMD will have a tough time justifying the cost of this system. Especially since there is no release date set for this (meaning it is 2-3 quarters away) and Intel could easily be at 3.8Ghz by that time with a cherry-picked Conroe--A 4x4 with FX-62 or FX-64 even if 70-80% better on 3 benchmarks will not beat a single Conroe at 3.8Ghz except in perhaps a couple of multi-threaded benchmarks or those dominated by FPU.

Finally, if this thing is released in 2007 you are competing against Kentsfield. At Xtremesystems forums there are many benchmarks showing the same 70-90% speedup vs. a E6600 Conroe (which is trashing the FX-62 right now). If you speed up both Intel and AMD by same amount; Intel still wins.... Forget for a moment that Kentsfield will have the same frequency headroom as Conroe (while the FX line has maybe 200Mhz left until AMD needs 65nm).

Now consider the costs of your 4x4....first you have 2 CPU's (if you use FX chips it is $2k to 2.5k today....if you don't use FX chips, you won't beat Conroe and are still around $.75 to 1k), then you have a specialized MB which won't be cheap, then you have double the TDP and associated cooling/heatsinks and the larger case it will take to hold all of this. A 4x4 system will run $2.5 to $5k....these are not mainstream systems.

Or, you could just buy a Kentsfield at about a third the price and have a 4-core solution which will handle any 4x4 configuration you could throw at it and probably take any benchmark except maybe a couple FP dominated or get a Conroe and win 85% of the benchmarks vs a 4x4 at about 20% of the price.

4x4 as I see it is dead for now. AMD needs to improve their single-core performance. Either 65nm (Q1-2-2007), K8L (Q2-2007????) are needed and have to deliver.

If AMD drops the ball with either 65nm or will K8L they could easily have the situation where the FX line is competing against 2MB Conroe Celerons with the 4MB Conroe owning the middle of the market and Kentsfield owning the top of the market. Under such a situation AMD would be in a world of hurt. The next couple quarters will determine whether AMD lives or dies.

3:00 PM, July 17, 2006  
Anonymous Anonymous said...

I'm an enthusiast who basically just wants a $150 video card and $100-$350 CPU which don't use too much power. Currently this is the AM2 family. If lots of people started buying E6600 Conroe's at $350 and there were no stability/compatibility/reliability/performance problems, I'd switch over in a heartbeat. But as it is, regular people can't buy it yet, so who knows if 4MB-cache E6600 Conroes that consumers can buy have any of these issues?

4-core GPUs and CPUs? That's not bad, but I'd wait until total power consumption came back in line with today's 1-core GPU and 2-core CPU systems before I got one for home use. For work, a 4-core CPU system wouldn't be bad: I could compile, run my co.'s software product, etc. at the same time, but work would probably only spring for a 2-core system until 4+-core systems were standard, and I bet there are similar concerns about power consumption. Probably idle power consumption should stay about the same as it is today, with 100% use of all four cores being 1.5x of today's 2-core systems, and 100% use of two cores being a little lower than the same for today's technology.

3:11 PM, July 17, 2006  
Anonymous Anonymous said...

I'd rather have a Intel Kentsfield running at 2.66GHz than two FX's running at 2.8 because the Kentsfield will be faster overall, and consume less power. And if I need even more power, I'd just overclock it.

http://www.xtremesystems.org/forums/showthread.php?t=107092

3:20 PM, July 17, 2006  
Anonymous Anonymous said...

Two X2 3800+ will frag a Con XE at 30% of the cost.

Cool! This would also mean, two cheap X2 3800+ for $300 will be about 50% faster than a FX62 for $1200.

3:22 PM, July 17, 2006  
Anonymous Anonymous said...

If Intel can bundle 10 Conroes into one package and get 10x performance, do it.

well. with the start of 4x4 kentsfield will be available. no dual cpu motherboard crap - just plug and play. :D

btw: real enthusiast buy xeon or operton!

3:40 PM, July 17, 2006  
Anonymous enumae said...

"Who cares? A computer user see a computer with specific performance and price point. Two X2 3800+ will frag a Con XE at 30% of the cost."

"The rest will just enjoy the fact that their 4x4 computer frags Conroe by 60% at lower cost."

First let me state that all information in this post is based off AnandTech's review.

I am a little suprised to see you release statements like these.

I went an ran the numbers, and the largest margin I found was 16%. Here is how I broke them down, if someone sees a fault in the way I did this please let me know...

2 - X2 3800 @ 2.0 = A2
1 - X6800 @ 2.93 = I1
1 - X6800 @ 3.22 = I2

A2 * 2 * 0.80 = 80% of A2

Quake 4 = A2 has a 16% advantage over I1, and 11% over I2.

BF2 = A2 has a 2% advantage over I1. I2 has a 3% advantage over A2.

HL2 = A2 has a 14% advantage over I1, and 9% over I2.

F.E.A.R = A2 has a 13% advantage over I1, and 8% over I2.

I have to say that I am not entirely familiar with AMD motherboards, so these maybe a little scewed.

1 AMD X2 3800 = 95 watts idle reference. For the sake of argument lets take 30% off since there will only be one motherboard.

A2 = 133 watts (idle)
I1 = 114 watts (idle)
I2 = 114 watts (idle)

1 AMD X2 3800 = 129 watts load reference.

A2 = 181 watts (load)
I1 = 172 watts (load)
I2 = 182 watts (load)

From the power point of view, AMD is very strong. I must point out that these chips are the low end AMD. To get the 60% your talking about they would have to be FX chips, and then the power really becomes an issue.

In regards to current prices, and soon to be released prices, I have a hard time believing that you could build a system for 30% of the cost, even with these chips.

3:44 PM, July 17, 2006  
Anonymous Anonymous said...

o no, Dr. sharikou starts his pre-school math again...

4:09 PM, July 17, 2006  
Anonymous Anonymous said...

what i would like to know is if amd will make 4x4 board that's compatible with my 4400 x2. also will you need to have two identical 4400 x2s or can you mismatch a 4400 x2 with a 3800 or 4200 so on and so forth.

4:11 PM, July 17, 2006  
Anonymous Anonymous said...

Mrmaloon, wondering why you're still using actual pricing even when MANY TIMES AS BEEN DISPLAYED EVERYWHERE. that AMD WILL REDUCE PRICES BY ALMOST 50% ( if not more ) on all their hardware except 1 MB per core based chips?

so stop claiming " ohh noez, its 2 x 1200 FX-64 vs 1 x 999 Conroe!!"

same with the X2's..
149 us x 2 = 300, not 300 per X2 3800+ ( 300 x 2=600 )

thus a 4x4 budged with double X2 3800's would be 300 US for the cpus.

4:29 PM, July 17, 2006  
Anonymous Anonymous said...

AMD's 4x4 is genius. Not only do they improve performance by 80% for the user who wants to add the extra processing power, they also sell more chips to these users.

I for one would love to pay an extra $25~$50 for a computer to have that extra 4x4 socket. I could easily add more horse power to my machine in the future with minimal effort or I might be able to add a special coprocessor to speed compile times.

As I understand it, the cost is minimal for this extra socket. In large volumes the cost might only be a couple dollars extra. If that is the case, all AMD main boards in the future will probably have one of these. AMD should definitely pursue this direction as it would increase their processor sales.

4:29 PM, July 17, 2006  
Blogger mrmaloon said...

The other thing that nobody pointed out is that most software (including MS) is licensed by the CPU.

4x4 would require 2 licenses of most everything, cost more than Intel solution, use more power, and run slower.

Truly genius!!!!!

5:03 PM, July 17, 2006  
Anonymous Graham said...

Anonymous coward wrote:
who cares?
A Ferrari needs 12 cylinders to be 50% faster than a Volkswagen 4 cylinder.
only thing that matters: Ferrari is faster.


HOwever, a Ferrari costs about 8 times as much as a VW. And the AMD 4x4 will surely cost quite a bit more than a mainstream desktop PC Core 2 Duo. So a strict performance comparison is apples to oranges as the first poster said. Therefore Sharikou is posting a pointless article (again!). When he can talk about performance/watt or price/performance and stop quoting the National Enquirer of computer journalism maybe people will take him seriously. UNtil the this blog is just drivel and I will continue to call him on his crap.

5:07 PM, July 17, 2006  
Anonymous Anonymous said...

Lol, you've given up completely haven't you? This is ridiculous even by your standards. I really am starting to think this whole site is a social experiment, this is just silly even when compared to your usual nonsense.

In other news, if I get two women pregnent I will have a baby in half the time!

5:44 PM, July 17, 2006  
Anonymous Anonymous said...

"Conroe XE has about 300 million transistors, so do two AMD64 CPUs combined."

So number of transistors = performance? That's an interesting theory.

What was tested in the Inquirer article? What were the benchmarks? There was no info in the article. Is the benefit only in mutli-tasking environment?

5:55 PM, July 17, 2006  
Anonymous Anonymous said...

"As I understand it, the cost is minimal for this extra socket."

Coud the person who wrote this direct the rest of us to where this information is? Thanks in advance.

If this is the case I see no more reason to ever by a single socket board with an FX62 or 5000 when you can do better with 2 3800's if all the performance and costs people are quoting are actually true. I'm curious if 2 Semprons would out perform a single FX62?

6:02 PM, July 17, 2006  
Anonymous Anonymous said...

"who cares? A Ferrari needs 12 cylinders to be 50% faster than a Volkswagen 4 cylinder.
only thing that matters: Ferrari is faster."

So I guess the whole price/performance ratio argument and power consumption are no longer important?

6:03 PM, July 17, 2006  
Anonymous amd's retribution said...

You intel fanbois are so stupid. 4x4 isn't just about adding a new processor. You can add a coprocessor as well. This would potentially boost application performance by over 100% because it be a second processor that loads Vista and 64 bit apps and specializes in speeding up particular applications. Intel doesn't have this advantage and never will due to its sad little bandwidth limitations.

See? Not only does AMD hold the ability to trash Intel by 80%-120% even after you factor in the biased reviews and fabricated benchmarks, it remains a leader in power consumption. Once AMD makes Energy Efficient FX's, then AMD will be able to double Conroe's performance at half the price and use less power. What does Intel have? Lies and money to make more lies.

There's a reason that good triumphs over evil and this demonstrates that quite clearly. No matter how much Intel lies and cheats, the truth will always come out. With dual K8L's and 4x4 being mass distributed in just a few months, Intel is once again on the defensive.

Just wait out a few quarters and by scientific analysis, Intel will be on the retreat and be potentially bankrupted. I cannot wait until this day comes. Once it does, AMD will finally be avenged for the media lies of the day.

