AMD should create a quasi quad core
AMD should be able to do this with a blink of eye. Just connect the ccHT links on two Opteron dies and arrange to deliver power to the second die. This should be much easier to do than Intel's shared bus double die solution, where two dies generate two loads.
BTW, after I posted a criticism of Rahul's rather primitive way of demonstrating Kentsfield performance (running multiple SuperPIs), I found all my comments are gone.
While I am not saying there was a causal nexus between the two events, I do think, sometimes, you need to be a Doctor of Philosophy to be able to face criticisms and adhere to fundamental principles.
posted by Sharikou, Ph. D. @ 10/26/2006 06:59:00 PM
51 Comments:
I don't think this is possible.
4x4 sucks btw.
Thanks for the argument anonymous.
You're retarded by the way (yay logical deduction!!!).
It would be fairly simple, just give 2 banks of memory to each memory controller. But this would literally be 4x4 in a single package. From a distribution point of view, 4x4 is much more elegant, and that's what matters most to AMD.
All I want to know..
Who has the first quad-core.
If it could be done in a second why isn't it out there?
Dr PhD please pray tell why?
WHy isn't INTEL not 3 quarters away from BK already? Why are you revising all your claims... Because they are all false... LOL
How about quasi quad core with reverse hyperthreding on zram.
Not like your doing much more :).
Given I do not agree with everything you say Sharikou, I at least get where your coming from and find many valid points in your statements. Pentium is exactly what you have said, a dead brand, and they have been riding a dead horse for a few years now. They should have cut their losses and developed a core like architecture a few years ago, its too close to AMD's next generation processors for it to have lasting effects like AMD has had. AMD has built of a reputation for the king of gaming, even if its currently not true, Core 2 does NOT have that reputation with people.
lol, I love your comment about being a PhD sharikou, it keeps me laughing. Sock it to them!
Jori said...
"AMD has built of a reputation for the king of gaming, even if its currently not true, Core 2 does NOT have that reputation with people."
Maybe not yet, but given the fact that it will have about 8 months (plus the last few) without something from AMD to compete with, I would think it is fair to say things will change.
People who want a new gaming system now will go Core 2 Duo, or do you disagree?
I have Athlon 64 3200 CPU. I'm a gamer sometimes. The only chip I upgrade in the future is X2 393 chip.
I'm just waiting for 4800+ to cost 100$.
You're too late Sharikou. They have already created one :) here
Oh dear! Bad news! Inquirer said AMD can't produce enough chips! No exact launch dates for 65nm chips except AMD vaguely says Q4 2006, but its already Q4 2006, when will it be? Two months left November, December? An AMD representative declined to comment on the die size for its first 65-nanometer products. No engineering samples or official demos. Is AMD's 65nm SOI in deep trouble?
"People who want a new gaming system now will go Core 2 Duo, or do you disagree? "
They won’t because Core 2 Duo isn’t that cheap. I'm better off with some AMD 3500+ or 3800+ and OC it a little , and of course a better GPU.
With the price of the cheapest Core 2 Duo I buy the AMD processor + motherboard. And besides that I get future upgrades to Dual core and Quad when they get cheaper.
When the Core 2 Duo (full version/4MB) not the one with the disabled cache gets into the price of the current cheapest version I may consider it.
quasi quad from amd is not really possible. second die will have to access memory through ccHT and IMC of first die, ie it will be terrible slow.
enumae said...
"maybe not yet..."
I have to disagree with you. Every serious gamer I know that doesn't have a sick infatuation with AMD has already bought or is planning to buy a Core2Due soon. The performance gap is just to big to ignore. Kentsfield is going to increase that even further. AMD's 4x4 will barely compete with a single Core2Duo system which isn't even Intel's extreme product by the time it comes out.
The shift to Intel has happened. Intel doesnt need 8 months.
Nobody wants a sluggish, very hot, unclockable Atlon.
Good to hear the phd talk about his doctorate. Surely no one can still doubt he's the real thing :-)
enumae said...
"People who want a new gaming system now will go Core 2 Duo, or do you disagree? "
Only if they are clueless...
TRUE gamers don't play at 640x480 or even 1024x768 with low detail turned on.
The fact is there is barely ANY DIFFERENCE in performance between mid-range vs. Hi-end CPUs when running games at Higher Resolutions with all the video features turned on. So everyone that pays more than $300 for a CPU is an idiot. Unless, they like to brag about how many seconds it took to run Super Pi. Spend it on your Video cards and shutup... btw, AMD now has better Gaming GPUs then Intel, so nah, nah, nah... :)
enumae said...
