Saturday, December 16, 2006

Anand made another negative scaling discovery

Recall Anand's last great discovery in computer science? Negative scaling in server performance--adding an Opteron CPU leads to lower performance.

Now, he did it again. This time, negative frequency scaling, a 2.6GHZ Athlon slower than a 2.4GHZ one.

58 Comments:

Blogger Roborat, Ph. D. said...

AMD's supposed IMC and its scaling is a myth. Anand has already proven this with data. You on the other hand has shown nothing.
I wouldn't be suprised if the new 65nm processors from AMD is broken considering its coming from their obviously broken SOI process.
Guess we finally know who right about staying away from SOI. No, its not AMD -- if you don't have the brains to figure that out.

9:43 AM, December 16, 2006  
Anonymous Tritosh said...

taken from the Anandtech:

What we're looking at here is the core temperature of the 2nd core in all of the CPUs, under full load, as reported by Core Temp. While it's not necessarily useful (or accurate) to compare readings across two different motherboards, as is the case when looking at AMD vs. Intel, the comparisons between AMD chips alone are enough to showcase the reduction in temperature.

With both cores under load for 15 minutes (calculating Fast Fourier Transforms) the 65nm 5000+ manages to produce just about as much heat as the X2 3800+ EE SFF. While this won't always be the case, it gives you an idea of the reduction in temperatures you can expect from AMD's new 65nm chips.

What about overclockability? We were unfortunately not able to get that much more out of the new 65nm core as we could from mature 90nm chips. Our X2 5000+ was able to run at 2.925GHz, at 1.475V with stock air cooling. If equipped with better air cooling or something more exotic, reaching over 3GHz shouldn't be a problem, but we wouldn't expect to see anything too far over 3GHz.



not only that, they're triying to find anyway to make AMD look bad, rofl..
like AMD beats intel in temp here..and now they claim the inacuracies, and that its not always the same, and you might not get the temperature drop of their version..

LMAO, I havent looled that hard at how biased anandtech is

11:27 AM, December 16, 2006  
Anonymous Anonymous said...

So, from the bench mark test, it seems that C2D still mops the floor with the 65nano from AMD.. That is the real story!

12:10 PM, December 16, 2006  
Anonymous Anonymous said...

Hey, didn't you say some time ago that AMD improved its 65nm CPU FPU's quite a bit and K8L will be further improvement over them? I see no real improvement, just a dumb reduction as everyone else said.

1:47 PM, December 16, 2006  
Anonymous Anonymous said...

Anand is making a fool of themself! like similar biased website always ridiculing Amd vs Intel.

AMD is the Future.

6:29 PM, December 16, 2006  
Blogger Sharikou, Ph. D said...

So, from the bench mark test, it seems that C2D still mops the floor with the 65nano from AMD.. That is the real story!



Only an idiot would trust Anand's conclusions.

7:37 PM, December 16, 2006  
Blogger sharikouisallwaysright said...

By the end of 2007 AMD will wipe the floor with Intels outpaced and then obsolete C2D-Megacache-FSB-Crap.

8:04 PM, December 16, 2006  
Anonymous Anonymous said...

Hey dude how about defending your last pile of $hit, "Intel stuck with the hot and slow image" before dropping another load of $hit. You post crap and then run away like a coward.

8:43 PM, December 16, 2006  
Anonymous Anonymous said...

"Only an idiot would trust Anand's conclusions."

So are you saying the E6600 doesn't frag the X2 5000+?

9:24 PM, December 16, 2006  
Anonymous Anonymous said...

"Only an idiot would trust Anand's conclusions."

So, the intelligent guy/gal should not have trusted the Anand benchmarks which showed the supremacy of Athlons over Pentiums?

10:21 PM, December 16, 2006  
Anonymous Anonymous said...

"So, from the bench mark test, it seems that C2D still mops the floor with the 65nano from AMD.. That is the real story!



Only an idiot would trust Anand's conclusions."

And only a moron would believe anything you type. You are a pathetic imbecile. I salute you. One finger at a time. Cunt.

10:44 PM, December 16, 2006  
Anonymous Anonymous said...

If everyone is so biased, why don't you do your own benchmarking?

Any predictions for Q4 numbers?

12:00 AM, December 17, 2006  
Anonymous Anonymous said...

Anand is probably the last of the few saviors for Intel. I am sure Intel is making sure ($$$) that this last piece of oxygen support remains..!!

12:04 AM, December 17, 2006  
Anonymous edward said...