By the time Torrenza comes out in December, even Intel's Nehalem architecture will be too old to save it even if Intel managed to bring it back through time which Intel will probably lie to us and say it did. Haha. I bet Intel will actually decrease in performance with Nehalem and Gesher like with Netbust. Those losers...

After AMD buys out ATI, we'll be able to buy graphics cards built onto our processors and witness a true revolution in computing. As you can see, even the marketing shills behind the CONroe XE are no match for truth as the recent developments of theinquirer.net shows us.

6:09 PM, July 17, 2006  
Blogger Sharikou, Ph. D said...

So I guess the whole price/performance ratio argument and power consumption are no longer important?

Usually, the last 5% performance means 2x the price. Con XE 6800+ is 2x the price of Con 6700. Because of Con XE's 10% performance lead, AMD has to cut price by 50%.

4x4 will have the more effect on INTEL. Con XE 6800 will be sold at $300.

6:36 PM, July 17, 2006  
Anonymous Anonymous said...

Those Intel fanboys just don't get it... Go buy your Conroe and STFU!!!

I personally like the idea of 4x4. Spring up a little extra for a Dual Socket Motherboard, and when the prices drop for the Dual Cores... drop in another CPU.

No one says you HAVE TO buy everything at once!! You can upgrade at your own pace.

6:40 PM, July 17, 2006  
Anonymous Anonymous said...

Talk about a desperate company flogging to try and stay ahead, and a desperate blogger trying to keep himself erect...

Where are the numbers, I don't care about rumors! Show me a test setup and some real independent benchmarks.. Its all noise from a whiny company more capable of sueing then any real engineering.

They had to sue to get the original x86, they have to use consortiums to survive.

They can't even do their own silicon.

They can't even get a first tier company to do their manufacturing.

Where is the tests? I don't care about rumors of numbers. Show me the rig.

The Doctor

6:46 PM, July 17, 2006  
Blogger Mad Mod Mike said...

I am trying to not insult Intel fanboys as much, but you are too g*d damn thick headed and won't listen!

OFFICIALLY, IN AN INTERVIEW, HENRI RICHARD OF AMD HAS STATED (paraphrased) "The cost of traditional Dual Socket Opteron64 platforms is too much for most users and that is why we have created a cheap solution known as 4x4 to bring the performance of the high-end workstation to the finger tips of the end-user" - HOW THE HELL CAN YOU TELL ME 4x4 WILL BE EXPENSIVE AFTER HEARING THAT?!?!?

Jesus, AMD fanboys may be thick headed too, but it is for good cause; K8! Intel fanboys just don't listen because for some reason, they hate AMD...not sure why, maybe because AMD whooped their Networst Craptecture by 80% w/ half the clock speed, who knows.

Get it through your head. For $2,000 I can either buy a single Conroe and a nice GPU, or I can buy Dual FX-62's and a nice GPU, do the math please. I can also buy a $800 FX-62 and a $200 Co-Processor for the price of 1 Conroe and guess what? My applications just got a 100%+ boost in performance. Add in 64-Bit, and Conroe is now Networst; old and over heating.

7:06 PM, July 17, 2006  
Anonymous Anonymous said...

Intel zealots have been making the most idiotic arguments to counter AMD’s genius architecture. I’m convinced the problem is emotional and we are wasting are time. How is it possible one company could have so many ignorant followers?

7:15 PM, July 17, 2006  
Anonymous amd candide said...

"Who cares? A computer user see a computer with specific performance and price point. Two X2 3800+ will frag a Con XE at 30% of the cost."

4x4 is a scam.

AMD sat back and said "why even spend the money to make our chip better when we can dupe the market and sell twice as many chips?"

Do many people in the world need 4 cores worth of processing power? No.

Do 4 low speed cores beat 2 high speed cores on anything except server-type tests? No.

Do entry level chips even work with AMD 4x4? No.

Is there any real price/performance benefit for 4x4? No.

Are you really going to "frag" Sheriff Conroe? No.

Who are you going to frag then?
The dumbass AMD customer who will be paying a lot of money every month to power cores he doesn't even need. But he will have a cool sticker on his computer that marks him as driving the Hummer of micro-processors.

Is AMD just a scam company driven by corruption, greed, and lies? Yes.

7:21 PM, July 17, 2006  
Blogger Mad Mod Mike said...

"Intel zealots have been making the most idiotic arguments to counter AMD’s genius architecture. I’m convinced the problem is emotional and we are wasting are time. How is it possible one company could have so many ignorant followers?"

It would be one thing if Intel was this innovative, caring, and performance-centric company (much like AMD) and than I can see people following them, but supporting networst and hording over Intel after "OMG I HAVE A REDESIGNED PENTIUM 3 THAT DOESN'T RUN 64-BIT!!! WOOTZERLEET!!" is pathetic.

p.s. Ignorant Intel fanboys beware: 4x4 is not expensive!

7:44 PM, July 17, 2006  
Anonymous who's fragging who? said...

It looks like AMD is going to get fragged for the rest of 2006, possibly all of 2007. Basically until AMD is willing to spend the money to make a better chip vs. selling the consumer two chips and other scams.

In the mobile space, Merom (C2D mobile) is coming very soon. This will frag AMD's entire mobile lineup, forcing AMD into massive price cuts (coming in August).

For the forseeable future, Intel now has leadership positions across every segment of processors except high end 4P/8P servers.

Desktop Conroe beats AMD at virtually every benchmark and offers much more performance for your electricity and processor dollar. It is a winner, a nimble fast sportscar vs. AMD's 4x4 SUV.

1P/2P Woodcrest Xeon beats AMD also at virtually every single test -- INCLUDING server-type tests such as Apache.

As mentioned above, Core 2 Duo Mobile (Merom) is going to make Turion look like the fat ugly cow it is. Just like desktop Conroe killed desktop AMD, except the same results for mobile.

AMD's 4x4 "SUV scam" will soon be exposed. No one in their right mind, in a moral mind, can even contemplate this gross and disgusting approach to computing.

When the entire world is in conflict due to energy consumption, AMD is now saying that everyone needs a gas-guzzler. What a bunch of Enron-style crooks.

With Intel dominating in all the important market segments, we have the inevitable BK of AMD. They will be merge with Sun and make overpriced servers for esoteric markets.

The smart buyer will stay away from AMD until AMD management makes better chips, not SUV gas-guzzler designs. That looks like 2008. Until then, Intel is the smart buy.

7:52 PM, July 17, 2006  
Anonymous Anonymous said...

Sharikou said "Who cares? A computer user see a computer with specific performance and price point."

I thought YOU cared when you were wondering why DOE is not imposing power consumption limits. Now it's performance at all costs?

BTW, you need to update your "4x4 is a permanent solution to pin down Intel at 50% of AMD's performance or less" line. You used to claim that a 4x4 is 100% higher performance than Conroe. Now you're accepting 60% . Is this your final answer?

7:55 PM, July 17, 2006  
Anonymous Anonymous said...

You know, no games were run. They only used unspecified multithreaded applications so really it's just a theoretical number they're quoting. There's no way your getting 80% in realworld applications. Certainly not games.

7:56 PM, July 17, 2006  
Anonymous Anonymous said...

As far I know, 4.8GHZ is unattainable with Con XE even if liquid helium is used for cooling.

Conroe is now at 5305MHz and it's still pretty early in it's lifespan. It'll probably improve as the process is slowly tweaked over the next few months and as people have more time to play around.

Runs SuperPi in 9.64s too, although I know you don't really care.

7:58 PM, July 17, 2006  
Anonymous Anonymous said...

I don't think you are going to see X2 3800+s in a 4x4 platform. AMD 512k processors only have 1 physical HTT link so they can't be used. Only 1MB cache models have actually have 3 HTT links. You can look at the die shots to confirm. HTT pads wrap around the core, and DDR pads wrap around the L2 cache.

If AMD were to open up 4x4 to regular X2s it'd be for the upcoming X2 5200+ and X2 5600+ since they are the only 1MB models left.

8:01 PM, July 17, 2006  
Anonymous Anonymous said...

Add in 64-Bit, and Conroe is now Networst; old and over heating.

Somehow I doubt 2 FX62s are going to be cooler than a single Conroe.

8:02 PM, July 17, 2006  
Anonymous Anonymous said...

wow, looks like amd took the doc's advice on pricing. check anand's report on amd's upcoming price cut. now 4x4 looks alot more affordable and an excellent price/performance system. now if only amd announces support for s939 x2 and that you can use any grade cpu. i think it would be best and an advantage to amd to make a 4x4 board that supports different cpu. i'm sure alot of x2 owners would jump at the chance of upgrading their system to 4x4 with a cheap 3800 or 4200, or 4600. i can't wait for the conference this thur. what's your prediction on intel's conference doc?

8:04 PM, July 17, 2006  
Anonymous Anonymous said...

4x4 isn't gonna "frag" Conroe by 60%. Proof? Like this one http://www.firingsquad.com/hardware/ultimate_workstation_ultimate_gaming_pc/default.asp

Dual Opterons versus X2 versus single core Athlons.

"X2 3800+" only have a SINGLE non-ccHT link. This also applies to FX processors. Only high end Opterons have ccHT links.

Thus I don't see how 4x4 can be used on these processors.

8:21 PM, July 17, 2006  
Anonymous Anonymous said...

"Get it through your head. For $2,000 I can either buy a single Conroe and a nice GPU, or I can buy Dual FX-62's and a nice GPU, do the math please. I can also buy a $800 FX-62 and a $200 Co-Processor for the price of 1 Conroe and guess what? My applications just got a 100%+ boost in performance. Add in 64-Bit, and Conroe is now Networst; old and over heating."

OK I understand your arguments but now you are just making up cost #'s.

$200 co-processor? When, who makes it, where can I buy it?

$2000 for 2FX's+GPU means ~ 400GPU ($800per FX62 AFTER the price cuts +~400 GPU)

Conroe XE ($1000) + same 400 GPU = 1400 this is a little lower than trhe 2000 you claim (you must have been rounding up to nearest 1000?). If you drop down one speed grade to a 6700 (which you readily state many place is very close to XE in performance) and you are now at <1000.

You also assume board costs will be the same, this may or may not be true as I have yet to see any pricing on 4x4 boards. And will you need power a bigger power supply and pay more electricity or is the FX-62 suddenly going to be cut to a 65Watt product?

8:38 PM, July 17, 2006  
Anonymous Jeach! said...

I'm a little perplexed about this blog Sharikou.

This is why:

I would disagree with you and agree with another reader that you are comparing apples to oranges. Not because there are 2 processors versus one. If they were priced the same that wouldn't matter, but the fact that a two processor AMD system will cost probably 50% more, it's a weird comparission.

