Thursday, November 30, 2006

K8L demoed

See video, there are some performance comparions inside.

Intel BK by 2Q08.

51 Comments:

Anonymous Anonymous said...

There's actual silion that was recently released. Called Quad FX. Tell us about that:)

11:57 AM, November 30, 2006  
Anonymous Anonymous said...

Where are those 200%+ performance increases compared to C2?
Alsop why did they say "simulations"? Don't they have real HW to benchmark on?

12:43 PM, November 30, 2006  
Anonymous Anonymous said...

Couple of interesting things:

Going from 2 core/old architecture to 4 core/new architecture gives only 70% Int performance (simulated) and 40% FP performance (simulated)?

Shouldn't dual core to quad core alone (throwing out architecture) changes yield a more substantial improvement? What happened to AMD's core scalability?

Are these parts underclocked to meet thermal/power requirements?

Also noone seems to take notice the mid-2007 is SERVER only. I wouldn't bet on seeing a desktop chip until 2007 best case. Looks like upper end AMD desktop is done until late 2007 (best case)

1:05 PM, November 30, 2006  
Anonymous Anonymous said...

Sniff....sniff...I smelll Vapor...

2:21 PM, November 30, 2006  
Anonymous Anonymous said...

Sharikou, on Nov 4th, you wrote the following:

"AMD 4x4 ready to frag Kentsfield with scalable architecture

Four 3GHZ cores+ four GPUs, AMD 4x4 will frag Kentsfield, which is just like two CPUs glued onto a slower bus. With Kentsfield, 4 cores share a 1033MHZ bus, each getting 266MHZ. We know this story ends in tears from 4P Xeon performance.

That's not all. By 2Q07, 4x4 owners will get an upgrade with the Rev H (aka, K8L) true quad. Then you have 8 Rev H cores, thermal managed independently. It will take Intel another 5 years to catch up, and AMD is a moving target.

Intel managed to squeeze the last bit of performance out of Bob Colwell's Pentium 3 microarchitecture, they simply ran out of brain juice in this game of innovation at the platform level. As I pointed out long time ago, communications is the key to high performance modern computing. AMD's core-core, core-memory, core-i/o, proc-proc communications are all Direct. Intel has only managed to establish core-core communication using shared cache at dual core level, it's about 5 years behind AMD.

...

Of course, there is another possibility with Torrenza, instead of running OCR in software, we can have a Torrenza OCR chip, which does the most time consuming algorithms in hardware..."


I am a multiple AMD PC owner since 1998 (K6-2 days), so don't consider me an Intel fanboy -- AMD's done the job since, and I still believe in them to this day. Intel's fine too, these are the PCs here at work.

At any rate, your journal entry above was/is quite off, really. Putting it lightly, you make all these extreme "frag" statements without really knowing the real conclusion.

4x4 performance is OK, but the drawbacks such as high noise & power consumption are certainly disappointing, to say the least.

I believe "4x4+" will change things, but that's a while away:

http://www.theinquirer.net/default.aspx?article=36017

Assuming that the inq is correct, 4x4+ will really the only way to get REAL blazing performance of the 8 core future setup w/ HT 3.0. I guess I will hold out until then.

2:29 PM, November 30, 2006  
Anonymous Anonymous said...

sorry sharikou,

This is a hearbreaker!

http://theinquirer.net/default.aspx?article=36050

2:30 PM, November 30, 2006  
Anonymous roborat said...

rofl, graphs shows 40% to 70% performance increase from current AMD offerings.
Intel is currently doing that to AMD NOW!
oh..it just keeps getting worst...

2:32 PM, November 30, 2006  
Anonymous Anonymous said...

Don't they have real HW to benchmark on?
Obviously not.

2:36 PM, November 30, 2006  
Anonymous Anonymous said...

where the hell is the demo? This is just an AMD guy who is promising stuff.. Nothing more dude..

At least Intel showed actual demos for Conroe and woodcrest. Sorry man, you are just out of argument and out of luck..

BTW, Kentsfield mops up the floor with the 4x4.. Did you see the benchmarks referred to by your buddies a the INQ

2:46 PM, November 30, 2006  
Anonymous Anonymous said...

Sharikou, you're definitely welcome to write anything and everything - but still, given that [disappointing] 4x4 benchmarks/reviews were out today, and instead you link the K8L demo, your AMD-biased side becomes highly apparent ... in an attempt to stay hush-hush about 4x4. You may not know it, but you're all about marketing, too.