"I wouldn't be suprised if the new 65nm processors from AMD is broken considering its coming from their obviously broken SOI process."

Oh yes it's so broken that the AMD 65nm chips consume less power than Intel's and are going to be used by all those mission-critical servers and supercomputers. Yes IBM/AMD's SOI process must be very broken in the eyes of an Intel fan.

12:17 AM, December 17, 2006  
Anonymous Anonymous said...

Anandtech did place the series 8000 opterons in the top ten technologies of the year along with c2d.
Anand did reported that the 8000 series opterons simply outperform all the c2ds and they can scale very well.
The conroes(pentium 3s) simply cant compete with the hi tech 8000 series opterons.
The quad c2d is a joke since intels antique bus system splits the static bus speed by 4 leaving only antique bus speeds for each of the cores.
Anand reports that the 8000 series opterons simply trash the c2ds in performance in actual server workloads.
There are simply only AMD chips to run 8 core systems.
AMD and IBM have announced they are working on the 45nm, 32nm, and 22nm chips.
Intel has no plans for 32 or 22 nm chips and no money left to prototype them.
Intels 45nm chips have yet to be proven as workable but AMD 45nm is already tested and working.
Intel will fall out of the top 10 chip sellers sometime during the next 12 months.
The minute intel falls from the 1st place spot will mark the beginning of the end for intel.
AMD leaped to the number 7 spot with only 1 quarter of sales increases and there does not seem to be any stopping AMD/ATI now.
Intels has had a long run especially considering all there platform problems.
C2Ds are antique pentium 3s glued together and supercharged but they are not high tech enough to carry intel to future success.
Now intel tells us how there going to crush AMD/ATI, are they not 5 years late with this kind of talk????
The final quarter of 2006 is looking very good for AMD/ATI but may put intel into serious financial waters.

12:23 AM, December 17, 2006  
Blogger Tom said...

Anonymous fanboy said

AMD and IBM have announced they are working on the 45nm, 32nm, and 22nm chips.
Intel has no plans for 32 or 22 nm chips and no money left to prototype them.
Intels 45nm chips have yet to be proven as workable but AMD 45nm is already tested and working.


That is funny I was under the impression that Intel had already demonstrated 45 nm chips. They did so almost 1 year ago an January 25 2006 http://www.intel.com/pressroom/archive/releases/20060125comp.htm AMD did not do so until April 2006 but the first one to show the chips was in reality TSMC and they did so in 2004. As it is ovious that you don’t know what you are talking about I will just skip your comment about Intel not working on 32 nm. Forget that just scrool ti the buttom of this article where 32 and 22 nm are both mentioned http://news.com.com/Intel+shows+test+chips+made+on+future+processes/2100-1006_3-6030489.html

Intel will fall out of the top 10 chip sellers sometime during the next 12 months.
The minute intel falls from the 1st place spot will mark the beginning of the end for intel.


Did you actually look at the numbers they have today? Even if AMD were able to supply 50% of the world with chips (and they are not) Intel would still be very much within top 10 on the list.

AMD leaped to the number 7 spot with only 1 quarter of sales increases and there does not seem to be any stopping AMD/ATI now.

Please remember that of the 7.5 billion revenues 2.5 billion are from the ATI merger. AMD have no way to grow 91% again anytime soon as they don’t have the financial power to take over any further companies and for the next few quarters its outlook is quite bad with sinking ASP and still no new graphics card from ATI to help with sales.

C2Ds are antique pentium 3s glued together and supercharged but they are not high tech enough to carry intel to future success.

And still they outperforms any AMD solution on desktop and laptop markets. On the server side things are looking a bit different as servers are much more dependant on the link between memory and other subsystems than any desktop computer. So here AMD do have an advantage that Intel properly cant beat until 2008. But raising the FSB speed would help of curse so don’t expect to see Intel loose all their server business.

P.S. I run a website that I have hosted on a Quad opteron great system.

1:48 AM, December 17, 2006  
Anonymous Anonymous said...

"Only an idiot would trust Anand's conclusions."

Even when they're in line with all the other Core 2 benchmarks? Or are you seriously claiming that Intel has bribed every hardware website and printed magazine in the world and planted thousands of users to post in forums claiming how great their processors are?

2:56 AM, December 17, 2006  
Blogger Fujiyama said...

What a great company is Intel. You buy an upgradeable server and you cannot upgrade it...here.

4:47 AM, December 17, 2006  
Anonymous Anonymous said...