BUT, on the other hand I've always believed that the worlds greatest business-model is the 'subscription based' business model. And I believe that most 4x4 system purchased by users will be with only 1 CPU. That means that AMD is selling a subscription to another in the short future (an upgrade path). With a single processor both machines should be evenly priced (assuming AMD drops prices by 50% soon to match Conroe prices).

The beauty of this is that when processors are cheaper one can decide to add a second processor or sell it on Ebay so that someone can buy 'it' as their second processor while the first person goes out and buys one (or two) 4 core processor for their 4x4 platform.

Either way AMD gurantees sales!

Just wait for their 8x8 platform with up to 8 cores!!

8:56 PM, July 17, 2006  
Blogger Sharikou, Ph. D said...

I thought YOU cared when you were wondering why DOE is not imposing power consumption limits. Now it's performance at all costs?

Two X2 3800+ EESF, 35 watts TDP_max, total 70 watts TDP_max. Frag conroe XE6800 at lower energy and lower cost by 30%.

Those use FX62s can afford the energy tax.

8:57 PM, July 17, 2006  
Blogger Sharikou, Ph. D said...

If they were priced the same that wouldn't matter, but the fact that a two processor AMD system will cost probably 50% more, it's a weird comparission.

No. 60% more performance and 50% more cost, that's a bargain.

If you want to get 30% performance over CON XE 6800, you pay about only $350: two X2 3800+ at $175 each + $50 more on MB. To get a Con XE 6800+, you pay $1000

9:01 PM, July 17, 2006  
Blogger mrmaloon said...

That's great you plan to put a $200 co-processor into the second slot.....can you tell me who makes a non PCIe HT-based card that will fit in that slot? Know anyone who has one planned???? Alright, then.....you can drop that coprocessor into that slot in 2009 when one is available.

Face it, 4x4 may capture 2% of the market......it would be lucky to get 5% of those users to buy a given coprocessor.....you're talking about a product which might make it into 0.1% of the market in your coprocessor.

If you are lucky enough to get someone to build a chip for the second socket, it will trail the PCIe-based equivalent in technology because the volume is not there which will more than offset any advantage from HT. Why build a HT-based coprocessor for 0.1% of the market when you can build a PCIe one that will fit into billions of PC's??? Furthermore, the low volume means you'll pay through the roof for it.

Why would a company build a chip for such a low-volume application as the second socket on a 4x4 platform??? The answer is that they won't unless you own them (this is probably the basis for ATI). But you are still left with 2-3 years before you see a product from them. The product you'll get from them will be the equivalent of a graphics card from Intel....which even I'll freely admit wouldn't be worth a darn.

9:02 PM, July 17, 2006  
Anonymous Anonymous said...

So number of transistors = performance? That's an interesting theory.

More transistors means more execution units, and more ancillary equipment to keep those execution units filled with instructions. If the platform is designed well and doesn't waste transistors needlessly, more transistors does mean faster.


You can add a coprocessor as well...Intel doesn't have this advantage and never will due to its sad little bandwidth limitations.

By that logic the 8087 and up wouldn't have ever existed. Granted things are a little different now with the relative speeds of busses, memory, and floating point throughput. But don't say it will never happen, because it can. If 4x4 is as cheap as we all hope it is, Intel will respond (eventually). ATI did it when Nvidia brought out SLI, so too will Intel if AMD brings out a cheap enough 4x4 system. Granted it will probably not be an elegant solution, even duct-taped together if need be, but if it needs to come, it will.

9:04 PM, July 17, 2006  
Anonymous Anonymous said...

"Two X2 3800+ EESF, 35 watts TDP_max, total 70 watts TDP_max. Frag conroe XE6800 at lower energy and lower cost by 30%."

Please point me to these benchmarks. Come to think of it, please point me to the benchmarks with 4x4 FX62 as well.

"That means that AMD is selling a subscription to another in the short future (an upgrade path)."

Wasn't it not 3-6 months ago where an AM2 board was to supposed to be "future proofed" for the enthusiats folks? You know support DDR3, same socket for the next 2-3 years - for those wanting top of the line performance a single socket AM2 board has now got to be replaced by a 4x4 board. Yes, the 4x4 gives you more POTENTIAL future flexibility, but my point is there is no way of knowing what will truly be future proof.

9:08 PM, July 17, 2006  
Anonymous Anonymous said...

One thing which is relavent to the arguments about price vs performance is software licensing.

With most server software you get processor licenses (unlimited clients, X licenses per CPU) the interesting thing is that when you get a CPU license if it's 1, 2 or 4 cores the cost is the same so you get major savings buying a 4 core AMD over a dual cpu Intel. That said Intel is bringing out a 4Core cpu but its a hack to say the least.

9:14 PM, July 17, 2006  
Anonymous Anonymous said...

For the forseeable future, Intel now has leadership positions across every segment of processors except high end 4P/8P servers.

What about the low end where AMD is cutting the bottom out of their entire lineup? Isn't this where the most numbers of processors will be sold? And with 4x4, the top end will be 2 socket AMDs.

9:15 PM, July 17, 2006  
Anonymous hummer lovers vs. planet earth said...

"X2 3800+" only have a SINGLE non-ccHT link. This also applies to FX processors. Only high end Opterons have ccHT links.

Thus I don't see how 4x4 can be used on these processors."


They can't. It is physically impossible.

But don't expect AMD fanbois to understand more than "ccHT is God!!!". Even if there isn't actually ccHT on the frakkin chip.

Which makes it all too funny when some people take a very expensive "EE" chip, double it up, and compare that to a top of the line Conroe.

Unless you are running server apps on your little ATX double socket game board, you won't see great performance. Four low speed threads and minimal cache don't perform worth a darn for most apps vs. two high speed threads and lots of cache.

But if you are truly a dumbass fuck-the-planet Hummer lover, go out and get your little double socket AMD game board. AMD will sell two new expensive energy hungry FX2 chips that you can use to make yourself an AMD Hummer 4x4.

As an aside on this blog... I will say that Sharikou used to care about what was right/best, but he has taken a detour into a darkness, most likely due to AMD's absolute betrayal of all its customers.

AMD is not doing things right anymore. AMD is about scams, lies, corruption, overpriced old chips and milking old designs instead of making new good chips.

FYI, my top vendor told me today that the last batch of Opteron 265s has rolled out of the fab. That means that from now on there are no affordable 2P Opterons. AMD is going to milk the Opteron 940 and keep prices high -- maybe even raise prices soon -- to make up for the revenue losses from having to cut the prices on all their desktop chips.

It is now cheaper to swap an entire motherboard + RAM out and get Intel Woodcrest Xeons vs. buying anything from AMD.

AMD can go frak themselves.

9:16 PM, July 17, 2006  
Anonymous BigBadWolf said...

"Do many people in the world need 4 cores worth of processing power? No."



This is the stupidest argument i have ever heard.
One never really gets as much processing power as they want.

Why do't you go back to 486?

9:17 PM, July 17, 2006  
Anonymous Anonymous said...

As far I know, 4.8GHZ is unattainable with Con XE even if liquid helium is used for cooling.

Conroe is now at 5305MHz and it's still pretty early in it's lifespan. It'll probably improve as the process is slowly tweaked over the next few months and as people have more time to play around.

Runs SuperPi in 9.64s too, although I know you don't really care.


Sorry forgot the link.

http://www.xtremesystems.org/forums/showthread.php?t=107353

Conroe at 5305MHz.

9:19 PM, July 17, 2006  
Anonymous enumae said...

Sharikou, I have to ask do you read peoples post?

You said... "If you want to get 30% performance over CON XE 6800, you pay about only $350: two X2 3800+ at $175 each + $50 more on MB."

I already broke down your claims. I found it to be a maximum 16%, granted this was only gaming, but the other benchmarks (non gaming) were even more in Intels favor if I recall corectly.

Also, do you believe that a new mobo that is capable of doing 4x4 will only be $50, seems a little low?

Next... "Frag conroe XE6800 at lower energy and lower cost by 30%."

4x4 will not be expensive but it also wont be entry level prices, AMD has a chance to make a little money here. 2 * X2 3800 = $300, plus a new motherboard (enthusiast price) $200, for a total of $500. Intel X6800 = $1000, new motherboard (enthusiast price) $200, for a total of $1200.

All this puts the price at about 42% which is not bad, add 2 better processors and your really going to have a good system.

2 * X2 5000 = $575, new motherboard (enthusiast price) $200, for a total of $775.

Now all of a sudden it looks very nice and at 65% of the Inetl price.

All prices shown are reflecting the price cuts.

9:33 PM, July 17, 2006  
Anonymous Anonymous said...

Shairkou:
"If you want to get 30% performance over CON XE 6800, you pay about only $350: two X2 3800+ at $175 each + $50 more on MB. To get a Con XE 6800+, you pay $1000"

So by this logic AMD will never sell another new single socket FX62 solution as it would be better to buy a 4x4 with 2 3800's? Since this solution "frags" a Conroe XE and XE matches/surpasses a single socket FX62, this solution would also have to "Frag" a single socket FX62, no?

So the only market now for FX62 are the 4x4 adopters (which AMD themselves state is targeted at ENTHUSIAST market) or for folks who have an existing board and are only looking to upgrade their CPU.

If you're 3800x2 4x4 speculation is true, AMD just cut the heck out of all it's high end single socket solutions.

9:37 PM, July 17, 2006  
Anonymous Anonymous said...

Serious question: I understand how a dual socket solution might benfit heavy mutli-tasking performance, but how will this benefit single threaded applications without SW significantly re-coded to take advantage of a second socket?

What were the specific applications the 80% performance gain was measured on?

Also who did the testing? (Was this AMD or a third party website?)

9:41 PM, July 17, 2006  
Anonymous Tritosh said...

" Anonymous said...

As far I know, 4.8GHZ is unattainable with Con XE even if liquid helium is used for cooling.

Conroe is now at 5305MHz and it's still pretty early in it's lifespan. It'll probably improve as the process is slowly tweaked over the next few months and as people have more time to play around.

Runs SuperPi in 9.64s too, although I know you don't really care. "

and apart from running superpi, tell me, what does it can do else?
keep working 24/7?
even 2 hours?
even 1 hour?
15 minutes?
or it was JUST ENOUGHT stable for a superpi before it crashed?

"The other thing that nobody pointed out is that most software (including MS) is licensed by the CPU.

4x4 would require 2 licenses of most everything, cost more than Intel solution, use more power, and run slower.

Truly genius!!!!!"
dude
you're smoking weed or something?
Microsoft sells most of its software PER COMPUTER(at least almost all of its consumer's ), wich means LICENSE PER SYSTEM, not by CPU retard.
unless you mean high end server licenses such as Microsoft server stuff. (wich is by cpu, and preety expensive.. but only solution for windows based systems.. )
and last time I seen 4 x4 was giving the "normal users" the "power and structure of high end workstations".