The problem with your "64-bit pervasive computing" blog is that you will never accept/acknowledge the other side of the coin, and that is Intel. Again, by all means, write what you want, but if you expect to be respected, acknowledge the two horse race, and acknowledge that AMD can't win 100% across the board all the time. It's the nature of the industry - both Intel and AMD have talent, and they will swap #1 positions for the years to come. Engineering is never one-sided; competing companies will always find a way to innovate.

Me, I'm mainly AMD, but the work PCs are Intel - both do the job, and for the most part, I'm just interested on the software side of things; I'm a mechanical engineer.

In addition, MCM package or not, the average consumer doesn't really care. If I talk about "multi-core on a single die"
to non-geek relatives and friends, they really go to sleep and just say "uh-huh, yeahhh". Point being, the masses won't care what's inside the box, as long as it's quiet, fast and reasonably priced - and this time around, Intel is where it's at.

You can try to argue that 4x4 is the way to go for 8 core future proofing, but what happens when 4x4+ (w/ hyper tranport 3.0) comes around? Then you have a ~$300 board that won't fully utilize the upcoming native quads ... in addition to having "obsolete" FX70s by then. With the current version of 4x4, lose-lose situation for future proofing, if you ask me.

Finally, if indeed you feel like adding that phd credential all the time, do us all a favor and show us what happens inside the "black box" of these CPUs - whether it's calculus, differential equations, systems science and math, and all that other "higher education" jazz. Show us what a phd is all about. Because really, most of your blog content can be written with a bachelor's.

4:40 PM, November 30, 2006  
Anonymous Anonymous said...

IT journalist? My ass. THis blog hasn't shown one shred of journalistic integrity so how can I take the claim that you're an IT Journalist seriously. Further evidence ... Sharikou links to AMD web pages to back up his arguments. Not exactly an unbiased source.

6:20 PM, November 30, 2006  
Anonymous Anonymous said...

Uhhhgggg. Thanks for posting that SHarikou. I smell vapor launch. I think it might be time to sell my 500 AMD shares and pick up some INTC. This is NOT good news. AMD needs K8L NOW!!!

6:21 PM, November 30, 2006  
Anonymous Anonymous said...

The 4x4 setup is very nice for megatasking, but oh god the power consumption. They need to cut about 200 watts off of that.

Onto the Bankruptcy. Is that during the 2nd quarter or the end?

6:27 PM, November 30, 2006  
Anonymous Anonymous said...

Demo????

Demo???

It's a friggin' animation.

All we've seen so far is some animations and a wafer. WTF?

4x4 is a dog according to the latest benchmarks. It gets it's but whooped by a Core Duo let alone a QX chip.

And the power? Thru the roof.

KL8 (K-Late!) will be hit by the train that is the Xeon MP quad-core (Tigerton) that has already been demoed by Intel. Individual links to memory controller(s) should take care of that bottleneck.

Oh, and if you need memory that's attached to another processor? Oh, wait that's AMD's design where you need to go to another processor to get memory.

Quad Quad Cores. Now that's a real 4x4!

6:47 PM, November 30, 2006  
Anonymous Anonymous said...

Anonymous said...

Sniff....sniff...I smelll Vapor...


This is worse that Vaporware. This is FARTWARE. It stinks....

6:48 PM, November 30, 2006  
Anonymous Anonymous said...

http://enthusiast.hardocp.com/article.html?art=MTIzMywxMiwsaGVudGh1c2lhc3Q=

Real World Quad FX Speed & Agility

Our real world testing today disproves my preconceptions entirely, and shows that in quite a few cases, the FX-74 is as fast, or even faster than the QX6700. We ran the numbers over and over again, and the FX-74 simply has the horsepower to compete with the QX6700 in the most CPU intensive applications. When the systems are completely maxed out and you’re running rendering, encoding, or multi-media applications, the difference between the two CPU’s is quite minimal, usually between 1 and 3 percent.

Sorry to burst your bubble but the benchmarks don't lie. QFX def kicks serious ass and holds its own agenst Intels best. And costs less then QX6700 right now by as much as $500. Newegg offers a QX6700 at $1500. The FX74's are only $999 a pair, besides why not get some FX70's for $599 and OC them to 3ghz eh?

http://www.amd.com/us-en/Corporate/VirtualPressRoom/0,,51_104_609,00.html?redir=CPPR01
http://www.amd.com/us-en/Corporate/VirtualPressRoom/0,,51_104_543~114483,00.html

7:45 PM, November 30, 2006  
Anonymous Anonymous said...

Anónimo dijo...

where the hell is the demo? This is just an AMD guy who is promising stuff.. Nothing more dude..

At least Intel showed actual demos for Conroe and woodcrest. Sorry man, you are just out of argument and out of luck..