Trust but verify.
I have seen no one ever report that they could reliably repeat the conroe benchmarks as reported by Tons,TG,and Anand without the intel referance platform.
All the benchmarks were done on hardware from intel that is not obtainable by the fanboys or anyone else for that matter.
In my own tests with the e6400 and e6600 I have never been able to verify the benchmarks reported by the press.
I have verified the benchmarks for AMD AM2 5000+ even with the lowest price platorms.
The intel benchmark platforms are just more vapor hardware from intel for a select few only.
Intel has never been known as a reliable source of unbiased data.
After all, intel counts VIA cpu sales as there own, is that not a lie?????
A antique reworked pentium 3 beating a opteron, in what universe does this make sense?????
I have heard a rumor that the conroe is actually a reworked and rebranded opteron, thats probably as true as the c2d benchmarks.

8:19 AM, December 17, 2006  
Anonymous Anonymous said...

The test=linux linspire 5.0
platform e6600 at 4791 bogomips
platform AMD AM2 5000+ at 5207 bogomips
Running std. random number generation sequences with generation of 10001 samples. timed event

Results(faster is better)
It took the conroe e6600 2.17 sec to complete task.
It took the AM2 5000+ only 1.13 sec to complete the same task.

However, if this test was run by intel and anand I am sure the results would be different.
So AMD runs std. software at nearly twice the speed of c2d and for less money with less power.

10:02 AM, December 17, 2006  
Anonymous Anonymous said...

"What a great company is Intel. You buy an upgradeable server and you cannot upgrade it..."

Just some time ago Sharkou was talking about using it's AM2 3800+ on 4x4 ...

2:37 PM, December 17, 2006  
Anonymous Anonymous said...

"Oh yes it's so broken that the AMD 65nm chips consume less power than Intel's and are going to be used by all those mission-critical servers and supercomputers. Yes IBM/AMD's SOI process must be very broken in the eyes of an Intel fan."

You talking at Cool n quiet idle mode or full load? Idle on AMD runs at 1.0GHz, vs Intel's 1.6GHz?

All IBM (uh, I mean AMD) has done thus far on 65nm is drop the Vt which enables a lower Vcore, hence the power reduction. As expected we have not seen any speed improvements yet because the 65nm transistor (right now) is essentially the same as the 90nm version due to AMD's "advanced" CTI (continuous transistor improvement) philosophy. This means they will improve transistor speed over time. Thus the only early benefit is the lower Vt of the 65nm tech node (leading to lower power). Which is why you see all the high end chips still on 90nm (which has mature yields and presumably better bin splits)

Intel on the other hand puts their transistor benefits in early BEFORE they ramp.

Before you start flaming I'm not saying this is better, but if AMD did what Intel does they would be nowhere near ready to ramp 65nm yet - this essentially allows them to start the 65nm introduction earlier. I expect AMD's 65nm process will not be mature (in terms of shipping product with 65nm transistor performance similar to Intel's) until at least the end of 2007.

2:56 PM, December 17, 2006  
Anonymous Anonymous said...

"What a great company is Intel. You buy an upgradeable server and you cannot upgrade it...here."

I think you mistyped - you mean HP right? The upgradeable parts were available had HP chosen to use them (and not try to cut costs on the back of their customers).

Another well informed AMD fan...go buy a 4x4 so you can keep reading INQ links for all of your technical information... I hear it's faster if you have 10,000 instances of IE/Firfox running - you know "Megatasking"

3:01 PM, December 17, 2006  
Anonymous Anonymous said...

LOL the PhD pretender is the most sorry example of a Geek!

A disgruntled ignorant geek who can't even put a compelling argument togather for AMD.

I respect AMD for a lot they have done, but lauch at his sorry attempt put INTEL down...

Sharikou you need to go back to school to get a new PhD.

Hey when is INTEL going BK? Want to put a wager down? I'm good for 1000:1 Odds, no BK in the next 2 years... way past your prediction.

Bottom line AMD behind,
bottom line AMD technology sucks
Bottom line AMD benchmarks are sad

AMD may have Dell but who cares.. INTEL got Apple...
LOL

4:58 PM, December 17, 2006  
Anonymous Anonymous said...

You promosed to do your own Core 2 benchmarking. I kept wondering why did you back off from that promise?

I know the answer!

http://tinyurl.com/yd4387

I certainly understand why you backed off on your promise to benchmark.

9:08 PM, December 17, 2006  
Anonymous edward said...

"It took the conroe e6600 2.17 sec to complete task.
It took the AM2 5000+ only 1.13 sec to complete the same task."