" amd candide said...

"Who cares? A computer user see a computer with specific performance and price point. Two X2 3800+ will frag a Con XE at 30% of the cost."

4x4 is a scam.

AMD sat back and said "why even spend the money to make our chip better when we can dupe the market and sell twice as many chips?"

Do many people in the world need 4 cores worth of processing power? No.

Do 4 low speed cores beat 2 high speed cores on anything except server-type tests? No.

Do entry level chips even work with AMD 4x4? No.

Is there any real price/performance benefit for 4x4? No.

Are you really going to "frag" Sheriff Conroe? No.

Who are you going to frag then?
The dumbass AMD customer who will be paying a lot of money every month to power cores he doesn't even need. But he will have a cool sticker on his computer that marks him as driving the Hummer of micro-processors.

Is AMD just a scam company driven by corruption, greed, and lies? Yes."

Yeah? please, tell me why we're on dual cores now, can you tell me please?
can you explain it?
I dont think you can,can you?
you cant scalate the power by increasing always the frecuency or adding more stuff, the real solution is adding MORE CORES or MORE SOCKETS.
Ie, multiprocessing..
these guys are each more time more retard.
and you guys should be thankful that even sharikou who accepts the "freedom of speech" in his blog, lets you post your constant bigotry and ignorance coming from intel fanboys who eat everycrap intel's marketting tosses.
at this rate IWould easily figure out that you guys think that the war on iraq was legal, and that sadam indeed had atomic bombs :P

9:43 PM, July 17, 2006  
Blogger Sharikou, Ph. D said...

So by this logic AMD will never sell another new single socket FX62 solution as it would be better to buy a 4x4 with 2 3800's?

Yes. That's true. But remember, there are people who needs two FX62s, or those who buy one FX62 and buy one more when they have the money.

do you believe that a new mobo that is capable of doing 4x4 will only be $50

I was talking about the added cost to a MB.

9:43 PM, July 17, 2006  
Anonymous Anonymous said...

I love watching this. All of you kooks are self destructing. If someone came and read this stuff for 2 minutes they would realize what utter nutters you are.

I used to not let your rambling sour me on AMD, who really does have some good technology and has done a great job with K8.

But this display of desperation on their part is just kind of sad. Releasing half-assed benchmarks (and these really are rigged, unlike the ones Intel released which have been confirmed since) showing very specific benchmarks which benefit from 4 cores.

The number of applications which benefit from 4 cores is tiny.

But you know what's funny? Not long after "4x4" Intel will have a quad core CPU that can be dropped into existing P965 motherboards.

I think AMD will get over this sometime next year, but it's going to be very fun watching you all fall apart.

9:48 PM, July 17, 2006  
Anonymous Anonymous said...

"Dr" FanBoi. Apple to Oranges. The 4x4 ain't out yet is it? And won't be till quite a bit after Conroe and Woodcrest have quite a bit of time to dig in deeper.

You yell and scream when Intel cuts prices, but I notice you make little mention on the fact that AMD both lost 1000 employees from Geode for the most part and also will be cutting prices on all of their processors by UP TO 50%. AMD doesn't have that kind of money to keep cutting prices across the board.

For the moron that made the Ferrari comment, the Ferrari is also 100x the price of the VW.

Blah blah blah goes the fake doctor how a upcoming, yet to be released processor can do much of anything. No wonder you are the laughing stock of any tech place...

9:52 PM, July 17, 2006  
Anonymous enumae said...

"I was talking about the added cost to a MB."

Ok, I didn't really think that, buthad to clarify.

Thanks

9:59 PM, July 17, 2006  
Anonymous Anonymous said...

Sharikou - thanks for answering my question about the 2 3800x2 vs FX62.

I'm wondering if you could shed some light on my other 2 questions:

1. What were the specific applications the 80% performance gain was measured on?

2. Also who did the testing? (Was this AMD or a third party website?)

Thanks in advance.

10:01 PM, July 17, 2006  
Blogger Entrophos said...

The sheer stupidity and ignorance displayed by the pro-Intel crowd posting here as a whole makes me despair for the future of mankind.

I see few real facts, many outright lies, incorrect assumptions and a lot of parroted Intel marketing-speak. You claim Sharikou is full of it, but most of the arguments against his viewpoint are smoke and/or amount to "LAWLZ INTEL PWNS AMD SUCKA! KEKEKEKE". The fact that many of the worst flamers write like they learned English in Usenet and IRC is painful, to say the least.

Sharikou might be hardcore pro-AMD and biased in some of his opinions and he does have some colorful viewpoints and commentary. He is often right, judging by what I've read so far, but he can also be wrong. He's only human.

At least he doesn't resort to the juvenile trolling exhibited by his detractors. He even tries to discuss things rationally with people that are completely irrational (rarely seems to work but I give him kudos for trying_. I think I've seen him lose his cool a couple times and it was never nearly as viscious as most of the stuff lobbed his way.

Even if I disagree with his viewpoints from time to time, I can respect him for his conviction in what he says. I *highly* respect the fact that he actually allows some of these comments to even make it past the bit-bucket. I would not have anywhere near that much patience or willingness to believe in free-speech on my own blog.

10:10 PM, July 17, 2006  
Blogger Sharikou, Ph. D said...

1. What were the specific applications the 80% performance gain was measured on?

2. Also who did the testing? (Was this AMD or a third party website?)


INQ didn't report details. But you can just look at Opteron 1P to 2P scaling, and see in general, there is a 80% boost in both integer and floating point performance.

10:12 PM, July 17, 2006  
Anonymous Anonymous said...

"INQ didn't report details. But you can just look at Opteron 1P to 2P scaling, and see in general, there is a 80% boost in both integer and floating point performance."

Again this is not meant to be a "fanboy question", but aren't the server benchmarks on SW designed to take advantage of multicores/multisockets?

Is typical SW for desktop designed the same way? I'm not familiar with server SW so much so I don't know if this is the case.

Also how do you quote performance to an article which neither mentions the benchmark or the person doing the benchmark (doesn't this send up a "red flag")?.

Thanks again for answering my questions.

10:17 PM, July 17, 2006  
Anonymous BigBadWolf said...

Stupidity reigns supreme amongst Intel fannies.
All AMD is trying to do here is to allow average user to have more horsepower at an affrordable price and you cry foul.
Just cuz your crappy processors wo't scale due to inferior architecture) does't mean its foulplay .
Intel is doomed untill it pulls out something shiny outta its ass to counter 4x4.

" who needs 64 bit said the great Intel"
And later licked their own chops.

And now i hear you poops stating who wants 4x4 and why, just to mention a coupl of examples.

You fannies just love to live in the dark ages , do't you?
You are so stupid that hereinafter i am gonna call you either *netBust" or FSB.

10:17 PM, July 17, 2006  
Anonymous BigBadWolf said...

To all the Intel girlie morons :

Listen good , sooner than later programmers are going to take advantage of multiple cores using more multithreaded enabled apps.
So its future proof investment.
Software developers not as stagnant as you might think sitting on your laurels.
So get a life and stop opposing the innovation you retarded pieces of crap.

10:26 PM, July 17, 2006  
Anonymous Anonymous said...

"All AMD is trying to do here is to allow average user to have more horsepower at an affrordable price and you cry foul."

With 4x4, AMD is saying if you want 80% performance improvement over their own single socket solution (and I'm skeptical of this 80% figure until some benchmarks get published); they will charge you >100% price (100% more for the 2nd processor + some added cost for additional motherboard complexity).

In other words if I have a 4400X2 and want this 80% improvement I will need to double my investment (from the CPU/motherboard perspective as I will need to buy another CPU and another motherboard)

For the enhthusiast user I'm sure this is not an issue, but for the average consumer I would prefer AMD focus improving performance on their single socket solution so they won't just double the end price to the consumer).

Note - in this analysis, I'm taking Intel out of the equation and just comparing AMD current solution to AMD future solution.

10:49 PM, July 17, 2006  
Blogger Michael said...

4x4 is truly amazing. It may be a little expensive but I agree with sharikou that two X2 3800+ with destroy Conroe in most benches. What I'm curious about though is, any thoughts on what it would be like with two X2 3600+ when they are released later on this year or early next year? That could be the cheaper alternative to a Conroe based system.
CyberSurge

10:58 PM, July 17, 2006  
Anonymous Anonymous said...

Intel processors won't scale?

How do you know? The Conroe is the first of this new gen processor. It's already faster and a LOT less expensive than anything AMD currently offers. Thats the plain and simple truth.

It's faster in both 32 and 64 bit.

I LOVE how the AMD Fanboi's need to constantly change the goal post. Before Conroe and Woodcrest if an Intel fanboi said anything "you just wait till Intel releases x or y" they were ripped apart by people with fake PHd's and those that support is inane rambling...

Now that AMD is on the other side, it's perfectly fine to discus 4x4 which isn't going to be out for a while.

When Intel slashed prices, it was all doom and gloom for Intel. But when AMD does it "Intel is gonna die now!!!! OMGZ!"

When Intel cuts 1000 jobs, no one even mentions that AMD cut 1000 jobs too.

Fanboi's led by a fanboi who has to lie about his PHd.

11:24 PM, July 17, 2006  
Anonymous Edward said...

X2 3800+" only have a SINGLE non-ccHT link. This also applies to FX processors. Only high end Opterons have ccHT links.

Thus I don't see how 4x4 can be used on these processors."

They can't. It is physically impossible.


Somehow I remember it's possible to break a wide HT link (say 32-bit) to multiple thinner ones (say 16-bit).

I could remember wrong, though. BTW, that'll also cut the memory bandwidth in half.

As for cache coherency, it's true that for multi-processor to work efficiently it is necessary to have that. Either 4x4 will require special, non-Opteron, ccHT Athlon64 chips, or AMD has to find a way to avoid cache coherency problem between the two processors (if possible).

As for the claim that 4x4 is just a scam to sell more processors, it is not. For one, you are not required to use 4x4, even if you buy a 4x4-capable motherboard. Then, instead of trashing the CPU+MB (probably also memory) and migrate all your data to a new computer after 3 years, with 4x4 motherboards you can first leave a socket empty and spend $300 later to drop in another processor to upgrade performance. While we yet not know how AMD's going to play out 4x4 exactly, it'll be attractive if it's done right (and I know the last thing Intel fanboys wish to recognize is that AMD can do things right... well, whatever).

As for the claim that 4x4 will use more power, it's only partially true. It's true if your multithreaded app needs that extra computing power; OTOH, it *might* be possible for one processor to go to low-power state while the other is working full-load. Again, we need to wait and see how AMD plays out 4x4 with its MB partners.