BTW, Kentsfield mops up the floor with the 4x4.. Did you see the benchmarks referred to by your buddies a the INQ

2:46 PM, November 30, 2006

werent intel promising the same and you all called "REAL" to it?
so now AMD spits it, and now AMD is
demoing, its VAPORWARE NOW?

chezz..

9:05 PM, November 30, 2006  
Blogger Jori said...

Roborat, its nowhere near those numbers. Its maybe 20% maxed in some benchmarks, C2D wins, C2D is superior, but C2D is not the god chip, and doesnt do 50% more than a K8.

9:26 PM, November 30, 2006  
Anonymous Anonymous said...

The 4X4 frags the intel chip by almost 2x. 510W vs 240W. Intel will surely go By 2008.

9:29 PM, November 30, 2006  
Anonymous SaintGreg said...

sorry sharikou,

This is a hearbreaker!

http://theinquirer.net/default.aspx?article=36050



How is that a heartbreaker? If anything that is good news for AMD. To quote from the article: "Test scores reveal that Quad FX manages to achieve equal performance to Core 2 Extreme QX6700 in SiSoft Sandra 2007 and PCMark05"

AMD dual core gets beat by Intel's dual core by 10% - 30% in those same tests according to this link. The fact that when you stack two of those same chips together you see the 2 AMD chips performing significantly better proportionally to the single quad core Intel is very good for AMD. The price point is also comparable: 1k for 2 top end AMD chips, the same as for the 1 top end Intel chip. That means a very competitive option from both companies.

But that also means that when K8L comes with the new architecture and 2xquad core (giving its own boost of 40-70% per core) AMD will flat out trounce Intel in performance by a very clean and decisive margin. We know whats in the the works for AMD, does Intel have anything in the works to deal with it? I gut feeling is no - if they try to make their own dual CPU device like 4x4 the FSB just won't hack it. We know another new architecture (possibly involving their high speed interconnect) is in the works but that isn't supposed to be until 2008 at the earliest.

where the hell is the demo? This is just an AMD guy who is promising stuff.. Nothing more dude..


Not only that, but the sad part is that he also just says the same stuff we already knew. Waste of 3:00 if I've ever seen one...

11:05 PM, November 30, 2006  
Anonymous Anonymous said...

Techreport.com has a good review of QuadFX which shows it can be competitive depending on application used. However the main problem is the high power consumption, noise and overall price to get equal to Kentsfield in just some applications.

Xbitlabs also show an interesting graph where finally Kentsfield runs out of steam when running 5 applications, however enthusiasts tend not to run that many, they run one or two more quickly !

So in summary AMD will still be strong in the server market but for the high end desktop Intel is still gaining the benefit of it's new architecture and AMD will not until it arrives mid 2007. I wish it was sooner.

Bandwidth, 65nm and QuadFX are now shown to be just straws to grab hold of until the real deal turns up in K8L.

11:35 PM, November 30, 2006  
Anonymous Anonymous said...

"Intel BK by 2Q08."

You had said 7 months 7 months ago.

Now you are again changing your prediction?

Now what will you say in 2Q08? 4Q 3010?

1:38 AM, December 01, 2006  
Blogger S said...

With 4x4, I think for the first time in a good few years, AMD has its backers doubting if they are backing the right horse.

AMD is back to playing catch up with Intel.

3:20 AM, December 01, 2006  
Anonymous Anonymous said...

Bachelors?

I doubt Sharikou has ANY degree, let alone a PHD or bachelors.

However, posessing a degree is hardly any indication of intelligence. Most of the innovators in the world weren't bound by institutional thinking, which made them the sucesses they are. (Granted, many HAVE also, in spite of their academic background).

---

So we get back to the K8L - what a disappointment. AMD just can't work their mojo these days. They're investing too much into the K8L when, it's generally agreed, that K8L won't appeal to the consumer (unless it's price is so low that it can be afforded by most common people).

I wholeheartedly wanted K8L to blow Kentsfield out of the water, but it doesn't. All that R&D invested in making multicore processors better hasn't paid off. This just isn't AMD's year.

7:04 AM, December 01, 2006  
Anonymous Anonymous said...

Most of you don't getQuadFX.It lost most of the simplistic benchmarks but it got decent reviews because it is the only enthusiast platform that:
-upgrades to 8 cores
-4 pcix graphic card slots with appropriate thermal spacing for SLI
-12 internal SATA ports

On the thermal side, yes, it requires an extra 125 watts, but given the baddest ass GPU cards can suck 250 watts and over 350 watts for a quad config, who cares? Maybe those who drive a Prius, I guess.

9:08 AM, December 01, 2006  
Blogger Dr. Yield, PhD, MBA said...