AMD's CPUs have better crypto (incl. random number generation, I suppose) and source/channel coding performance than Intel counterparts.

It just "happens" that none of those review websites thought cryptography is important. They love SuperPi very much, though.

9:50 PM, December 17, 2006  
Anonymous Anonymous said...

"Just some time ago Sharkou was talking about using it's AM2 3800+ on 4x4 ..."

I don't recall AMD ever making that claim. Nice try making a point. Better luck next time.

9:54 PM, December 17, 2006  
Anonymous Anonymous said...

"The test=linux linspire 5.0
platform e6600 at 4791 bogomips
platform AMD AM2 5000+ at 5207 bogomips
Running std. random number generation sequences with generation of 10001 samples. timed event

Results(faster is better)
It took the conroe e6600 2.17 sec to complete task.
It took the AM2 5000+ only 1.13 sec to complete the same task.

However, if this test was run by intel and anand I am sure the results would be different.
So AMD runs std. software at nearly twice the speed of c2d and for less money with less power."

How in the bloody *hell* is any test on Linspire considered standard software? Standard for what? You and the 9 other people on the planet using it?

Congratulations. You've joind Sharidouche in the ranks of complete and utter F*cktards.

10:23 PM, December 17, 2006  
Anonymous Anonymous said...

They are only stressing the CPU's computation power - which is clearly Conroe already wins K8. The benchmark such as these just further verify the claim, but not benchmarking the bandwidth. I really want to see one. Either one wins, we... customers win.

Heck, so far no benchmark site are free enough to stress bandwidth. Duh. :(

10:37 PM, December 17, 2006  
Anonymous Anonymous said...

"The test=linux linspire 5.0
platform e6600 at 4791 bogomips
platform AMD AM2 5000+ at 5207 bogomips
Running std. random number generation sequences with generation of 10001 samples. timed event

Results(faster is better)
It took the conroe e6600 2.17 sec to complete task.
It took the AM2 5000+ only 1.13 sec to complete the same task.

However, if this test was run by intel and anand I am sure the results would be different.
So AMD runs std. software at nearly twice the speed of c2d and for less money with less power."

Perhaps you can do a full blown review yourself and show the world how AMD has been cheated?

I mean really, AMD must be faster than C2D in almost everything right? All the benchmarks and reviewers must have been paid off by Intel - I know it!

10:43 PM, December 17, 2006  
Anonymous Anonymous said...

Anand just did a review of the Woodcrest v Opteron and the Opteron was doing fine and they concluded that.

When people say the websites are biased they invariably mean they have come up with a conclusion that is against the bias shown by the reader to one particular company.

11:06 PM, December 17, 2006  
Anonymous Anonymous said...


I have seen no one ever report that they could reliably repeat the conroe benchmarks as reported by Tons,TG,and Anand without the intel referance platform.


Yes! It's a conspiracy! Intel has bribed every tech site out there - including AMDzone, and millions of consumers into posting false benchmarks for the Core 2 processor.

Everyone knows that Core 2 wipes the floor with AMD's desktop offerings. The 4x4 is a joke. Two processors to compete with one!? Double the power use too!

AMD used to have a benchmarks site on their site, showing how they were faster than Intel. Now all that section strangely contains is comparisons between different AMD processors!

11:38 PM, December 17, 2006  
Anonymous Anonymous said...

"The test=linux linspire 5.0
platform e6600 at 4791 bogomips
platform AMD AM2 5000+ at 5207 bogomips"

P4 3.0 GHz with HT enabled I get 6145.36 bogomips for both virtual CPU's.
Zomg !!!!1112!!1 P4 630 beats the c**p out of e6600 and 5000+.

If you want to do some real tests then at least do something meaningful. A full kernel compiling would be a nice measurement. My e6300 @3.1GHz can do it in about 3 minutes IIRC, old P4 920@3.8GHz did in around 5 minutes. My Athlon XP @2.36GHz does in 15 minutes. All are running Gentoo.

1:41 AM, December 18, 2006  
Anonymous Anonymous said...

And only a moron would believe anything you type. You are a pathetic imbecile. I salute you. One finger at a time. Cunt.

You're hard, where's your handbag?

If you dont like what the nutty professor says then go elsewhere, you disrespectful imbecile!

4:32 AM, December 18, 2006  
Anonymous Columbus said...

Doesn't really important if the idle state of Athlons 64 (not ibernation) is reached at 1 GHz or low by lowering the Vt and the same idle state at 1.6 GHz for Intel's Woody or Conroe, this is an Intel problem. In short words, isn't important HOW you get the cats in the cage, is important a cage filled up of cats!