As for the claim that 4x4 will be noisy, it's plain wrong. Two lower-speed fans will be quieter than a high-speed one. You'll need to spend more on the fans, though, since you need to buy two of them (which I suppose are good ones).

As for the claim that 4x4 w/ 3800 will frag one Conroe (and thus FX-62), it's only partially true. For some multithreaded apps, yes, but for all single-threaded apps 4x4 simply won't help. Even for many multithreaded apps, Conroe's dual-core w/ shared cache will be more efficient than four cores connected with a HT link.

The only hope on AMD to compete with Conroe on the mainstream market is K8L, which is due out mid-2007. Maybe rev.G later this year will help a bit, too. OTOH, 4x4 is not and was not supposed to be compared to Conroe; it is aimed only at multithreaded, memory intensive workloads - for those a single dual-core is not enough but a two-socket certified server is too costly.

11:54 PM, July 17, 2006  
Anonymous Edward said...

"Intel processors won't scale?

How do you know? The Conroe is the first of this new gen processor.
"

FSB architecture doesn't scale. It'll be really sad if Conroe is the first of its generation. Within a few years this FSB generation will end.

"It's already faster and a LOT less expensive than anything AMD currently offers."

Faster, yes; less expensive? Not true if you can't buy them, also not true if AMD price cuts. Besides, Conroe is not what Intel *currently* offers.

"When Intel slashed prices, it was all doom and gloom for Intel. But when AMD does it "Intel is gonna die now!!!! OMGZ!""

I believe it's the other way around, with Intel fanboys saying AMD's going to die due to Intel price cuts. AMD's price cuts are aimed at Intel, specifically Conroe.

"When Intel cuts 1000 jobs, no one even mentions that AMD cut 1000 jobs too."

Did you read the Inquirer's update on that news? AMD cut much less jobs, most of which were offered other positions in the company.

It's truly fun to see Intel fanboys getting all excitingly emotional for the yet unreleased Conroe. It's almost like a 10-year-old waiting for Christmas present since October... or more accurately, July?

Come on, Intel did a good job, and secures 15% performance lead in average to AMD. That's all. That happened before, and will happen in the future, from both Intel & AMD side against the other. What's truly exciting is the 4x4, if it plays out right, and its possibilities.

12:09 AM, July 18, 2006  
Blogger nyx said...

Intel fanboys guarantee job security for computer system repair technicians.
Some very insightful posts from the AMD community as usual. It's evident to most that AMD's roadmap allows for considerable innovation and performance increases. It just seems that most Intel fanboys lack the basic knowledge of hardware architecture and efficient software implementation to realize that.

12:24 AM, July 18, 2006  
Blogger nyx said...

Such software developers as Bioware, Cakewalk, Crytek, Havok, Irrational Games, Midway, Nero and Sony have been working on software applications and games to take advantage of multi-core processors and benefit from AMD’s direct connect architecture.
While there are much more software designers, who have been creating applications that take advantage of multi-threading on systems featuring several processors or processing engines, AMD implies that titles from the aforementioned companies would be able to take advantage of the 4x4 – AMD’s multi-socket platform for enthusiasts.
We’re looking at a decidedly multi-core processor, multi-threaded application future for the PC industry.

The above are points extracted from:

http://www.xbitlabs.com/news/cpu/display/20060626004842.html

in order to further drive the point home.

12:35 AM, July 18, 2006  
Anonymous Anonymous said...

Edward - you made a lot of good points but this one is off:

"Faster, yes; less expensive? Not true if you can't buy them, also not true if AMD price cuts. Besides, Conroe is not what Intel *currently* offers."

Putting aside the endless (and childish) "it's not available OFFICIALLY debate" (as if the chips will somehow get pushed back 6 months?), Conroe when released will be cheaper than the equivalent performance to single socket AMD solutions, even when advertised AMD price cuts kick in. The only exception to that is the XE, however that outperforms the highest priced AMD desktop part, so there is no comparable AMD single socket part at this time.

I am intrigued by your argument that you can potentially just choose to use the 4x4 as a single socket - thus the only added cost to end user is the delta cost between single socket and 4x4 board. I think this is a very good point.

My only concern here is realistically how long that window is - things are fairly dynamic on the component front betwen CPU socket type changes, memory, GPU's, PCI to PCI-x, etc so the actual upgrade/"insurance" window may be soemwhat less than 3 years. It is something to consider though for people on a budget who think they may get the upgrade itch/money later.

I have not seen a board yet - any guessed on whether you need to have separate DIMM's for each socket? (I'm assuming you do?) If so, you would either need to buy the extra memory up front or that too might be an obsolescene issue down the line. (not sure if speeds and/or size need to be matched between sockets - if they don't then this will not be an issue).

And I think your analysis on Aps dependency on performance benefit is dead on - many people here, including the blog owner, have seemed to latch on to this one random article which has no actual data or source and assume this 80% claim will be the average performance benefit of this solution.

Before folks start flaming me - I'm not saying there won't be a benefit from this 4x4 approach, I'm just saying let's wait and see until some actual benchmarks come out.

12:52 AM, July 18, 2006  
Anonymous BigBadWolf said...

My point here is that its just a beginning.Once 4x4 solution is made available to the enthusiasts and it gains popularity ( as i am sure it should see overwhelming response ) , definitely it will see improvements , performance wise and so on
And why just for hi end users , why can"t an average joe have extra muscle at an affordable price.
Will be good for evey one , users getting more bang for the buck and AMD ending up selling nearly twice as many processors.
Someone here accused AMD of trying to sell greater number of cpu's thro 4x4 as if its a federal crime.
Please get it throug your head.
It's a solution that requires 2 brains instead of one.
Now tell me whats wrong with?
Put yourself in the pic and think what if you happen to be a heavy multitasker , what would you do then?
Go for a 4x4 or or opt for some hi end offering that might burn a hole in your pocket.
Let me reiterate , its a beginng.
How did the first airplane look like?
Trashing a good Product just because it does't live up to fullest expectations as yet is plain dumb.
Where we would been had that first airplane been trashed?
As for the argument that not many of 64b apps available , does't hold ground.
Its just a matter of time before 32 bitters shall diminish.As a matter of fact the demise of 32b will begin the moment vista hits the shelves.

Cheers

3:31 AM, July 18, 2006  
Anonymous BigBadWolf said...

On second thoughts :
If 4x4 proves livesup to the hype ,why should single sockets exist anymore?
Go out grab a dual socket mobo and insert a processor.
At a later stage should you need more muscle drop in another.
It should save at least one upgrade cycle.
Correct me if wrong.

3:49 AM, July 18, 2006  
Anonymous Richard said...

Well, AMD's prices have already considerably dropped and they are hurting at the moment.

http://blogs.techrepublic.com.com/Ou/?p=271

50% share price drop and it isn't to do with performance, it's to do with marketing.

So, without being a fanboy here, Intel have business skills (although unethical), they've acheived their goal quite nicely at this stage.

To think, all they had to do was give away a few CPUs to hungry reviewers...

I have never ever in this industry come across such marketing and anticipation. Maybe it's the sign of all in or all out from Intel?

4:39 AM, July 18, 2006  
Anonymous Richard said...

I forgot to mention that, even if AMD can frag Intel in the not too distant future, if Conroe takes off due to it's marketing, it'll still put AMD on the back foot.

But then again, if Intel are all lies, AMD will once again reign supreme.

Only time will tell.

5:18 AM, July 18, 2006  
Anonymous Anonymous said...

Most digital media and games are already multi-threaded. I challenge any naysayers out there who have a dual core processor today to watch their CPU meter as they are running apps. If you see the two meters running at the same time then it is a multi-threaded app. Check out the DIVX site as even their latest, multi-core optimized encoder and player. Also, look at all the multi-core patches available even for older games. Newer games will all be writen for multi-core if they want cross-platform compatibility. XBVox has 3 cores and PS3 has 12 CPU cores.

5:35 AM, July 18, 2006  
Anonymous Anonymous said...

no need to wait for 4x4.

conro is dusted in amd64 mode. all you need is just 64bit os and applications. some generic C2 features do not work in 64bit mode!

http://aceshardware.com/forums/read_post.jsp?id=120062277&forumid=1

to quote that japanise "the reason where intel stops emphasizing 64bit"

5:47 AM, July 18, 2006  
Anonymous Anonymous said...

imagine the possibility. if 4x4 board is am2 and k8l is rumored to work with socket am2. the possibility of running 2xk8l quad cores for a total of an 8 cores dual socket system. wow that thought makes me shiver. if anyone from amd is reading this, make it happen please. if this becomes a reality, who needs conroe. anyone with me on this. lets get a petition going.

5:58 AM, July 18, 2006  
Anonymous Anonymous said...

Since the 2.93 GHZ Conroe XE 6800 is about 10% faster than the 2.8GHZ Athlon 64 FX62

Wow, master of choose and pick.Isn't that 2.67GHz E6700 already beating 2.8Ghz Athlon 64 FX62 by more than 15%?

6:46 AM, July 18, 2006  
Anonymous Anonymous said...

who cares?
A Ferrari needs 12 cylinders to be 50% faster than a Volkswagen 4 cylinder.
only thing that matters: Ferrari is faster.


Good analogy. Same thing on the Intel vs AMD. You can cry all you want that intel is without IMC and HT. It is much faster and energy saving than the AMD for 2p and below, and cheaper. who cares.

6:54 AM, July 18, 2006  
Blogger mork said...

"Intel fanboys guarantee job security for computer system repair technicians.
Some very insightful posts from the AMD community as usual. It's evident to most that AMD's roadmap allows for considerable innovation and performance increases. It just seems that most Intel fanboys lack the basic knowledge of hardware architecture and efficient software implementation to realize that."

Then again most AMD fanboys seem to lack the most rudimentry knowledge in economics, so it all adds upp :p

8:29 AM, July 18, 2006  
Anonymous Anonymous said...

and apart from running superpi, tell me, what does it can do else?
keep working 24/7?
even 2 hours?
even 1 hour?
15 minutes?
or it was JUST ENOUGHT stable for a superpi before it crashed?


Sharikou said that Conroe couldn't get above 4.8GHz even with liquid helium. All I'm saying is that they've reached over 5.3GHz. No need to get all worked up.

In fact the highest clocked Conroe is at 5267MHz.

http://valid.x86-secret.com/show_oc.php?id=102876

You show me a AMD dual core that can pass just 4GHz for a few minutes and I'll be equally awwed. These are short pipeline processors afterall, and this is Netburst type clock speeds they're squeezing out of them using still early B0 and B1 steppings so you can't really complain.

Kentsfield's hit 3906MHz also.

http://valid.x86-secret.com/show_oc.php?id=103729

8:35 AM, July 18, 2006  
Anonymous Anonymous said...