To say this "demo" is disappointing is an understatement. The "demo" was a powerpoint marketing presentation, with nothing to back it up.

The key takeaways:
+ AMD does not have K8L/Barcelona silicon working well enough that they are comfortable showing it or generating performance data with it.

+ "Simulated" performance improvements of 70% integer and 40% FP not earth-shattering or category re-defining. They also show a shocking lack of scalability- so much so that I'm rather puzzled.

My conclusions:
+ K8L/Barcelona is best case just going to meet its planned ship date given that server parts typically require at least 6 months of Si validation. This of course assumes that they get the bulk of their issues resolved in the next mask revision.

+ K8L/Barcelona will be at best performance competitive with Cloverton in the 1P/2P space. 4P will still leverage the advantage of AMD's memory architecture. Of course, need for 4P goes down with 4 cores/socket...

+AMD has effectively ceded leadership of the desktop space for at least 1 year. C2D/C2Q are available in volume now, and have the performance crown for desktop apps. AMD's only answer until K8L/Barcelona ships into the desktop (Q407?) was QuadFX, which has now proven itself to be a slower, hotter alternative. Exactly what AMD whomped Intel over the head for for several years. Not good when the shoe is on the other foot.

+AMD needs to grow its mobile business and hang onto high-end servers, because their desktop business has been commodotized and will not generate margins. Look at Dell advertising- AMD is primarily the "value" option in desktops. Same thing is happening in the 1/2P server space due to loss of performance and performance/watt leadership.

And you wondered why I said that I was concerned that the huge ATI deal was going to distract management from the core business? Eye off the ball, and now look. Hector needs to regain focus and fast. AMD doesn't have the cash cushion Intel did to allow it to make a few blunders.

9:48 AM, December 01, 2006  
Blogger Dr. Yield, PhD, MBA said...

http://eetimes.com/news/semi/showArticle.jhtml;jsessionid=AYP1YFHFOMHE2QSNDLPSKHSCJUNN2JVN?articleID=196600862

Just to further my point about ATI acquisition being a distraction for management. See the above reference eetimes.com article- AMD seems to have acquired a DOJ monopoly investigation, as ATI is being investigated for its practices in the graphics market (along with NVIDIA). Money that would be better spent on processor development, process technology, and capacity expansion.

9:56 AM, December 01, 2006  
Anonymous edward said...

"Money that would be better spent on processor development, process technology, and capacity expansion."

You should say this to Intel, whose legal & marketing money percentage is way higher than AMD.

11:21 AM, December 01, 2006  
Anonymous edward said...

"+ AMD does not have K8L/Barcelona silicon working well enough that they are comfortable showing it or generating performance data with it."

AMD actually demoed the box to a bunch of Industry Analysts. Unfortuantely and apparently you aren't one of them.

You call 70% performance improvement lack of scalability? Do you know what benchmarks they used to make such comment? Do you know anything about integer performance at all? I'm sure your doctorship, if any, has nothing to do with computer science (or you're just humiliating your school & advisor).

"+AMD has effectively ceded leadership of the desktop space for at least 1 year."

Oh I don't know C2D has been out for at least 1 year... ??Which planet did you exactly inhabit? Doesn't seem like earth to me....

"+AMD needs to grow its mobile business and hang onto high-end servers, because their desktop business has been commodotized and will not generate margins."

AMD's desktop has always been commoditized and generate much fewer margin than Intel. They only did better in 2005, 2.5 years into K8, after they filed the lawsuit against Intel, something you deemed a waste. Apparently you have very wrong/contradicted ideas about business, too.

Why do you spend make such a long bullshit analysis to embarass your own phd-ship? Your higher education looks little to me just because of this. All those critics against Sharikou suddenly seem proper on you...

11:34 AM, December 01, 2006  
Anonymous edward said...

"They're investing too much into the K8L when, it's generally agreed, that K8L won't appeal to the consumer (unless it's price is so low that it can be afforded by most common people)."

When does AMD market K8L to "consumers"? It's always been marketed as a server chip. Now, how good a consumer process it will make, when AMD cuts it down to dual-core, is another question.

11:37 AM, December 01, 2006  
Anonymous Anonymous said...

Got this link from AMDZone...

AMD Demos Native Quad-Core Processor

It amazes me to see so much ignorance coming out here from Intel fanbois minds. Vapor launch?

11:49 AM, December 01, 2006  
Anonymous Anonymous said...