"People who want a new gaming system now will go Core 2 Duo, or do you disagree? "
Only if they are clueless...
TRUE gamers don't play at 640x480 or even 1024x768 with low detail turned on.
The fact is there is barely ANY DIFFERENCE in performance between mid-range vs. Hi-end CPUs when running games at Higher Resolutions with all the video features turned on. So everyone that pays more than $300 for a CPU is an idiot. Unless, they like to brag about how many seconds it took to run Super Pi. Spend it on your Video cards and shutup... btw, AMD now has better Gaming GPUs then Intel, so nah, nah, nah... :)
Your comments are still there it looks like Sharikou, it seems Rahul just put your moniker/name on each of your posts as anonymous instead. Crafty, but unfortunately due to your preceding reputation the drooling Intel trolls, perhaps a wise move after all to keep the flamebait to a minimum.
To the fool who thinks 4x4 sucks reason this:
Would you rather have a slow, outdated FSB running through a northbridge controlling the CPU requests and memory bandwidth for 2 sandwiched dual core processors on a single socket who are already burdened with cache thrashing?
Or would you rather have a fast, point to point interconnect with a memory controller directly accessing memory bandwidth at full processor speed to each dual core processor in a seperate socket?
The obvious choice is clear.
As glorified as Intel(and it's legions of fanboys) tries to make Kentsfield, it's just as much a stop gap solution to true quadcore as 4x4 is, PERIOD. A MCM dual core processor "sandwich" is not what engineers refer to as a TRUE quad core solution and because of that, it's performance is going to be overshadowed by any native quad core processor and even dual-core processors in 2P solutions. The bad thing about it as well, even when Intel will get a native quad core out, 4x4 will still allow for future quadcores in the future, adding even more scalability on AMD's part up to 8 cores in the high-end desktop/workstation platform.
Semantics of who gets a "psuedo quad-core" processor first be damned; do it right, do it natively and do it right the first time is who gets my money.
enumae said...
"AMD has built of a reputation for the king of gaming, even if its currently not true, Core 2 does NOT have that reputation with people."
Maybe not yet, but given the fact that it will have about 8 months (plus the last few) without something from AMD to compete with, I would think it is fair to say things will change.
People who want a new gaming system now will go Core 2 Duo, or do you disagree?
Are you forgetting about 4X4?
If you choose to answer, make sure that you truly understand what 4X4 really is:)d
has anybody came across this article yet... "Analyzing Efficiency of Shared and Dedicated L2 Cache in Modern Dual-Core Processors"
http://www.digit-life.com/articles2/cpu/rmmt-l2-cache.html
the L2 cache on C2Ds are very inefficient.
so 4x4 just may be able to take advantage kentsfield's weaknesses...
i think K8L will really raise performance to a whole new level unreachable by intel.
Sharmeister,
I'd much rather hear you figure out a way to make a single application work faster with multiple cores.
Let us know when you figure that out!
Anonymous said...
"Spend it on your Video cards and shutup... btw, AMD now has better Gaming GPUs then Intel, so nah, nah, nah... :)"
lol!
------------------------------------
To those who answered my question about upgrading, think about this...
1. Most of the Core 2 Duo motherboards above $150 support Kentsfield.
2. The E6600 at $319 plus a little overclock, a decent motherboard, and a decent Heatsink would be less than the FX62, by about $120, and have better performance.
Thats $120 towards the X1950XT, or a 7950GX2.
I have to believe that people will be moving to Core 2 Duo, marginal gains or not the performance is better.
Hey, isn't that what we all want anyways?
Anonymous said...
enumae said...
"AMD has built of a reputation for the king of gaming, even if its currently not true, Core 2 does NOT have that reputation with people."
Actually I didn't say that.
"Maybe not yet, but given the fact that it will have about 8 months (plus the last few) without something from AMD to compete with, I would think it is fair to say things will change.
People who want a new gaming system now will go Core 2 Duo, or do you disagree?"
That I did say :)
"Are you forgetting about 4X4?"
Actually no I am not, but keep in mind enthusiast are a very small segment, then factor in of that small segment how many can afford 4x4, then think about who would do it if they even had the money.
The number just gets smaller and smaller.
