Tuesday, June 20, 2006

Woodcrest semi-officially fragged before launch

After massive amount of benchmarketing fraud, Woodcrest is set to launch in a week. A French site got hold of two Woodcrest CPUs (2.6GHZ, 4MB L2) and did some compression/decompression testing against one Opteron 180 (2.4GHZ DC, Socket 939) on the linux-2.4.32.tar.bz2 file. Since the Opteron 180 is a one way dual core CPU, and a 2 way Woodcrest has four cores, we should only look at the numbers for 2 simultaneous processes.

For decompressing the bz2 file, the Opteron 180 dual core CPU finished two simultaneous tasks in 10 seconds. The two way 4 core Woodcrest finished in 14 seconds. Opteron 2.4GHZ is 40% faster.

For compressing the tar file into bz2, the Opteron 180 finished two tasks in 46 seconds, the Woodcrest finished in 58 seconds. Opteron 26% faster.

This should be a very reproducible and verifiable test. You should be able to get the linux-2.4.32.tar.bz2 file from linux.org.

I know Intel fanbois are going to complain that "bzip" wasn't compiled with an Intel compiler, which would have crippled the AMD chip.

36 Comments:

Anonymous Anonymous said...

I don't mean to criticize, but no test setup. If this were Intel who set this up or if the test showed Intel wining, Jesus!, I could hear it now, Intel is cheating, paid pumper...etc.

I do not know anything about server chips, but why are we only looking at the first coulmn?

As you go to the right, (you say AMD is capable of adding another chip and almost doubling performance), it seems that AMD would be slower. And substantially.

Please explain thanks.

7:29 AM, June 20, 2006  
Anonymous Anonymous said...

Another one:

http://realworldtech.com/forums/index.cfm?action=detail&id=68853&threadid=68853&roomid=11

8:07 AM, June 20, 2006  
Anonymous Anonymous said...

Great job at cherry picking the benchmarks to make it look like the Opteron is faster. Look at all of the other benchmarks. Woodcrest is faster in almost all of them.

8:08 AM, June 20, 2006  
Anonymous Anonymous said...

Quoted from that web:"
A detail is rather interesting: the management of the multitask seems more effective on the Intel platforms than on those of AMD. Indeed, that one compares Opteron 150 64bits with the Xeon-HT, this same Opteron 150 64bits with the P4-D, Opteron 180 64bits with the Bi-Xeon-HT, or this same Opteron 180 64bits with very new Woodcrest 64bits, processors AMD is left better than their competitors with a reduced number of simultaneous processes, but loses this advantage there as soon as this same number of processes becomes relatively important.
"

8:36 AM, June 20, 2006  
Blogger Ajay S. said...

Quite unexpected results,

too bad the site has not published the system configurations used.

8:39 AM, June 20, 2006  
Anonymous Anonymous said...

Is that it one benchmark ?

I suppose they are too busy watch France play footbalol badly in the World cup to do more than that ! :D

Very strange.....

8:53 AM, June 20, 2006  
Anonymous Anonymous said...

A. Opteron 180 (2x2.4GHz) 64bits
B. Dual-Woodcrest (4x2.6GHz) 64bits

Difference in MHz = 8%.
Simple math: 2xAx8% = B.

Process: 5 6 7 8 9 10
2xAx8%.: 47, 63, 68, 75, 88, 104
B......: 70, 77, 92, 117, 123, 132

Now imagine the result with a quad or octo core...

9:31 AM, June 20, 2006  
Blogger Sharikou, Ph. D. said...

Great job at cherry picking the benchmarks to make it look like the Opteron is faster.

This is not true. I covered almost al Woodcrest benchmarks. I am curious about it. But Intel dudes are always playing hide and seek. Even in results published by Intel, Opteron 2.6GHZ is faster than Woodcrest 3GHZ in some cases.

11:03 AM, June 20, 2006  
Anonymous Anonymous said...

"I know Intel fanbois are going to complain that "bzip" wasn't compiled with an Intel compiler, which would have crippled the AMD chip."

Jeez you really do like clutching at straws..

What an intel-optimised compiler will do is allow SSE to come to the fore. It doesn't necessarily follow that the intention is to 'hobble' K8.

I think you'll find both comp/decompression and also en/decryption (another area where Core 2 hasn't yet dominated) will improve markedly once code is written to take advantage of the newer SSE optimisations.

11:03 AM, June 20, 2006  
Blogger Sharikou, Ph. D. said...