"Most of you don't getQuadFX.It lost most of the simplistic benchmarks but it got decent reviews because it is the only enthusiast platform that:
-upgrades to 8 cores
-4 pcix graphic card slots with appropriate thermal spacing for SLI
-12 internal SATA ports "

If you are enthusiast:
- You do not need 12 SATA ports (you would build a NAS box to have that much HD capacity). Unless of course you need >6Terabytes instantly available?
- You will not upgrade to 8 cores while losing some of the capability on those future quad core chips... it is pin compatible but the MOBO/chipset will not be fully compatible with all of the chip features in the future - AMD themselves have stated that.

Enjoy your 4x4 by all means, but it is not "future prrof" and not any better/cheaper than an Intel Kenstfield system when you factor in cooling needs, bigger power supply, bugger power bill.

And if you do like to OC, then the Kenstfield is a much better chip.

12:43 PM, December 01, 2006  
Anonymous Anonymous said...

I found a good use for this 4x4: use it for winter.. It can also be used to cook!

3:55 PM, December 01, 2006  
Anonymous edward said...

"If you are enthusiast:
- You do not need 12 SATA ports (you would build a NAS box to have that much HD capacity). Unless of course you need >6Terabytes instantly available?
"

He WILL need this many SATA ports if he is a megatasking user. You haven't heard of RAID, have you?

"- it is pin compatible but the MOBO/chipset will not be fully compatible with all of the chip features in the future - AMD themselves have stated that."

AMD stated nothing like that, which is another FUD from Intel believers. The mobo and chipset are fully compatible. You don't have the ability to use DDR3, that's it.

But honestly, say he invested on 4GB DDR2 memory today; will he trash them for DDR3 just because quadcores are available?

I see you have a very silly type of "enthusiast" in mind.

5:59 PM, December 01, 2006  
Blogger Dr. Yield, PhD, MBA said...

You should say this to Intel, whose legal & marketing money percentage is way higher than AMD.

Last I checked, Intel is not resource constrained for manufacturing. AMD is- so they need to be a little more careful how they spend those dollars. And generally, I think you would be hard pressed to show that Intel's marketing expenditures have been anything but effectively spent. Remember all of those hot Pentium 4's everyone bought even though K8 was the superior offering at the time?

It's like the guy with a take home income of $100k vs. the one with a take home of $25k. On a percentage basis, the $100k guy can afford to spend a higher percentage of his income on discretionary expenses. Or do you disagree?

7:03 PM, December 01, 2006  
Blogger Dr. Yield, PhD, MBA said...

edward whined and cried:
AMD actually demoed the box to a bunch of Industry Analysts. Unfortuantely and apparently you aren't one of them.

Oh yeah, I read the reports of the demo too. It boots Windows and runs 16 instances of task manager on a 4 way. If it ran more, I think they would be demoing apps and reporting something other than simulated performance.

You call 70% performance improvement lack of scalability? Do you know what benchmarks they used to make such comment? Do you know anything about integer performance at all?

No, no, and oh you betcha. Comapred to scaling of X2 to 2P X2 servers, it's not as impressive as I expected. But you're right- they could be underhyping performance.

I'm sure your doctorship, if any, has nothing to do with computer science (or you're just humiliating your school & advisor).

I'll ignore the personal jabs- it's the sign you don't have much useful to input. Computer science no- but as you've demonstrated previously here, I've forgotten more about process and device technology than you've demonstrated proficiency in...

"+AMD has effectively ceded leadership of the desktop space for at least 1 year."

Oh I don't know C2D has been out for at least 1 year... ??Which planet did you exactly inhabit? Doesn't seem like earth to me....

Comment is based on AMD's desktop roadmap- but obviously you were foaming too much at the mouth to assume anything other than the worst intention. They are currently not reporting anything that will challenge Intel's high-end processors now. Could they steer Barcelona to desktop before this time next year? Could 65nm give them speed bumps that they aren't publishing as planned releases now? Sure! Hopefully! I'd like to see AMD competitive. But nothing on the desktop roadmap for the next 12 months indicates thats the case. For AMD's sake, I hope they have another ace up their sleeve.

AMD's desktop has always been commoditized and generate much fewer margin than Intel. They only did better in 2005, 2.5 years into K8, after they filed the lawsuit against Intel, something you deemed a waste. Apparently you have very wrong/contradicted ideas about business, too.

I never said that suit was a waste. What I said was the ATI acquisition has diluted management focus, and brought additional expenses that drain resources away from critical needs.

Why do you spend make such a long bullshit analysis to embarass your own phd-ship? Your higher education looks little to me just because of this. All those critics against Sharikou suddenly seem proper on you...

Why are you so whiny and bitter? I've pointed my concerns about AMD not executing. I've previously stated that I'd like to see two successful competitors in the space. You seem incompable of analyzing any statements that appear to "hurt" your favorite company. Intel made bad choices- they got called for it, and they are suffering for it. Shoe is on the other foot now.