"If you choose to answer, make sure that you truly understand what 4X4 really is:)d "
Thanks, but I think I get the idea of what 4x4 is
It's not made for gaming, more for "mega tasking", of which a gamer may or may not be but lets look at the price vs Kentsfield, as that will be my deciding factor.
4x4 = About $1000 for processors (low end 2.6GHz), say about $300 for a motherboard.
And then what kind of power supply do you need for the 2 x 125W FX chips? $100?
Next is windows XP which is sold on a per socket basis, say you have a copy, well you still need another one $150
So AMD 4x4 = $1550 MB, PR, PS and OS.
Kentsfield = $1000 PR, $150 MB and a decent PS $75, and you already have your copy of windows...
So Kentsfield = $1225, that leaves you with enough money to buy 2 Gigs of some good low latency ram.
Whats your take?
quasi quad from amd is not really possible. second die will have to access memory through ccHT and IMC of first die, ie it will be terrible slow.
Who knows?
"MSI didn't give the second CPU local memory. Despite this limitation, the MSI K8Master-FAR2 proved to be an excellent performer in our database server tests."
http://www.anandtech.com/printarticle.aspx?i=2447
All AMD has to do now is wait for Intel to announce their QuadCore Processor on November 2.
Then when AMD announces their 4X4 on November 3/4 also show off their A1 True-QuadCore processor. Even if it is still just a demo version in a server box. This should take the wind out of Intel's "quad-core-sails".
People are still deciding whether to pony up dough for the Core2Duos, and will probably wait a few months for the PC maker choices to appear. Then they won't be far off from receiving AMD's True-QuadCores.
So AMD 4x4 = $1550 MB, PR, PS and OS.
Kentsfield = $1000 PR, $150 MB and a decent PS $75, and you already have your copy of windows...
You pay for what you get. Kentsfield is a double die chip, each core has 266MHZ bandwidth, less than a Athlon XP with 333mHz bus.
Sharikou said...
"You pay for what you get."
Great point.
It will be expensive though :)
"MSI didn't give the second CPU local memory. Despite this limitation, the MSI K8Master-FAR2 proved to be an excellent performer in our database server tests."
http://www.anandtech.com/printarticle.aspx?i=2447
Exactly. In any case, a dedicated memory controller is much much better than the Intel bus which can't even handle the memory.
"Kentsfield is a double die chip, each core has 266MHZ bandwidth, less than a Athlon XP with 333mHz bus."
We shall see how much slowe/faster is the ancient FSB compared to uber high-end cHT really soon.
I wonder what would people say if a CPU sitting on old and slow FSB is beating CPU's sitting on several times faster HT link.
You love AMD when it comes to synthetic benchmarks!! But for the rest ....
Will be fun to see how this plays out in a quad core system. Extrapolate....
http://tinyurl.com/y3g9zh
http://www.tomshardware.com/2006/10/26/intel_woodcrest_and_amd_opteron_battle_head_to_head/
AMD should be able to do this with a blink of eye. Just connect the ccHT links on two Opteron dies and arrange to deliver power to the second die.
Theoretically this is true. However, practically you'll need twice the number of memory access pins on the socket, or one die will have to access memory only via the ccHT link.
The problem with this approach is two-fold: First, performance will not be great since you are dividing the memory bandwidth by two dual-core dies. Second, the MCM module may not worth it if the sales volume is low, and with 4x4 AMD could still sell two FX chips separately.
Kentsfield also shares the first problem, and partially the second. But I guess Intel has much more resource to waste than AMD simply because its production volume is 4x greater.
Front Side Bus
Yes we will have to see how the ANCIENT FSB on the Kentfield performs against AMD AMAZING IMC.
In Other News:
[ironicmode]Intel test how the FSB cripples the whole prosessor[/ironicmode]
http://www.legitreviews.com/article/393/2/
I quote:
When running Ice Storm Fighters on a quad-core processor (we assume it was the QX6700) with a 1066MHz Front Side Bus (FSB) vTune showed that 13% of the FSB was being used. This goes to show that even on intense upcoming games/benchmarks that are heavily threaded don't require a higher front side bus. Why would Intel point that out?
Please publish it Sharikou, orelse I will have to go to sharikou180degrees to post it.
I wonder what would people say if a CPU sitting on old and slow FSB is beating CPU's sitting on several times faster HT link.
Some would say "But it's not an elegant solution!" and would go with the slower, more elegant solution. Why? Because to some people, principle or cool factor matters more than performance.