5:36 AM, December 18, 2006  
Anonymous Rechargeable said...

Hey boys! This Sharikou *seems* too silly too many times, and saying too I mean TOO silly to believe him! I think Sharikou must be an Intel employee, payed to write false AMD "defenses" and false Intel bashing to convince the readers that AMD sucks!
His pro-AMD and versus-Intel arguments are too often too stupid to believe who write this crap is a genuine AMD enthusiast. Am I wrong?

6:02 AM, December 18, 2006  
Anonymous Anonymous said...

Sharikou, any comment about this news??

http://news.moneycentral.msn.com/ticker/article.aspx?Symbol=US:AMD&Feed=Bcom&Date=20061218&ID=6284240

Advanced Micro Nollenberger Capital initiates Sell . Target $13. Nollenberger initiates AMD with a Sell and a $13 tgt saying AMD appears to have rested too long on its laurels in 2006 and is now quite vulnerable in 2007 in terms of CPU/GPU market share as well as ASP declines. The firm says AMD's Athlon64/Opteron micro-architecture is now aging, and a technically elegant native quad core update in 2H07 for high-end servers does not change the co's weakening position in high-volume desktop/notebook segments. The firm says Intel's new Core 2 Duo products have unambiguous performance/price/power advantages in 95% of all PC applications vs. AMD's Athlon64/Opteron. The firm believes aggressive internal capacity expansion of 33%-plus and the addition to multiple product lines from ATI increases execution risk substantially in 2007.

6:53 AM, December 18, 2006  
Anonymous Anonymous said...


You talking at Cool n quiet idle mode or full load? Idle on AMD runs at 1.0GHz, vs Intel's 1.6GHz?

in both, you retard....

AMD 65nm uses only 7.5 watts on idle
conroe uses 30WATTS ( revision C1 wich was out 2 days ago, is susposed to bring the wattage consumition at only 15 watts )

AMD 65nm uses also a MAXIMUM of 65 watts at full load...
Conroe uses 65 watts at "TYPICAL LOAD" wich means 80 watts at FULL LOAD.

diferences are clear.

2:55 PM, December 18, 2006  
Anonymous Anonymous said...

Results(faster is better)
It took the conroe e6600 2.17 sec to complete task.
It took the AM2 5000+ only 1.13 sec to complete the same task.

However, if this test was run by intel and anand I am sure the results would be different.
So AMD runs std. software at nearly twice the speed of c2d and for less money with less power."

How in the bloody *hell* is any test on Linspire considered standard software? Standard for what? You and the 9 other people on the planet using it?

Congratulations. You've joind Sharidouche in the ranks of complete and utter F*cktards.

just like another useless benchmark you intelfanboys love to go all smoochy smoochy.. *cough cough* SuperPi anyone?

2:57 PM, December 18, 2006  
Anonymous Anonymous said...

"You're hard, where's your handbag?

If you dont like what the nutty professor says then go elsewhere, you disrespectful imbecile! "

You negated any point you were trying to make with this post by your last comment.

Any self-respecting, intelligent, individual with a half a grain of independent thinking knows this asshat for what he is. A phony.

He is NOT a ph.d.
He is NOT a journalist.

He IS an ex-Intel employee, and from what I've read, a low-level one at that, who was fired for systems abuse.

So instead of owning up to his mistake and being a man, he trashes the company that rightly fired him, posting gibberish, half-truths (See the comment about C2D being unsuitable for Vista elsewhere in his Blog), and at times outright lies.

I'm amazed that AMD even lets him speak on their behalf. It's embarrassing.

I feel sorry for the people who defend this clown.

3:16 PM, December 18, 2006  
Anonymous Anonymous said...

""Just some time ago Sharkou was talking about using it's AM2 3800+ on 4x4 ..."

I don't recall AMD ever making that claim. Nice try making a point. Better luck next time."

It was Sharikou (and other AMD fanboys) who said that, not AMD itself. Next time read what is written.

3:44 PM, December 18, 2006  
Blogger Scientia from AMDZone said...

Roborat. If you can show me where Anandtech has proven that K8 cannot scale with IMC I'll do an article on this on my blog.

Pessimistic projections for AMD stock are nothing new. These are often based on very little other than faith in Intel.