7:19 PM, December 01, 2006  
Anonymous Anonymous said...

Yo PhD pretender.. you have been quiet for a while.

What no commentary about AMD's new product 4x4. Tell us about the power numbers, tell us about the performance of that great 4x4 competing with Kentfield please! Please use your higher intellect!

K8L demo'd how come it was demoed as a x16? how come not as a native 4 core ? I find that very very curious. How come now benchmarks. Pss.. I know why.. any guess from you AMD fanboys?

The PHD pretender please tell us! Why as a 16 way and not as a native 4 way and why no benchmarks please???

Lots of FUD no Meat!!

Where is the count down to our trip to BK for some burgers and fries?

8:09 PM, December 01, 2006  
Anonymous Anonymous said...

They demonstrated it.... running task manager! Nothing else! Wowee, isn't that awesome? Intel showed off Tigerton, which is also due next year, back in October. However, Intel actually Showed Tigerton running some stuff, not just task manager: http://www.techreport.com/onearticle.x/11099

This would suggest to me that K8L is not as ready as AMD hoped for. If it were capable of running some applications to show how well it runs... then why wouldn't they have done so?

8:12 PM, December 01, 2006  
Anonymous Anonymous said...

"You call 70% performance improvement lack of scalability? Do you know what benchmarks they used to make such comment? Do you know anything about integer performance at all?"

Edward - you do realize that all of this was SIMULATED and there was no actual performance demonstration other than showing task manager with 16 CPU util shots, right? They didn't even show CPU speed!

And no I'm not impressed 70% scaling (and only 40% in FP) with 2 core --> 4 core AND old architecture to new architecture both happening. Haven't folks on this board said they expect >80% scaling going from 1 socket -> 2 socket (or 2 to 4).

I find it interesting that within the processor going 2 core to 4 core (AND again "revolutionary" new architecture) only nets 70 (or 40% FP)%...

If this data holds are you better off going dual core 2P to dual core 4P server instead of going dual to quad core within a 2P server? And then just eating it on the power consumption and extra memory, just like the 4x4 solution?

I also find it interesting that certain people screamed BS! when Intel did their own demo (with actual applications running) of Core 2 and now those folks are more then happy to believe SIMULATED numbers with no actual demonstration behind it (biased or unbiased)...As said with previous product demos - we may actually want to wait for some real demos.

12:50 PM, December 02, 2006  
Anonymous The Architect of Taos said...

THE REAL STORY ABOUT KL8

Due to the fact that the Opteron team left AMD, the KL8 has gone through several redesigns.

AMD is struggling to make a chip that works as well as the Opteron, whose design is now 10 years old.

Rumor has it that KL8 will not work in existing motherboards due to incompatibilities in how HT is implemented. But without the newer HT in KL8, the quad-core processor does not work well with another quad-core.

Meanwhile, Intel is shipping quad-core Xeons that work just fine and have been shown to work in existing systems for the past 6 months.

Maybe the FSB is old and ancient. But it is reliable and mature, something that cannot be said about AMD's HyperTransport.

Overall, AMD is now 1 year behind Intel in quad-core and falling further behind every day. Once Intel gets their Churchill embedded microcore designs going, then AMD will be in big trouble. Imagine 4 main cores surrounded by 32 microcores. This microcore architecture is going to make mincemeat of AMD.

12:51 PM, December 02, 2006  
Anonymous Anonymous said...

"They demonstrated it.... running task manager!"

Hey task manager is a heavy megatasking applictaion (doesn't it show all CPU's running?!?!) - much better than all those "fake" benchmarks people use. They really should put 4x4 on it and see how it does as well!

I prefer my data SIMULATED! (much easier for me to remain in my state of denial) And while I'm in that state; I'm also going to assume that Intel will make absolutely 0 improvements to the Core2 architecture - it's clear from the OC'ing reports that the architecture has 0 headroom for incremental speed improvements. In fact as the 65nm process further matures and 45nm starts coming online the speed may actually get worse as we all know Intel's manufacturing capabilities are complete crap! I mean it's not lilke they're already sampling 45nm chips as AMD's 65nm is just starting to ship.

And now it's time for me to get back to planet earth....

7:46 PM, December 02, 2006  
Anonymous Anonymous said...

There's just so much BS here.

Here's the real deal with the new sockets/processors.

A new socket was originally planned to cater for HT3 (will support PCI-E v2 properly) and DDR3. That was AM3.

DDR3 official specs have been delayed, so there is an interim solution supporting PCI-E v2 with HT3 and running DDR2.