It's kind of like comparing a Dodge Viper (brute force V-10) with a Ferrari (highly tuned V-8). Both go fast as hell, but sometimes the brute force approach wins on raw power. Depending on your application, get the engine that makes the most sense. For most well-threaded apps, Kentsfield will be king for a period of time. At some point, AMD may take the crown back. We'll have to wait and see. K8L/Barcelona is still idle speculation on performance and a minimum of 8 months away.
"I wonder what would people say if a CPU sitting on old and slow FSB is beating CPU's sitting on several times faster HT link."
It depends on the workload. If an app requires intensive memory access, then Core 2 at quad core is easily FSB limited, whereas for apps such as SuperPi FSB is seldom execersized.
On the cryptographic application that we run (which uses standard OpenSSL), K8-939 ties or beats Core 2 on clock-per-byte for almost all tests, except ones that are optimized for SSE2 (where Core 2 performs about 2x better). Under 64-bit, K8's lead is even more (up to 40%). Netburst is no contest here, in average only 1/3 speed at the same clock cycle.
Cryptography is a typical example of highly computational, high memory bandwidth application. Thus Core 2 can compute Pi very fast, but it cannot verifies the authenticity of your messages as well. Tell me which one is more important.
Of course, there is no doubt that Core 2 has the better SSE2 implementation - which also manifests itself in some media codec benchmarks; that is, if they are not bandwidth limited (e.g., if they use highest computation setting, resulting in lowest data throughput).
Theoretically this is true. However, practically you'll need twice the number of memory access pins on the socket, or one die will have to access memory only via the ccHT link.
There is a low end 2P MSI Opteron board with only one bank of memory.
I wasnt talking about hardcore gamers that know things. Im talking everyday people who get $1000 and want to get a new comp, they have heard somewhere I gaurantee that the Athlon64 is better than Intel for gaming, and buy it out of sheer ignorance. I had 3 guys I worked with that still thought Athlon64 was better than Core2, that shows alot I think, they considered themselves fairly "computer" literate.
Ok so we get it. Intel's Core is beating the K8. Being that this is the case, I expect to see a lot FUD being spread on this blog for say the next 6-8 months. I can see it now, the K8L is 300% faster than the Core. The K8L's superior 64bits equals 10X the performance of the Core and yes an occasional reference to an ad from Sun stating that the Opteron is 80% more efficient than the Woodcrest. Oh and of course the 4X4, I can't even imagine what crazy FUD will come out of that :)
"When running Ice Storm Fighters on a quad-core processor (we assume it was the QX6700) with a 1066MHz Front Side Bus (FSB) vTune showed that 13% of the FSB was being used."
This "vTune" result is pure crap from simple logic. Why? Because if the 13% had any truth, Kentsfield wouldn't have needed 1GHz FSB at all. 533MHz or even 266MHz would've been amply sufficient.
In any rate, the absolute value is meaningless without a comparison. How much better does it run on 533MHz FSB (with supposedly only 26% usage) vs. 1066MHz? How much better does it run against a Core 2 Duo, or a same-clocked Athlon64 X2?
Read this
http://news.zdnet.com/2100-9584_22-5895002.html
Mr. Rahul Sood knows computers and business. He is a good man.
His company sells both AMD and Intel based systems. If anything, it would seem he would rather work with AMD.
Why then would he promote Intel's chip wantonly?
Answer? He's not. He's being honest when he praises this chip.
A previous poster mentioned he did not delete Sharikou's post. He merely made them anonymous.
Doctor Sharikou, if you want to criticize Mr. Sood's test, PLEASE do you own unbiased tests with equally configured systems to enlighten us all.
Until then, I'll trust those words spoken by Mr. Sood.
Doctor Sharikou, if you want to criticize Mr. Sood's test, PLEASE do you own unbiased tests with equally configured systems to enlighten us all.
I think user experience is a valid test, which I did in comparing Core 2 Duo's responses to user interaction. Rahul is of course a good guy with good insight -- like he predicted the DELL Alienware acquisition. But I found him sometimes a bit defensive of his views. (BTW: I knew HP would acquire Voodoo after DELL acquired Alienware, but that wasn't a surprise).
WHERE IS AMD's QUADCORE is all I want to know! Where is it? How come there are NO Barcelona's in the wild. IF AMD is so great and is kicking INTEL's A$$ where are they quadcores? Where where where.
I'll tell you where they are all non-fuctional garbage.. Back to the drawing board AMD.