It is interesting though to suggest that AMD growth is over. AMD still seems to be taking about 0.6% volume share from Intel every quarter. It did this in Q3 even after C2D as released. With this rate of growth AMD can roughly double its size by end of 2008. This would put AMD in 3rd or 4th position on the top 10 list. However, Intel is in no danger of losing the #1 spot. I also do not expect AMD to overtake Samsung which is #2 on the list. I think it is likely that AMD will build a third 300mm FAB and this could bring AMD up to about 40% volume share but this wouldn't happen until around 2011.

Both Intel and AMD are doing experiements two generations ahead. 45nm is pretty well nailed down by now and the heavy R&D is on 32nm with some experimentation on 22nm. There is also a lot new technology moving into manufacuring like submersion lithography. The only thing that Intel is doing that I'm puzzled about is the Tri-Gate. The Tri-Gate should work on 32nm but will be too large for 22nm. In contrast, the AMD FinFet which is double gate should still work at 16nm. I guess that is still a bit in the future and maybe Intel plans to switch to double gate transistors with 22nm.

I've already said that Intel's and AMD architecture will be very similar by end of 2007 and both the architecture and corporate structure will be more similar in 2009. I think AMD can still grow quickly through 2008. I'm not certain though that AMD can gain quickly enough to use a third FAB; that remains a question.

4:40 PM, December 18, 2006  
Anonymous Anonymous said...

"Doesn't really important if the idle state of Athlons 64 (not ibernation) is reached at 1 GHz or low by lowering the Vt and the same idle state at 1.6 GHz for Intel's Woody or Conroe, this is an Intel problem. In short words, isn't important HOW you get the cats in the cage, is important a cage filled up of cats!"

My point is if Intel chose to retarget the idle speed down to 1.0GHz like AMD that would give a substantial idle power reduction - it's not a superior 65nm process or a "healthy" one as the poster I was replying to was implying!

I competely agree it's better (doesn't matter why), but the previous poster's argument about it being due to "healthy" 65nm process is ridiculous...

It is a DESIGN (not process technology) difference and to state AMD's 65nm technology is better or "healthy" based on an idle power consumption comparison of 2 different chip designs (like the poster I was replying to was) is ridiculous if you consider all of the other variables.

To the end user yes I don't care about how they get there - just like I don't care about MCM vs "native" (stupid term) quad core - if performance is similar who cares if it is 2 dual cores "glued". Likewise when Intel comes out with 45nm 1 year earlier than AMD I will not state it is not fair to compare Intel's performance with 45nm to AMD's 65nm if that is the only product available.

5:55 PM, December 18, 2006  
Anonymous edward said...

"I mean really, AMD must be faster than C2D in almost everything right? All the benchmarks and reviewers must have been paid off by Intel - I know it!"

Those Intel fanboys who are fed and raised up by Intel's marketing have no clue of computer performance. If you run a biased configuration/benchmark 100 times, you get 100 biased results.

As I said, the problem is that ALL those websites basically repeat the same set of Core 2-favored benchmarks, in doing which they created Core 2's unduly superiority.

Compared to K8, Core 2 perform worse on cryptography, probably one of the most important CPU-intensive task. In the same manner, Core 2 performs worse on non-sse optimized coding (e.g., in high-order integral fields which cannot be vectorized).

These mathematically intensive programs such as crypto, etc., used to be the standard benchmarks those websites use. Notice how they are lacking after Core 2's release (except when benching a 3Ghz Woodcrest against a 2.6Ghz Opteron).

It's obvious that all those websites have the same agenda to meet - make Core 2 looks better than it is. To be fair, Core 2 does have in average 15% higher IPC than K8, and that is no small feat. The important term here, however, is the "in average". If OTOH one only focuses on the good/better part, his conclusion can be nothing but biased.

7:34 PM, December 18, 2006  
Anonymous Anonymous said...

AMD 65nm uses only 7.5 watts on idle
conroe uses 30WATTS ( revision C1 wich was out 2 days ago, is susposed to bring the wattage consumition at only 15 watts )

That's doubtful, considering there isn't that a 23W difference in idle between Brisbane and Conroe in the Anandtech review, despite Conroe being paired with a chipset that uses 15W or so more than one based on a P965.


AMD 65nm uses also a MAXIMUM of 65 watts at full load...
Conroe uses 65 watts at "TYPICAL LOAD" wich means 80 watts at FULL LOAD.

Yet the Conroe system uses less power than the Brisbane system in all of the CPU intensive tests.

8:06 PM, December 18, 2006  
Anonymous Anonymous said...

"The Tri-Gate should work on 32nm but will be too large for 22nm."