With this hybrid socket design, you can use new quadcores on current Socket AM2 and Socket 1207 with PCI-E v1, maintaining your current equipment, and having investment protection. The new QuadCore's HT3 will run in HT1 mode.

This is like when you bought an AGP8x graphics card and slotted it into an AGP4x slot. Was there a compromise? Yes vs running in full AGP8x mode, but it was a comfortable compromise without losing too much performance.

Or you can get the new quadcores into newer Socket AM2+/1207+ socket motherboards which will support HT3 and PCI-E v2.

This is like if you changed your motherboard and graphics card at the same time to support AGP8x.

What's the Intel approach? The same. You also need to get a new motherboard for PCI-E v2.

So why all this outcry?

8:59 PM, December 02, 2006  
Anonymous edward said...

"Oh yeah, I read the reports of the demo too."

No, you didn't read the report, or you wouldn't have said the demo is "powerpoint marketing presentation."

You were simply wrong. You just don't have the gut to admit it.

"Comapred to scaling of X2 to 2P X2 servers, it's not as impressive as I expected. But you're right- they could be underhyping performance."

So you really think 70% int performance improvement is underhyping? Then you really know nothing about the current art of computer science and architecture.

"I'll ignore the personal jabs- it's the sign you don't have much useful to input. Computer science no- but as you've demonstrated previously here, I've forgotten more about process and device technology than you've demonstrated proficiency in..."

Apparently you can't ignore it - then don't pretend you will. And, I don't know/care how much you've forgotten on any topic else; just wanna tell you that your "analyses" on 4x4 were wrong and rediculous. That's the point here.

10:51 PM, December 02, 2006  
Anonymous Graham said...

edward said:
You should say this to Intel, whose legal & marketing money percentage is way higher than AMD.

THe major difference being that Intel HAS the money to spend on legal and marketing and still develop some kick ass processors. AMD on the other hand can do only one thing at a time because of lack of funds and low profitability (on the razors edge of losses).

8:37 AM, December 03, 2006  
Anonymous Graham said...

edward wrote incoherently:
Why do you spend make such a long bullshit analysis to embarass your own phd-ship? Your higher education looks little to me just because of this. All those critics against Sharikou suddenly seem proper on you...

I think you might want to take a weekend and bone-up on English. You grammar and spelling are atrocious.

8:41 AM, December 03, 2006  
Anonymous edward said...

"just wanna tell you that your "analyses" on 4x4 were wrong and rediculous. That's the point here."

Oops... I meant to say K8L...

9:20 AM, December 03, 2006  
Anonymous enumae said...

AMD's quad core should see performance increases of about 70% INT and 40% FP over current 2220SE as published in Spec.org (zoom in on the presentation).

I found the best SPECint_rate2000 score, then used the same system for SPECfp_rate2000.

Before you say anything about the selections keep in mind AMD is far behind Intel on integer performance and that is why I focus on integer, floating point is in AMD's hands already, they will only stretch the lead there, and as I am sure the Dr will point out, the Sun Microsystems is faster but is running Solaris 10 (and does not show a SPECint_rate2000 score), not Windows Server 32bit.

Hewlett-Packard Company ProLiant DL385 G2 (AMD Opteron (TM) 2220SE)

Hewlett-Packard Company ProLiant DL385 G2 (AMD Opteron (TM) 2220SE)

------------------------------
SPECint_rate2000
------------------------------

SPECint_rate2000 = 90.4

90.4 * 1.70 = 153.68 is the projected score for SPECint_rate2000 in 2P with 8 cores.

Now compare this to...

IBM System X 3500 (2.67 GHz Xeon X5355, 8MB L2 Cache)

SPECint_rate2000 = 197

153.68 * 100 / 197

AMD has about 78% of the performance of Intels 2P quad core in SPECint_rate2000.

------------------------------
SPECfp_rate2000
------------------------------

SPECfp_rate2000 = 80.1

80.1 * 1.4 = 112.14 is the projected score for SPECfp_rate2000 in 2P with 8 cores.

Now compare this to...

IBM System X 3500 (2.67 GHz Xeon X5355, 8MB L2 Cache)

SPECint_rate2000 = 101

101 * 100 / 112.14

Intel has about 90% of the performance of AMD's 2P quad core in SPECfp_rate2000.

What this tells me is it doesn't matter who you are for, Intel or AMD, they will both have advantages over the other, but as it looks now the performance of AMD's quad core is a little dissapointing.

The main thing to keep in mind is that AMD's presentation is talking about 2P, not 4P or above, also we do not have any idea about clockspeed, these may be quite low as to be able to fit in the same power envelope, but at the same time the 2220SE 2.8GHz part has a TDP of 119W.