I think you'll find both comp/decompression and also en/decryption (another area where Core 2 hasn't yet dominated) will improve markedly once code is written to take advantage of the newer SSE optimisations.

Compression/decompression are integer opreations. I wonder why Intel got such a ultra high score in spec_int_base2000 but can't handle compression/decompress well.

11:29 AM, June 20, 2006  
Anonymous Anonymous said...

"As you go to the right, (you say AMD is capable of adding another chip and almost doubling performance), it seems that AMD would be slower. And substantially."

Sorry for my comment, I was looking at the Opterom 150.

Could somebody check these numbers... vs. the number "Wirmish" did.

I am not sure about this and I am really trying to understand it.

Also...

Does "Bi-dual-Woodcrest (4x2.6GHz)" mean 4 chips (8 processors), or is it 2 dual core 2.6Ghz? That whole Bi dual thing threw me.

Thanks alot.

11:50 AM, June 20, 2006  
Blogger "Mad Mod" Mike said...

That website is trying to say the Intel systems multitask better, but the 180 consistently beats Woodcrest in up to almost 3 processes. The 180 beats the P-D in every single test. Comparing a 180 to Quad Woodcrests is just freakin' stupid, even 4 Netbursts can beat 2 Opterons, albeit by a small margin.

1:14 PM, June 20, 2006  
Blogger "Mad Mod" Mike said...

"Could somebody check these numbers... vs. the number "Wirmish" did."

IDK about Wirmish, but that's comparing 2 CPU's to 4, and the 2 CPU's aren't being beaten that badly for having a 100% deficiency in performance power, which goes to show the real supreme architecture.

1:16 PM, June 20, 2006  
Anonymous Anonymous said...

Thanks Mad Mod Mike.

1:29 PM, June 20, 2006  
Blogger Sharikou, Ph. D. said...

Does "Bi-dual-Woodcrest (4x2.6GHz)" mean 4 chips (8 processors), or is it 2 dual core 2.6Ghz?

Woodcrest/Bensley is incapbale of 4P (8 core) operation. Woodcrest is for low end DP market only.

1:42 PM, June 20, 2006  
Anonymous Anonymous said...

Yeah, that bit about the Two 2.6 GHz woodcrest chips scaling better to a higher number of threads than the single 2.4 GHz Opteron seemed pretty funny. You mean 4 cores beats 2 cores in multiple threading? You don't say. How about showing 2 Woodcrests vs. 2 Opterons? Or, even better (heh heh) 4 vs. 4? :)

The 'secret weapon' of a huge, shared L2 cache for the dual-core Woodcrest is showing its limitations...

1:59 PM, June 20, 2006  
Anonymous Anonymous said...

"What an intel-optimised compiler will do is allow SSE to come to the fore. It doesn't necessarily follow that the intention is to 'hobble' K8."

not true anymore. in old good days, icc (that is intel c compiler, best x86 compiler for linux) produced better code for k8/opteron than for p4 (of course with p3 generic optimization, something which could be seen as optimization-level-equally-good-for-all-cpus-we-can-meet-at-our-customers, ie p3/p4/k8). this could not last long and in brand new, 9.0 version check for k8 was included to disallow optimal code generation.

competition in its best.

2:37 PM, June 20, 2006  
Anonymous Anonymous said...

the amds get fragged after the 3rd or 4th simulan task. why is that so? has someone an idea why the performance for one task is much better for amd than intel?

3:23 PM, June 20, 2006  
Anonymous Anonymous said...

My numbers are an approximation. For them to be correct, the Opteron scaling must be perfect... and I known is not possible.

Real world Opteron scaling:
http://www.amd.com/DAM/data/cordacharts/105411.jpg (2P=177%)
http://www.amd.com/DAM/data/cordacharts/91687.jpg (2P=181%, 4P=370%)
http://www.amd.com/DAM/data/cordacharts/93335.jpg (2P=196%, 4P=372%)

Seems like 85% is more 'accurate' for 2P.
(But look at the Linux score in the first link)

The 4P scores (scaling ~93%) are excellent!
Not bad for a 1MB L2 cache chip...

3:33 PM, June 20, 2006  
Anonymous Anonymous said...

Cherry picking single benchmarks again? What about the rest of the benchmarks which Woody pwned Opterons... Like to see real world benchmarks such as SpecInt, SQL server , web server, etc.. cause these are the REAL stuff that server runs..

7:08 PM, June 20, 2006  
Anonymous Anonymous said...

Sharik...how much does AMD pay you? For once be fair.