Huh? The wraparound portion of the trigate is in the opposite direction of Lg - not sure what you mean by this comment.

You can gang multiple gates together (as you can see in Intel's publicly available photos) to address things like heat, fin size, etc... The "height" of the trigate I'm sure will scale down to keep AR under control, just as it does with conventional planar transistor technology. The only issue I see is keeping the Si fin (active Si, not the gate) aspect ratio manageable for some of the other processing steps. This has to be done yet you need to keep a reasonable amount of active Si for performance (thus, I think, the concept behind ganging these multiple fins together as seen in the photos with one large source/drain)

Could you explain what you mean by too large? (in what geometric direction? too large for what? a specific process step?)

Scientia - my background is in transistor development so I'd be curious in having an intelligent conversation on this. (as opposed to most of the other posts on this board, i.e Edward please chime in, spewing some google info about Si processing / "APM3.0" which you know nothing about...)

10:53 PM, December 18, 2006  
Anonymous Anonymous said...

Compared to K8, Core 2 perform worse on cryptography, probably one of the most important CPU-intensive task. In the same manner, Core 2 performs worse on non-sse optimized coding (e.g., in high-order integral fields which cannot be vectorized).

Oh man, stop, you're cracking me up. Now Cryptography is super important? That's hilarious. Far more people do media encoding/decoding tasks and that was one of the few things the P4 was as good or better at compared to K8. But you dismissed this back when K8 pwned P4 in most things.

Now you have the gall to a.) lie (Core 2 is equivelent to Opteron, it's not beat on a per-core basis) and b.) make encryption sound like some big important thing?

In fact, if you take off those big geeky AMDroid eyeglasses you can maybe see that the blue line is _still_ on top of the red line all the way up to 4 threads - ergo core 2 wins this one too.

You are truly a laughable dope. On close to _every_ real life benchmark Core 2 'frags' the K8. You can put your head in the sand all you want, but even most of the more rational AMD fanbois have finally admitted this.

10:54 PM, December 18, 2006  
Anonymous Anonymous said...

Scientia - I appreciate your posts - don't always agree with them but as least they aren't blind fanboy post like most around here are...

"45nm is pretty well nailed down by now and the heavy R&D is on 32nm with some experimentation on 22nm."

45nm is pretty well nailed down for Intel (they are already building out D1d for 45nm and process is essentially done); I don't think this is also true for IBM/AMD unless they are truly only doing geometric scaling with minimal performance scaling. (Although this would help explain how they may be able to move to the next node in <2 years)

AMD I believe stated they expect a ~15% improvement on 45nm (I think this may have been at recent analyst day?), if this is true they may be more or less done, but a 15% improvement honestly is not very good (it should be 30+% per node). If true this means their high K/metal gate work has definitely pushed out to 32nm. There is not much blood left to squeeze out of the strain stone and unless they scale oxide gate electrically this would jive with the 15% - meaning essentially no Tox (gate) scaling, probably some minimal improvements on strain and perhaps some minor improvements in the backend interconnect. There are other areas like salicide and implant which can also give some modest fmax improvements.

I think Intel will be a generation ahead on high K/MG as it will likely be announced they are doing this on 45nm (they have mentioned substantial power breakthroughs) - this would be consistent with addressing gate leakage. I suspect we will hear in Q1/Q2'07 when samples start going out the door more liberally and people can start to cleave the chip and do material analysis on it.

Also immersion litho will be much harder than believed in manufacturing - I think it will require a great deal of more effort in terms of yield & manufacturability. Printing development wafers using one immersion tool which is well understood and finely tuned is one thing. Dealing with 5-10 tools that may have slight optics differences, reg issues, etc.. is another.

11:11 PM, December 18, 2006  
Anonymous edward said...

"Now Cryptography is super important? That's hilarious."

Cryptography is more important than you think. It is more important than playing games at 800x600 and hope to get 130fpx instead of 100. It is certainly more important than SuperPi. It is more important than encoding wmv and divx back and forth. Not only crypto and coding, for high-IO bandwidth tasks such as complex database access and web serving, K8 outperforms Core 2, too.

Saying cryptography isn't important for computing is both ignorant and rediculous, since one of the major problems the computer was invented to solve is to encrypt and decrypt.

If you run a mission-critical server, for example, one in the millions of small businesses to handle on-line transactions, cryptography is not only a must, but a performance-limiting must. Any small-business owner who believes the "Core 2 frags all" crap is going to pay more and get less.