INTEL BK in 2008 remember that!
Sharikou: "I think user experience is a valid test, which I did in comparing Core 2 Duo's responses to user interaction."
Well lets just say that your extreme bias towards AMD will probably negate the validity of the test.
Gawd, they said it with Core2 - "yer gonna saturate a-tair front side bus with dem screamin' fast cores!!" And "they" were wrong, yet again. FSB bottlenecks Core2 performance - just because you keep repeating it doesn't make it true.
Yet again, Intel proves it has smart engineers and lots of resouces. Intels quad core is dominant. Mr. Sood just sold his very successful company for a bundle. Mr. Sharikou runs a very humorous blog, kinda low on the success-o-meter.
What kills me is the even the antiquated PDP processors took some benchmarks in media encoding away from current X2's of the time. People soon for get that PDP was competitive with X2, not superior, but competitive.
That AMD has no answer for the entire upper end of Core2 is not lost on performance-minded enthusiasts.
AMD can only compete with C2Q with a TWO SOCKET solution - doesn't that strike everyone as a little ODD?
Now, Sharikou, post my damn comment for a change.
This is all FUD. Exploding chips, 100 billion dollar company going bankrupt, they can't sell their old products, they can't sell their new products, they can't make enough of their new products to sell, no one will buy their new products, I can go on and on but one thing is for sure and that this is all FUDDDDDDD!!!!!!!!!!!
http://zquake.frag.ru/vansmiths/intel.htm
Edward said...
"When running Ice Storm Fighters on a quad-core processor (we assume it was the QX6700) with a 1066MHz Front Side Bus (FSB) vTune showed that 13% of the FSB was being used."
This "vTune" result is pure crap from simple logic. Why? Because if the 13% had any truth, Kentsfield wouldn't have needed 1GHz FSB at all. 533MHz or even 266MHz would've been amply sufficient.
You logic as has critical flaw: you assume the Intel CPU is used for the "Ice Storm Fighters" or alike apps only which in reality is not the case. The 13% FSB usage by the said apps is one example how much would a apps load the FSB. There will be apps or multitask that load the FSB higher or lower. I know you all like to say that when the FSB is fully loaded, it will be the bottleneck. But what we need to think is that, in a particular usage model, for desktop or laptop, how will this happen and how frequent it happens. The same goes with some server apps.
I'm not here to start a word fight with you again. Just wanna point it out to you that you cannot simply ignore the fact that under a lot of normal usage of the desktop/laptop and some particular usage of the server, the FSB is really not loaded that much and of no concern or its impact to the overall system performance is simply negligible.
With the increased FSB freq, Intel CPU can target more usage models ... but not all, put it in another word that is sweet to your ear, those heavy memory-reference apps. Neverthelss, just to re-iterate, desktop/laptop or even some server, do not really use those apps. Even if it is used, question is, how frequent, what os the % of its usage compare to the rest.
"AMD should be able to do this with a blink of eye."
Makes you wonder if it is really so easy why they have not done it...
How would both chips access offchip memory? With an IMC the socket would need significant mod to ensure both CPU's are being fed from DRAM (unless you are planning to do this through a die to die CPU interconnect within the package which would not seem to be very efficient)
How would both chips access offchip memory?
It is called ccHT or Torrenza technology.
"How would both chips access offchip memory?
It is called ccHT or Torrenza technology."
Through a common socket? Just wired in parallel with each other to communicate to memory? I understand how this is how 1 chip does that but would it really hold for 2 dies sharing 1 socket?
N4CR said...
"Huh.. it's a single socket dude. They ain't trying to beat preshott with 250w per socket."
What are you talking about?
4x4 is dual socket, each FX chip is going to have a TDP of 125W, hence 2 x 125W = 250W.
"If they can shoehorn quad 3ghz cpus into 125w tdp then they've done a bloody good job."
Again, what are talking about?
90nm + 3.0GHz + Quadcore = 125W...
That would be incredible, but its not going to happen, not at 90nm, maybe at 65nm.
Anonymous said...
http://zquake.frag.ru/vansmiths/intel.htm
Gee, is that Sharikou from six years ago!? Six years on... Intel is still the dominant CPU maker. Guess what? In six more years Intel will still be the dominant CPU maker.
N4CR said...
"I assume I'm correct in stating it is single core now? "
Sorry man but its 2 sockets, here is a picture...
And here is the article.
Post a Comment
<< Home