FSB architecture doesn't scale. It'll be really sad if Conroe is the first of its generation. Within a few years this FSB generation will end.
No, the conroe is not the last generaton that use FSB. And indeed, i was wondering why a general would care so much how how the implementation is, as long as the CPU is faster and greener.

Anyway, FSB (and no IMC) is not a bad design as of its target market. It serves nicely for 1P to 2P market . The FSB is not on its own, it is backed with large and efficent cache at the back, and clocked at higher frequency. The FSB is just one of the CPU 'component'. The important thing is that it is able to provided the needed bandwidth. The Core 2 Duo arcitecture has been able to reduce FSB accesses (bigger cache, shared cache, etc) For the same program (generally), AMD CPU would access DRAM more often than Intel CPU.

and for those that still want to argue on the 2P end, intel has chipset that has 2 FSB. (i'm not sure on 4p)

Admittingly, Intel need CSI to match AMD's scalability at 4p and above.

8:42 AM, July 18, 2006  
Anonymous Anonymous said...

I know Sharikou is an extremist AMD fanboi, but others on this thread can answer this: where are these stability concerns regarding Conroe arising from. Can anyone point out a single source, other than INQ Theo Valich, who has seen this problem. I have read at least 100 independent reviews on Woodcrest/Conroe, and not a single of them (other than Theo Valich's one) mentions stability issues. And as I understand, Theo is as short on credibility as Sharikou is.

For long, Intel has been known for its stability. Intel has never shipped OCed processors, AMD has.

8:58 AM, July 18, 2006  
Anonymous Anonymous said...

Who cares? A computer user see a computer with specific performance and price point. Two X2 3800+ will frag a Con XE at 30% of the cost.

well, i'm not sure how you come out with the 30% cost. let's ue the cheap CPU that you are refering to.

2 CPU = USD 330
Premium motherboard with 4x4 support , say USD 100 extra.
if both system configure to have same amount of RAM, then by performance wise, there will be a hit at the 4x4, as it has less ram at 1 CPU. if you choose to double the ram, then the DRAM cost, say USD300.
Then, you can choose to have more graphic card or not. sau you choose not, then performance wise, not as compelling.but if you choose to, then there will be a hit.

then, you are forced to use XP Pro which is about USD 200 more expensive than the home edition. (i;m not sure about VISTA)

then the special casing and power support to support the energy monster.

large energy bill.

some software needs different license to run in MP system.

And why compare to the X6800, why not compare to the E6700 which is much cheaper, USD5xx. To be fare, the 4x4 system using x2 3800 would beat the E6700 in heavy Multithread, but it won;t be much. In addition the E6700 would still out perform this 4x4 system using 3800 in single threaded (most game is) and light multithread environment.

Oh, i should use the x2 3800? then put in more $$.


If Intel can bundle 10 Conroes into one package and get 10x performance, do it. But that's not possible, because each of the core will only get 1066/20 = 50MHZ of bandwidth.

Do you have engineering background? Intel can't and so do AMD. Any way, as per current CPUs, AMD CPU scale better than Intel at 4p and above (still lose to Intel at 2p, as there is chipset with 2 FSB)

9:02 AM, July 18, 2006  
Anonymous Anonymous said...

Anyway, I hope I am wrong... but unless we see more real results, your conclusion on 4x4 fragging Conroe is just too early.

I agree. And the funniest par tof his statement is on the lower cost side. refer to my previous post on my argument on that end.

9:06 AM, July 18, 2006  
Anonymous Anonymous said...

Cool! This would also mean, two cheap X2 3800+ for $300 will be about 50% faster than a FX62 for $1200.

wow, how come i never think about? :) AMD frag AMD, at lower cost :) Allow me to immitate Sharikou style, their ASP goes way too low and AMD is going to bankrupt in 6 quarters. :) Just kidding. (i know i'm inviting attack here, hahaha)

9:18 AM, July 18, 2006  
Anonymous Anonymous said...

It would be one thing if Intel was this innovative, caring, and performance-centric company (much like AMD) and than I can see people following them, but supporting networst and hording over Intel after "OMG I HAVE A REDESIGNED PENTIUM 3 THAT DOESN'T RUN 64-BIT!!! WOOTZERLEET!!" is pathetic.

p.s. Ignorant Intel fanboys beware: 4x4 is not expensive!


ok, before the Core 2 duo, release, you play down on it and say it is hype.

when benchmark sites give review, you said they are intel pumper (inclusinve of AMD fan site)

when the product is out (some already has conroe), then you claim
the 4x4, or K8L WILL be better.

And then also label the Intel Core 2 Duo as old, P3 arhcitecture, blablabla and never realized that it beat AMD's supposing super duper architecure, by large margin and lower power consumption.

so, who is being ignorant here? Even if you strongly believe AMD's future product WILL be better, you should at least give the credit where it is due to the Core 2 Duo, as it is indeed current outperform AMD's CPU

9:28 AM, July 18, 2006  
Anonymous Anonymous said...

Intel advised that it will be at the low end of its target estimate earnings, so what does the BofA annalist advice; BUY with a target of $25.00 per share, stating Intel will regain lost market share 3Q06. BofA annalist is the one who must be composing the stupid fanboy posts on this blog.

9:32 AM, July 18, 2006  
Anonymous Anonymous said...

"When Intel cuts 1000 jobs, no one even mentions that AMD cut 1000 jobs too."

Maybe that's because the RUMOR of Intel firing 1000 people because reality, and the RUMOR of AMD firing 1000 people became a reality of ~180 jobs in the Geode division, and 75 of those people are being moved to a different area in AMD. Oh yeah, and Intel is planning on cutting a BUNCH more, rumored up to 10,000. AMD looks pretty confident, Intel looks scared.

AMD Geode cuts.

""X2 3800+" only have a SINGLE non-ccHT link. This also applies to FX processors. Only high end Opterons have ccHT links.

Thus I don't see how 4x4 can be used on these processors."

They can't. It is physically impossible. "

Maybe you should read this before you look like more of an idiot Intel fanboy.

10:19 AM, July 18, 2006  
Anonymous Jeach! said...

I never thought that Intel fans could be such morons and receptive to great new technology.

First of all AMD is improving their processors! A little late in my opinion, but it will come eventually. As for the 4x4, its probably some employees pet-project which everyone found really cool and easy to produce. Why NOT build a dual-socket MB for the extra few bucks? As someone said, they will probably all be made like that in the future.

Second, Intel's Conroe is a HACK! It's that simple! It only scales to dual core due to their shared cache, apart from that it couldn't do it. And the worst thing is that it is NOT economically feasable for a company to keep producing processors with that much shared cache. Do you think they will have 2MB shared L2 cache on a 4-core/1-die processor? If they do, it will be unprofitable!!

Until they come up with a real solution, they revert to large shared caches.

Note that I'm not dissing shared caches, just 'LARGE' shared caches. AMD will also utilize shared caches in future processors.

10:44 AM, July 18, 2006  
Anonymous Anonymous said...

"/*

Q1. I notice you make little mention on the fact that AMD both lost 1000 employees from Geode for the most part.

Q2. When Intel cuts 1000 jobs, no one even mentions that AMD cut 1000 jobs too.


MORRONS !!!


A. We got some clarification from AMD on the Geode job cuts, and it looks like the numbers aren't as big as we initially heard. The official word is that the Longmount Design Center is being 'phased out', and about 180 people are affected. Of the 180, 75 or so have new jobs at AMD waiting if they want them, and the others are on the short list for any new AMD jobs.

10:46 AM, July 18, 2006  
Blogger TheKhalif said...

When Intel cuts 1000 jobs, no one even mentions that AMD cut 1000 jobs too.

The Ing recanted and stated it was only 78 jobs from Geode.


To those who think tha FX has 3 links it doesn't. According to AMDs site ALL 93 chips have 1 HT link. Thisi snot a problem though because HT connects to IO using DCA.
By placing the chipset in between the two chips with 2 separate HT ports both chips have dedicated access to RAM.

I'm tired of you 3rd graders. i disagree with Sharikou about some things but I won't be name-calling.

He can have his opinion.

Anyway witht he 500 series chipset a simple reroute is needed so that the chipset acts as the arbiter(crossbar).

This maybe more efficient than Opteron 2xx if they limit the disance between the chips and chipset.

It will be likely that the board will ahve only the necessities (1 PCI) 2 16x PCIe, 1 4x PCIe, 2 USB 2.0 port, 1 IDE, 1floppy, 4 SATA.

They could actually start with a MicroATX board and then add on as necessary.

This will not be a wksta. Wkstas NEED ECC RAM.

It will force Intel to do soemthing in response.

I don't know if they will consistenty get 80%, but if those weere the first tests I can only see it getting better. Maybe not 100% on the current tested apps but higher on ones that aren't getting that increase.

Also rumors about RHT are coming around again while AMD has yet to confirm or deny its existence.


One of their Analyst Day slides shows that addign extra transistors will enable this tech, so I can't say if it will come before 65nm.

Knwowing AMD it will come first for Opteron and then later for X2, or X4. People think that AMD is goign to jump right to 65n K8L.....NOT. They wildo X2 65nm then X4 65nm then K8L.

I could be wrong but that seems to be AMDs modus operendi.As far as comparisons to Intel. There are none. Intel makes CPUs, AMD makes platforms. Intel lives on an island alone and AMD has partners.

And for the peopel with the consortium comments have you ever heard of JEDEC or PCI SIG?

AMD is smart. WHy do you think Intel didn't go Itanium alone? HP and other companies have done as much work as Intel.

11:43 AM, July 18, 2006  
Blogger Entrophos said...

Let's compare the two on a very simple level and you can see the differences in approach to performance improvements.

Conroe:

- Heavily optimized for 32-bit code at the expense of 64-bit code (based on existing benchmarks)
- Dual-core is more functional than P4's implementation
- Huge shared cache to relieve latency issues with the off-die memory controller and FSB

Pros - Runs existing applications that are not heavily SMP and/or 64-bit very well and beats AMD K8 chips clock-for-clock in many of these applications by a decent margin

Cons - 64-bit is less robust and SMP scaling is questionable. The future performance of this CPU is questionable depending on if current trends continue. Still runs hot despite the shrink to 65nm.

Conroe is a brute-force approach to running existing 32-bit single-threaded codebase very well. And it does excel at this, there is absolutely no question. Intel specifically designed the processor to shine in this environment. They want to recapture the performance crown today at all costs because to be honest, they needed to to stave off more losses to AMD.

The processors will be reasonably priced and the boards will be affordable as well. Obviously the 4MB cache models will be at a premium due to the large real estate due to the larger die size.

However, expect Conroe to suffer when more applications move away from single-threaded 32-bit support. Conroe is designed to blow the socks off most existing applications today, but it's not future-proof if the current trends in software development continue.