At the end of the day, the real "practical" programs that Core 2 runs much better than K8 are the media encodings. Yet first, that's (SSE) really the easy part of the core to improve; and second, servers and (non-graphics) workstations need almost no SSE entirely (which is probably why AMD just made their SSE "good enough" to compete with P4).

"You are truly a laughable dope. On close to _every_ real life benchmark Core 2 'frags' the K8. You can put your head in the sand all you want, but even most of the more rational AMD fanbois have finally admitted this."

The problem of your statement is that while I recognize the programs that Core 2 runs faster, you don't those that Core 2 runs slower. If I have to spend 4 min to encode a video clip instead of 3 min, I'll just wait a bit. If a business owner finds his transaction-handling server handle 20% less secure transactions, he loses revenue. Now try deny that, and you are the one(s) who buries your head in the sand of Intel's web-marketing.

11:24 AM, December 19, 2006  
Anonymous Anonymous said...

Not only crypto and coding, for high-IO bandwidth tasks such as complex database access and web serving, K8 outperforms Core 2, too.
Except Opteron gets easily beaten by Woodcrest in all major enterprise benchmarks, like TPC-C, SAP-SD, SPECjbb2005.

3:02 PM, December 20, 2006  
Anonymous Anonymous said...

Anonymous said:
"Except Opteron gets easily beaten by Woodcrest in all major enterprise benchmarks, like TPC-C, SAP-SD, SPECjbb2005."

I think they compared an Opteron blade with very limited HDD storage (an old Blade LS20 Rev E) vs a full blown Woodcrest with lots of HDD space for the SAP-SD benchmark, which benefits greatly from the extra spindles.

It's not rocket science.

Fit an Opteron out with similar specs, then do a more valid comparison.

6:16 PM, December 20, 2006  
Anonymous edward said...

"Except Opteron gets easily beaten by Woodcrest in all major enterprise benchmarks, like TPC-C, SAP-SD, SPECjbb2005."

TPC-C is a more meaningful measurement of system scalability than sheer CPU performance. People compare TPC-C for system , not processor, decisions. AFAIK, SAP SD benchmark is the same. Also in both case, Power5 and even Tulsa outperforms Core 2, and both Power5 and Opteron has much better scalability than Woodcrest/Clovertown, too.

SPECjbb2005 OTOH is highly memory intensive and could benefit from large cache and FB-DIMM. Even so, when concurrency is high, Woodcrest doesn't win much over Opteron on this benchmark (also depends on whether NUMA is enabled, and which JVM you use).

The main problem I have with your statement however is that "all major enterprise benchmarks." The "major" in your POV may not be sensible at all, or maybe in your world enterprise uses no cryptography?

7:20 PM, December 20, 2006  
Anonymous Anonymous said...

The main problem I have with your statement however is that "all major enterprise benchmarks." The "major" in your POV may not be sensible at all, or maybe in your world enterprise uses no cryptography?
Or maybe in the realworld, cryptography doesn't take very much CPU time. Perhaps you can show the transaction processing tasks which have Opteron even ahead of Woodcrest?

9:09 PM, December 20, 2006  
Anonymous Anonymous said...

I think they compared an Opteron blade with very limited HDD storage (an old Blade LS20 Rev E) vs a full blown Woodcrest with lots of HDD space for the SAP-SD benchmark, which benefits greatly from the extra spindles.
HP made two recent BL series submissions, one for the 5160 Xeon and one of the 2218 Opteron.

9:21 PM, December 20, 2006  
Anonymous Anonymous said...

By the way, sharikaboob, there is negative scaling with these laughable 65nm Brisbanes. The L2 cache is slower, more latency.

Who's laughing now, idiot? Your sarcasm was idiotic to begin with, now that you're shown wrong on every level it's truly hilarious.

10:07 PM, December 20, 2006  
Anonymous edward said...

"Or maybe in the realworld, cryptography doesn't take very much CPU time."

Our conversation will have to end here because you are plain ignorant.

"Perhaps you can show the transaction processing tasks which have Opteron even ahead of Woodcrest?"

Almost everything involving crypto, and a few (not all) web & SQL.

11:26 AM, December 21, 2006  
Anonymous Anonymous said...

Almost everything involving crypto, and a few (not all) web & SQL.
In other words, you can't find any important enterprise benchmarks.

1:41 PM, December 21, 2006  
Anonymous Anonymous said...

http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2893&p=3

65nm has higher L2 latency.

Anandtech frags Sharikou.

2:48 PM, December 21, 2006  

Post a Comment

Links to this post:

Create a Link

<< Home