11:05 PM, June 20, 2006  
Blogger Sharikou, Ph. D. said...

Like to see real world benchmarks such as SpecInt, SQL server , web server, etc.. cause these are the REAL stuff that server runs..

In any real indepdendent tests not using Intel compilers, Opteron can frag Woodcrest

11:49 PM, June 20, 2006  
Anonymous Anonymous said...

"In any real indepdendent tests not using Intel compilers, Opteron can frag Woodcrest" - Proof?

1:49 AM, June 21, 2006  
Anonymous Anonymous said...

"Sharik...how much does AMD pay you? For once be fair."

I was wondering this myself. Everywhere i have looked on this blog, i have seen sharikou talking garbage about intels products. Either this guy is payed by AMD, or the Intel chairman drover over his dog. At any rate, NOTHING this guy says can be taken even remotely serious.

4:46 AM, June 21, 2006  
Anonymous Anonymous said...

In any real indepdendent tests not using Intel compilers, Opteron can frag Woodcrest

which opteron frags which woodcrest?

4:57 AM, June 21, 2006  
Anonymous Anonymous said...

Looks like AMD has lost the game this time around:

Conroe beats Athlon in this review:http://www.digit-life.com/articles2/cpu/intel-conroe-2-13-ghz.html

Yonah beats Turion X2 (so Merom will definitely outperform Turion x2 then: http://www.pcworld.com/reviews/article/0,aid,126190,00.asp

Maybe in 2007, either with 65 nm or quad core or both, AMD will be able to beat Intel again.

5:15 AM, June 21, 2006  
Anonymous Anonymous said...

SpecWeb2005 results:

2 x Opteron 280, 2.4 GHz - 8394
Woodcrest - 9182.

6:30 AM, June 21, 2006  
Anonymous Anonymous said...

Intel's Frag...

http://www.theinquirer.net/?article=32555

10:27 AM, June 21, 2006  
Anonymous Anonymous said...

Whether or not Sharikou is paid for his words matters nothing. As an individual he is under no obligation to be objective. He has an opinion and does not claim otherwise.

He doesn't publish his own benchmarks, he just offers his opinions on publicly available information, which he always references.

No one is forced to take his words at face value. You can read Sharikou's opinion, check the referenced facts, and research the issue more if you want, then come to your own conclusions. "Blaming" someone for having an opinion is downright silly.

You have eyes to see, a brain to think and a heart to judge. Use them.

11:30 AM, June 21, 2006  
Blogger Altamir Gomes said...

"No one is forced to take his words at face value. You can read Sharikou's opinion, check the referenced facts, and research the issue more if you want, then come to your own conclusions. "Blaming" someone for having an opinion is downright silly."

It's not all that hard, buy hey this is multibillionaire IT industry. Talk smoke and mirrors when somebody like Sharikou puts the big guys to stress (remember Van?)...

12:44 PM, June 21, 2006  
Blogger Sharikou, Ph. D. said...

Whether or not Sharikou is paid for his words matters nothing.
These Intel dogs know I am not paid for my writings, they just want to throw mud and confuse the issues.

3:46 PM, June 21, 2006  
Blogger "Mad Mod" Mike said...

'These Intel dogs know I am not paid for my writings, they just want to throw mud and confuse the issues."

What do you expect them to do after countless times you've proven how lying and cheating Intel is, you give them proof that their products are severely inferior, and of course the Intel fanboy morons will get mad, it's common nature for those idiots.

5:47 PM, June 21, 2006  
Anonymous Anonymous said...

Another Conroe article shown crushing AMD...

http://www.pcper.com/article.php?aid=265

11:19 AM, June 22, 2006  
Anonymous Anonymous said...

Uhhh, do servers do a lot of bzip decompression? Excuse me for not being impressed with your findings. Show some server software benchmarks where Woodcrest does this bad and I may think you're onto something. But otherwise I will just say it is more Intel bashing.

2:06 PM, June 22, 2006  
Blogger Sharikou, Ph. D. said...

do servers do a lot of bzip decompression

If you enable mod_gzip, then every apache response is gzipped before sent back.

Intel fanbois do SuperPI everyday on their servers and enjoy a 80% performance boost

2:16 PM, June 22, 2006  
Blogger "Mad Mod" Mike said...

'Intel fanbois do SuperPI everyday on their servers and enjoy a 80% performance boost"

They're waiting for the day that SuperPI does their taxes and is a database application.

2:51 PM, June 22, 2006  

Post a Comment

<< Home