Still, if you must upgrade now and only care about performance in the short term and want to run your existing software fast, it's a good choice.

AMD 4x4:

- Relies on low-cost dual-socket SMP to boost performance on applications
- Very robust 64-bit support and performance
- Can use one processor to save money with the option to pop in another to get a performance boost
- Will benefit greatly from the movement of software to optimize for SMP/64-bit
- Torrenza might work with this chipset allowing special co-processors to boost specific functions that are deemed important to the user

Pros - Scalable, possibility for dedicated co-processor chips using Torrenza as an option instead of a second processor. Future-proofing in the form of excellent SMP implementation and strong 64-bit support. Should take advantage of coming advances and trends in OS/software development to gain even more performance.

Cons - Power consumption will be higher than a single CPU system. Performance won't improve much on applications that aren't SMP/64-bit optimized which includes many of today's programs and Windows XP 32-bit

AMD is banking on the current trends of software to take advantage of multi-threading and 64-bit extensions to improve performance as opposed to raw megahurtz and IPC of single-threaded CPUs. 64-bit and SMP are their two big strengths in the server world and they are bringing them to the desktop arena to combat Intel's muscle with finesse and the promise of future-proofing and better performance with tomorrow's software.
Expect applications optimized for 64-bit and SMP to allow the performance of 4x4 systems to surpass Conroe-based systems.

Although detractors will say this will be more expensive for the end user, it shouldn't be as the AMD CPUs have less real estate on silicon wafers and can be produced cheaply and sold at a reasonable price. Hence the smart move by AMD to move to cores with less cache so they can be produced more cheaply and sold at a price where they can recapture the price/performance crown and to sell the 4x4 platform to the budget-conscious. Most speculation is that 4x4 motherboards will not carry a huge premium over single-socket boards, further bringing down the overall price.

Expect 4x4 to be even more attractive from a price/performance standpoint with the shrink to 65nm and further reduction of TDP.

...........

Personally, I am not upgrading from my current system until I see how things play out but I am extremely interested to see how 4x4 plays out and if they can bring Torrenza support to it. I can imagine plugging in a dedicated Physics PPU into one of the sockets and getting some real bang in games due to the HTT access to main system memory, the CPU and the graphics subsystem. This is what AGEIA should have waited for instead of releasing their PCI-based card as it doesn't do their product justice.

11:55 AM, July 18, 2006  
Anonymous Anonymous said...

where are these stability concerns regarding Conroe arising from

They arise from a unique 4MB shared cache architecture and a slow ramp that has been recently reported to be for the desktop space 10% in Q3 and 30% for Q4.

On the one hand, you could say the reason for the slow ramp is to ensure quality and stability for consumers, but arguing the other way, you could say review web sites got only the very best samples.

Regardless, the true metric of success is how soon a large number of end users can purchase a 4MB-shared-cache E6600 Conroe for $350. Any issues with the average production sample will come out soon after.

12:40 PM, July 18, 2006  
Anonymous Anonymous said...

"Hence the smart move by AMD to move to cores with less cache so they can be produced more cheaply and sold at a price where they can recapture the price/performance crown and to sell the 4x4 platform to the budget-conscious."

4x4, as expressed by AMD, is for the ENTHUSIAST market, not the mainstream.

So my guess, based on that language, is that it will not be cheap to get yourself a 4x4 system.

With only one processor having a link to memory, you get half the bandwidth of a real 2P system.

And what great benefit is having a custom co-processor in a Socket 940 vs. just plugging a co-processor into a x16 PCI-Express slot? The 8GB/sec that x16 gives you is pretty good for most things. And then your co-processor can be moved to Intel, another machine without 4x4, etc.

This whole 4x4 thing begs the question... why doesn't AMD just make Opteron 2P cheaper?

Then everyone could afford a real 2P system. Heck, throw a couple extra sockets onto that 2P system for 2P + 2CP, i.e. two extra coprocessors.

Anyhow, I'm looking forward to something other than the 4x4 vaporware that AMD has offered so far.

1:07 PM, July 18, 2006  
Blogger C2 said...

I'm completely agree with "Entrophos".

At last we'll have a powerful & scalable system solution!
As to the cars and engines, some comparisons up are funny and pointless because of minimum 2 reasons:
1. If you can to buy "Ferrari" there is amazing computer systems as performance, stability and ... $$$$$$, but it's for a “Ferrari”, isn't it?
2. About the cylinders & engines - in many cases 2 engines are the proper solution, not 1 engine - 2 times powerful!
And the most important - the analogy is certain way for someone to fall into error.

Please, ... just for a while ... think! WHY AGAIN THE INTELs CPUs ARE WITH SO MUCH (4MB) CACHE AND

SO QUESTIONABLE PERFORMANCE?!
At last about these who still think that to push some "benchmark-fit" shits with a questionable performance, advantage and stability, is a right and positive way...they are non-recoverable case. :)

1:25 PM, July 18, 2006  
Anonymous Edward said...

"To those who think tha FX has 3 links it doesn't. According to AMDs site ALL 93 chips have 1 HT link. Thisi snot a problem though because HT connects to IO using DCA.
By placing the chipset in between the two chips with 2 separate HT ports both chips have dedicated access to RAM.
"

I don't think it is a good idea, though. An additional chipset arbiter introduces extra latency to both memory access and inter-processor communication. Also from AMD's slide 4x4 clearly has separate links to memory banks from both processors. It won't be true if there is one chipset in the middle to arbitrate.

If 4x4 is indeed layed out as you said, it will not be NUMA, but just a 'hierarchical' quad-core SMP spreaded over two sockets. That looks more like Intel's Core 2 quad-core (although splitted to two sockets) than the 4x4 I had in mind.

Also, the reverse hyperthreading rumor is not true - I remember reading it confirmed false by either the Inquirer or some other website previously spreading the rumor. Although theoretically it is possible to spawn a side-kick thread, in hardware, to help the main processing, I seriously doubt AMD has implemented & tested the detail. RHT actually seems more natural with Conroe's shared L2 cache; and for AMD, it seems like a K8L or post-K8L improvement.

1:27 PM, July 18, 2006  
Blogger Howard said...

To those who think tha FX has 3 links it doesn't. According to AMDs site ALL 93 chips have 1 HT link. Thisi snot a problem though because HT connects to IO using DCA.
By placing the chipset in between the two chips with 2 separate HT ports both chips have dedicated access to RAM.

Correct, according to the AMD Product Data sheets, http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/33425.pdf
all Athlon64 processors have only one HT link.

Your comment about dedicated access to RAM is orthogonal - since each CPU has on-die memory controller, *of course* they always have direct access to RAM. However, that's dedicated per-CPU; without ccHT I don't see how the two CPUs will get access to each other's RAM or maintain cache coherency. This is the most interesting question about 4x4, IMO. For it to work, there has to be a ccHT link. Since AMD says it *does* work, then they must have enabled ccHT on the AM2 chips. That really brings the AM2 into Opteron territory though. While it may be a useful move, that further obscures the distinction between A64 and Opteron.

As for ECC or not - even A64 memory controllers support ECC, that's not a distinguishing feature. What distinguishes Opteron from A64 is use of registered vs unbuffered DIMMs. I think this is an important feature, that needs to get more attention. My X2 3800+ system has 2GB of DIMMs installed, and I see bit errors in my data once every couple of weeks. I'm going to swap in ECC DDR DIMMs. It's just crazy not to use ECC. With random cosmic-ray induced bit error rates and everything else to deal with, and 16 gigabits in play, it's just a fact that you'll have memory errors on an ongoing basis, workstation or not.

1:34 PM, July 18, 2006  
Anonymous Gandalf said...

Personally, I am not upgrading from my current system until I see how things play out but I am extremely interested to see how 4x4 plays out and if they can bring Torrenza support to it. I can imagine plugging in a dedicated Physics PPU into one of the sockets and getting some real bang in games due to the HTT access to main system memory, the CPU and the graphics subsystem. This is what AGEIA should have waited for instead of releasing their PCI-based card as it doesn't do their product justice.

Torrenza is a really interesting technology. The con on Tornenza is the needed (and propably lacking) support in software for (lots?) of "coprocessors". the mentioned AGEIA is a good example for this problem. the hardware is capapble of extremly high performance - but the few games that are actually use AGEIA aren't able to show this performance... :(

2:03 PM, July 18, 2006  
Anonymous Edward said...

"Anyway, FSB (and no IMC) is not a bad design as of its target market. It serves nicely for 1P to 2P market."

I don't agree.

FSB started back when memory speed was about the same or just a bit slower than CPU speed.

Ever since CPU clock rate became 2x faster than the memory access time (i.e., Pentium-II), FSB has been the bottleneck to realife applications. Since then FSB was life-supported by larger cache and higher memory bandwidth. Both on-die cache size and memory bandwidth increased ~16x, from the first P-III to the latest Core 2 Duo, while the CPU clock rate increased only 6x.

OTOH, Athlon64 X2 with cache size only 4x the original P-III performs only 15% less than the latest Core 2 Duo. Tell me that FSB is really a good design!

As for the 1p/2p market you're saying, IMO it is going away. The phase-out is already started with dual-core processors - now a 1-socket is actually the previous 2p and 2-socket the previous 4p. With FSB, Intel had to glue the 2 cores with a huge shared L2 cache, which granted was a good engineering. But what happens when it goes to dual-socket 4-core or 8-core? We're looking at 4 to 8 simutaneous memory access requests to one FSB, and 8 to 16 requests to one chipset. How is that going to perform? And I thought Intel intended to go 32-core by 2010. Tell me FSB is not going to obsolete in 4 years!

It is not from one single aspect, but numerous combined, that many believe (and I bet Intel itself does, too) that FSB is at the end of its life. People will be looking at direct connect, NUMA, and more distributed & multithreaded architecture to optimize performace. Of course, you won't care if you just buy a Core 2 Duo and feel happy for 2 years. But be sure that up until now it is AMD who is pushing foward the evolution wheel of x86 microprocessor architecture.

2:34 PM, July 18, 2006  
Anonymous Edward said...

"My X2 3800+ system has 2GB of DIMMs installed, and I see bit errors in my data once every couple of weeks. I'm going to swap in ECC DDR DIMMs. It's just crazy not to use ECC."

Nice... I think I'll take your advice and do that for my box, too. ;-)

2:37 PM, July 18, 2006  
Blogger Mad Mod Mike said...

"Correct, according to the AMD Product Data sheets, http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/33425.pdf
all Athlon64 processors have only one HT link."

Please read it yourself.

"HyperTransport technology TO I/O DEVICES: 1" - Says nothing about ccHT links.

2:40 PM, July 18, 2006  
Anonymous Edward said...