AMD will likely still have the scalabilty advantage over Tigerton (4P and up), but in 2P it will be going against the 45nm shrink of Woodcrest a few months after launch (supposedly). As many of you have read the clockspeed increases for 45nm will be significant, and most likely give Intel the ability to surpass AMD's future advantage in floating point.

As it stands it would seem AMD is in trouble, and that sucks for consumers, Intel will be able to raise prices and if you want the best performance you are going to have to pay for it.

10:15 AM, December 03, 2006  
Anonymous Anonymous said...

Question for Edward:

If AMD solely went from 2 cores to 4 cores on the K8 architecture, what performance would you expect from just this change?

Now throw in the architecture change, and yes a 40% FP is nice bu tthis underwhelming considering both the increased core count and the hype on K8L.

I'm not saying the quad core K8l will not be a good product, but between the doubling of the # of cores and the architecture change I was expecting at least 2X (100% better than)the old generation (and 1/2 the core count)

If we were talking about a dual core K8l, than yes these #'s would be VERY good.

6:21 PM, December 03, 2006  
Blogger Dr. Yield, PhD, MBA said...

edward mumbled:
So you really think 70% int performance improvement is underhyping? Then you really know nothing about the current art of computer science and architecture.

Oh great ridiculor of my knowledge, I humbly submit this question to your superior understanding of computer science:

If going from 1P/2C to 2P/4C with the existing architecture gives me a 91-99% improvement in peak SPEC2000 INT_rate, and a 97-99% improvement in SPEC2000 FP_rate, why would I NOT be disappointed in only 70% improvement in integer and 40% improvement in FP performance?

Data was collected from SPEC200 Q$2006 results, using 2 Fujitsu-Siemens syste,s (Opt285 and 290) and an HP Proliant System (Opt2218).

I stand by my statement- due to the less efficient scaling of 4 integrated cores on a single socket, a 2P/4C configuration can be expected to perform better than a 1P/4C system. NOT an expected result with the architectural enhancements included in K8L. Which leaves me disappointed.

I'm happy if your happy with it- but, frankly, I had hoped for better. And I sincerely hope that maybe a few more Si steppings might wring out some more performance. And I also stand by my statement having been involved with plenty of demo Si back in the day- if they could have demo'd more software running on the tool- they would have. I feel bad for the debug engineers- no holiday season for them this year. They are going to be busy.

9:46 AM, December 04, 2006  
Blogger Dr. Yield, PhD, MBA said...

edward, one final note regarding scaling:

I think we can all agree that generally speaking, the IMC of Opteron is superior to Intel's FSB. Yet Clovertown spec performance, which has no architectural improvements over Woodcrest shows integer scaling of 83-85%, and FP scaling of 35-38% over the 2C/1P solution. So with an inferior memory access architecture and no enhancements, it achieves better integer and nearly equivalent FP scalability as K8L/K8. So again, I'm disappointed.

Regarding If going from 1P/2C to 2P/4C with the existing architecture gives me a 91-99% improvement in peak SPEC2000 INT_rate, and a 97-99% improvement in SPEC2000 FP_rate, why would I NOT be disappointed in only 70% improvement in integer and 40% improvement in FP performance? , we can perform the same test for Intel. The 1P/2C to 2P/4C speedup for them are ~83% and 70% (INT/FP), which is worse than Opteron. But you'll note that the 4P/1C solution scales as well in integer, falling short in FP. Performance disparity between the two isn't as large (and again, no archtectural improvements between the two to help cover the FSB issue).

Please, use simple words so that I can understand why I shouldn't be disappointed. I really hoped for more, and continue to cross my fingers that they can squeeze more out of this.

11:04 AM, December 05, 2006  
Anonymous Anonymous said...

"This is like if you changed your motherboard and graphics card at the same time to support AGP8x."

I think the "outcry" (if you can call it that) - is AMD fans are/were purporting to buy this future-proof / upgradeable 4x4 system and just drop in two quads for 8x4. Not too long ago there was this theory of buying chips one at a time, mixing processor speeds until the 1207 news came out.

If you are truly planning to go 8x4, why would you not wait until that launch and buy a socket which took advantage of the new chips' capabilities?

I think the point is 4x4 is a temporary band-aid and folks are better off waiting until the inevitable bugs get ironed out and get a board that takes advantage of all of the new chip's capabilities - after all this is targeted at "megataskers" anyway and aren't those the folks that want to take advantage of all the new features?

11:24 PM, December 06, 2006  

Post a Comment

Links to this post:

Create a Link

<< Home