"The 8GB/sec that x16 gives you is pretty good for most things. And then your co-processor can be moved to Intel, another machine without 4x4, etc."

The issue here is latency, not bandwidth.

A 4x4 is not a 2p Opteron. They are designed with different target in mind. For one, I see no reason for Opteron servers to have dual PCI-E. Most servers don't even have a monitor. And, for 99% workstations, I see no reason to have more than one on-board GbE, or buffered RAMs, or full RAS as the servers require. So 4x4 is clearly aiming at enthusiasts AND workstations market.

Less than 2 years ago I heard people swearing right in front of my face that Pentiums going to 10Ghz will beat anything. I just laughed. They couldn't understand how an Athlon64 with slower clockrate can actually be faster. It's completely due to Intel's mischievous marketing - Intel engineers, it seems, were totally aware of the impracticality of the roadmap from their colleagues in the marketing department.

IMHO today many pro-Intel people are missing another evolutional trend: ccHT (or something alike). Core 2 Duo is a nice chip, no doubt, but frankly it's one of the last breeds on its genre. Direct connect, multithreading, and distributed processing is the future. I bet when CSI becomes ready, which would be earlier than we all expect, Intel's platform architecture will be very similar to 4x4, and Intel will use all its muscles and tactics to *encourage* every software vendor to adopt and force all consumers to upgrade. If you like Intel's forceful and late approach, but dismiss AMD's open and early innovation, I know no other word to call you than an Intel fanboy.

3:08 PM, July 18, 2006  
Blogger Howard said...

Mad Mod Mike you ignorant slut:
Please read it yourself.

"HyperTransport technology TO I/O DEVICES: 1" - Says nothing about ccHT links.


Of course it says nothing about ccHT links, because there are none, there's nothing to say about them. Want more proof? Try this document, section 2.4.1, table 1, page 11.
http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/31412.pdf

Again, clearly the A64 family has only one HT link, and that link is non-coherent. If they decided to make an exception for the AM2 socket, they would have called it out specifically when they added AM2 details to the A64 product data sheets. They made no such doc change, because they made no such CPU change.

5:34 PM, July 18, 2006  
Anonymous Anonymous said...

Sharikou said: "Two X2 3800+ EESF, 35 watts TDP_max, total 70 watts TDP_max. Frag conroe XE6800 at lower energy and lower cost by 30%."

Sharikou, isn't this bait-and-switch? Your performance comparison has the FX62 as the baseline, not a 2GHz, 512KB cache entry-level X2 part.

So let's do the math again:

X2-3800+ = 100%
FX62 = 131%
Conroe = 115% FX2 = 151%
4x4 = 180%

Now the best case performance for 4x4 is only 19% (180/151=1.19) over Conroe. Easily within reach of overclocking. Nevermind that most games will be lucky to see 150% scaling on 4x4. Also keep in mind that all your single-threaded applications are stuck with an entry-level core.

Let's face it. AMD cannot sell workstation performance at desktop prices.

6:31 PM, July 18, 2006  
Anonymous Anonymous said...

While it may be a useful move, that further obscures the distinction between A64 and Opteron.

For the same reason the military uses (or used?) the MK23 instead of an ordinary USP, so too will grid-iron servers use Opterons instead of Athlons. Same caliber, nearly the same gun, but they excel at different things.

9:08 PM, July 18, 2006  
Blogger Mad Mod Mike said...

"Again, clearly the A64 family has only one HT link, and that link is non-coherent. If they decided to make an exception for the AM2 socket, they would have called it out specifically when they added AM2 details to the A64 product data sheets. They made no such doc change, because they made no such CPU change."

I see the blatant insulting that Intel Fanmonkeys resort to to get their points, albeit rare, across. That tells me the Socket 940 Opteron64's & original FX-51 & FX-53 have 1 link, says nothing about Socket 939 or Socket AM2. Desktop can refer to what I have stated. Please find me one that refers to s939 or sAM2 and Dual-Core CPU's specifically; I also ask you do it in a non-retard fashion, e.g., no insults.

If you can produce this, I will gladly listen.

9:22 PM, July 18, 2006  
Anonymous Anonymous said...

CLARIFYING POWER

There has been lots of B.S. that shows Intel operates at lower power than AMD. Anand shows things differently.

http://anandtech.com/cpuchipsets/showdoc...

Athlon X2 is comparable to Intel "E" Series, not "X" Series. This is where over 95% of the unit volume is sold in the market.

Athlon X2 consumes 13% less power at idle and 3% less at full load. PCs are idle 95% of the time and do not run benchmarks all day. They run real software programs. Try it for yourself. Do Ctrl-Alt-Delete and then minimize task manager. If you don't see the little green bar go up then your CPU is idle, which is most of the time.

Don't be duped.

3:59 AM, July 19, 2006  
Anonymous Big Green Kidnapper Van said...

"If you like Intel's forceful and late approach, but dismiss AMD's open and early innovation, I know no other word to call you than an Intel fanboy."

4x4 is vaporware. It's nice to talk about stuff that doesn't exist as "innovation", but until it is in the market, in real users' hands, etc., it is just meaningless fanboy speculation.

I can understand everyone's reticence to jump on board AMD's big innovation Hummer to the future. Why?

Just look at the full-on raging DUD that is AM2. Memory makers are scrambling to get DDR2-800 and DDR2-1066 out the door so this AM2 from AMD has some performance.

And guess what? DDR is still outselling DDR2 in the world RAM markets.

Which means AM2 is not just a dud, it means there is no point to it. Instead of doing right by customers and improving the existing chips, AMD pulled a trick out of Intel's bag and said... "hey, you pieces of dirt... i mean customers... time to buy all new motherboards, processors, and memory." For what? For a bona fide dud chip that uses dud RAM. Gotta love that *innovation*.

So what's the scoop with 4x4?

The same damn thing. New motherboards, likely new chips. Okay, maybe if you were foolish and hopped on AMD's move to DDR2, you can keep your RAM. Well, if it ain't ECC all your RAM is going to be crashing your computer and corrupting your data files. So you will likely need new RAM as well.

AMD is *saying* 4x4 is futureproof. Just like the BIG FUCKING LIE that AM2 was futureproof.

I see nothing that makes AMD better than Intel *today*. When the Opteron came out, it was clear that Opteron was a great design. But that was five years ago.

All that AMD has done in five years for making the architecture of their chips better is... NOTHING.... NADA.... DIDDLY.

So AMD, Intel 2.0 in the making, wants you to pony up the big bucks for their multi-threading cache-coherent innovation Hummer so you can go on the big digital Rubicon adventure to see great Vistas. Uh huh...

Hey there pretty fan boi, wanna go for a ride?

4:25 AM, July 19, 2006  
Blogger TheKhalif said...

Your comment about dedicated access to RAM is orthogonal - since each CPU has on-die memory controller, *of course* they always have direct access to RAM. However, that's dedicated per-CPU; without ccHT I don't see how the two CPUs will get access to each other's RAM or maintain cache coherency. This is the most interesting question about 4x4, IMO. For it to work, there has to be a ccHT link. Since AMD says it *does* work, then they must have enabled ccHT on the AM2 chips. That really brings the AM2 into Opteron territory though. While it may be a useful move, that further obscures the distinction between A64 and Opteron.

Not RAM, IO. The way that DC works it won't cause too much of a latency problem. All accesses happen through this in a 1 link setup.

I doubt if I'm wrong. Things like this are the point of HT.

The Analyst Day slides show the two chips in series going directly to the PCIe bus and then going to the chipset. I will look at the Opteron 2xx layout to see how it's done there.

10:14 AM, July 19, 2006  
Blogger Howard said...

mad mod mike:
I see the blatant insulting that Intel Fanmonkeys resort to to get their points, albeit rare, across.

You've either never watched Saturday Night Live, or you just have no sense of humor. By the way, I'm no Intel fanboy, I use A64 X2's every day in my work and would never go back to Intel. Before A64 came along I was a diehard Atari/M68K programmer...

That tells me the Socket 940 Opteron64's & original FX-51 & FX-53 have 1 link, says nothing about Socket 939 or Socket AM2. Desktop can refer to what I have stated. Please find me one that refers to s939 or sAM2 and Dual-Core CPU's specifically; I also ask you do it in a non-retard fashion, e.g., no insults.

If you can produce this, I will gladly listen.

Sorry, but you're just intentionally twisting words to make your point. AMD defines "desktop processors" to include Athlon64 and to exclude Opteron, that's the only definition that matters in this discussion.

You also appear to have very poor reading comprehension skills. Socket 940 Opterons all have 3 HT links, of which 1 to 3 of them may be coherent, depending on the processor series. Socket 939 chips all have only 1 HT link, and the S939 pinout only provides pins for 1 HT link.
The early Socket 940 Athlon64 FXs are desktop processors; they only have 1 HT link. Go look at the Socket 939 datasheet and compare it to the Socket 940 datasheet, it's all explicitly stated there.

The previous document I pointed you at is specifically the Athlon64 X2 datasheet, it doesn't get any more relevant than that.

I've done my research and presented you with valid evidence. Now you show me the data sheet that shows that you're right and I'm wrong.

thekhalif:
Not RAM, IO. The way that DC works it won't cause too much of a latency problem. All accesses happen through this in a 1 link setup.

I doubt if I'm wrong. Things like this are the point of HT.

The Analyst Day slides show the two chips in series going directly to the PCIe bus and then going to the chipset. I will look at the Opteron 2xx layout to see how it's done there.


You're missing the point. In a normal 2P Opteron setup, one ccHT link is used to connect the two processors to each other, so they can directly exchange data. The other links are used for IO devices.
The ccHT link allows the processors to address each other's RAM and broadcast cache updates to each other. So for a normal 2P setup, one processor needs to have at least 2 HT links - the ccHT to talk to the other processor, and at least 1 for use with IO.

The Athlon64 chips and sockets all provide electrical connections for only one HT link per chip. If the HT link is used to connect the two processors directly to each other, then there's nothing left to connect to IO. And of course, since the A64 HT link doesn't support ccHT, then the cache synchronization protocol can't be used. So even if they provide a single bridge chip that has two HT links of its own, so that both processors can be connected to it, that would only give them fast access to IO, there's no connection available for memory accesses. Also, even if they can run both memory and IO accesses over this single HT link, it will become a bottleneck, it only handles 2GB/sec in either direction, which is tiny compared to the usual RAM bandwidth you'd expect.

11:46 AM, July 19, 2006  
Anonymous Anonymous said...

Hey guy, you're a complete idiot.
If the 4X4-system frags the dual-core Conroe you admit that the Conroe-EE is faster than the fastest dual-core Athlon64. Thanks god you agree to that.

4:44 PM, July 31, 2006  

Post a Comment

Links to this post:

Create a Link

<< Home