Barcelona 'openssl speed' benchmarks

[zpdixon42]

[edit: see the 2nd msg on the 4th page in this topic for 32-bit openssl scores for K10]

Anandtech ran a couple 64-bit Linux applications when reviewing Barcelona... I really wish they would have spent the 3 minutes it takes to run 'openssl speed' to see how Barcelona perform on common cryptographic and hash algorithms...

If anyone is aware of how well Barcelona performs on openssl, let me know.

-zpdixon

 

[zpdixon42]

Check this out people... Clock-for-clock, it appears that AMD's quad-core K10 Opterons are 30-35% faster than Intel's Clovertown and Penryn Xeons on RSA operations (Linux 64-bit OpenSSL benchmark performing 2048-bit sign & verify ops):

http://www.tecchannel.de/serve...en/1729228/index9.html

These RSA numbers are very predictable... directly proportional to the clock speed (look at Intel's numbers). By extrapolating the Opteron 8350 scores, a 2.3 GHz K10 Opteron should be slightly faster than a 3.0 GHz Penryn.

This is kind of unexpected for what is purely an integer-based benchmark... Now, to those of you lucky reviewers who have a pair of Opteron 2360SE as well as a pair of 3.0GHz Penryn, please, run a complete 64-bit Linux "openssl speed" benchmark on both platforms... I have no freaking idea why nobody has done this yet. This is a simple standard benchmark tool present on most Linux distros and the numbers it gives are meaningful to sysadmins running heavy SSL/TLS crypto...

 

[SniperDaws]

Riiiiiiiiiiiigggggggggggghhhhhhhhhhhtttttttt.

What i want to know is when you found that link how on earth did you contain your excitement?

 

[brxndxn]

I'll admit I'm pretty damn disappointed at the lack of Barcelona benchmarks out there.. And, I am still pissed off that the reviews don't seem to take into account the prices of the processor..

Why the fvck must every damn review be 'Intel's best' vs 'AMD's best' when sometimes there's a $1000 difference in price between the two processors?

How about platform price, also?

And, why the hell does every hardware review HAVE to have 37 pages when it's much easier for the reader to scroll through one? God.. it's annoying as hell. You do it to, anandtech.com!!

I have way more rants..

 

[aigomorla]

Originally posted by: brxndxn
I'll admit I'm pretty damn disappointed at the lack of Barcelona benchmarks out there.. And, I am still pissed off that the reviews don't seem to take into account the prices of the processor..

Why the fvck must every damn review be 'Intel's best' vs 'AMD's best' when sometimes there's a $1000 difference in price between the two processors?

How about platform price, also?

And, why the hell does every hardware review HAVE to have 37 pages when it's much easier for the reader to scroll through one? God.. it's annoying as hell. You do it to, anandtech.com!!

I have way more rants..

lol... you my friend need a beer :X

 

[firewolfsm]

I think you mean a blunt.

 

[Viditor]

Originally posted by: brxndxn
I'll admit I'm pretty damn disappointed at the lack of Barcelona benchmarks out there.. And, I am still pissed off that the reviews don't seem to take into account the prices of the processor..

What's even more amazing to me is the fact that not a single one of those sparse reviews is actually using one of the shipping chips...
The reviews were all done on the B1 stepping, but it's the BA stepping that's actually shipping...and BA is supposedly at least 5% faster than B1.

 

[dmens]

Originally posted by: Viditor
The reviews were all done on the B1 stepping, but it's the BA stepping that's actually shipping...and BA is supposedly at least 5% faster than B1.

im going to call shenanigans on that one. hard to believe release candidates will be let out in the wild with that much of a performance gap with the "real release".

 

[Viditor]

Originally posted by: dmens

Originally posted by: Viditor
The reviews were all done on the B1 stepping, but it's the BA stepping that's actually shipping...and BA is supposedly at least 5% faster than B1.

im going to call shenanigans on that one. hard to believe release candidates will be let out in the wild with that much of a performance gap with the "real release".

I am going by Kris Kubicki's reporting (and chatter I've heard as well...)
Kubicki Blog

"I waited on publishing this blog, half-expecting to see some more in depth coverage of benchmarks. The 2.0 GHz samples we saw on Monday were of AMD's B1 stepping of Barcelona. But these processors are not the ones we'll see on Newegg's shelves.

Production Barcelona samples come with the BA revision designator. These processors, manufactured after work-week 30 (WW30 for those who work in the corporate world) include errata fixes not present in the chips reviewed on September 10th.

One AMD developer, who wished to remain anonymous for non-disclosure purposes, stated, "B1 versus BA should be at least a 5%, if not more, gain in stream, integer and FPU performance."

An AMD engineer, when confronted with the claim, stated that 5% gains when moving from B1 to BA processors "seem conservative.""

Also noteworthy is that it was a 3rd party developer and not AMD who is making the 5%+ claim...the AMD engineer merely said that it seemed conservative.

 

[Regs]

Viditor, I think we can all agree that AMD is still trying to make the Barc more competitive and AMD is not releasing anything in quantity until they do.

My only worry is when.

 

[Viditor]

Originally posted by: Regs
Viditor, I think we can all agree that AMD is still trying to make the Barc more competitive and AMD is not releasing anything in quantity until they do.

My only worry is when.

I do think (and have heard from OEMs) that Barcelona is already shipping in quantity, at least it's "in quantity" for a server chip that hasn't been qualified yet...that means only 10-20k chips. However, that quantity is pretty much spot on for all server chips at launch (I don't think Woodcrest did any better their first month).

The big test on quantity will be Phenom...we should know right away if there's a lingering problem when Phenom gets launched next month.

Edit:BTW Regs...my real problem at the moment is 2fold.
1. As has been mentioned, there are VERY few reviews so it's hard to get an objective picture.
2. The reviews we have seen don't list the stepping, and if they are indeed stepping B1 they don't represent shipping silicon

Let me be clear that I lay the blame for this squarely on AMD...this has to be one of the most botched launches ever! That said, I really would like to see a BA benched...

 

[AlabamaCajun]

You can get them from New Egg and NCIX (UK). I'm sure others have them but these two I can vouch for. HP and Dell will sell systems soon if not already, I just could not find them aroung the (Sept 15-7). I know a lot of people are waiting on the Benchies, I'm just waiting for the AM2 chips and will get at least 2 regardless of what other product brands are available. I prefer the architecture and power saving features.

 

[OneEng]

That SSL bench was a good find.

I believe that K8 also performed better in SSL vs non-secure HTTP applications. It still was bested by Core 2 in IPC, but it was much closer with the SSL benchmarks.

I think that this must imply some amount of FPU dependence in the SSL benchmark. I don't see how any INT operations would result in such a decisive IPC win for Barcelona over Penryn.

 

[zpdixon42]

SniperDaws: well these performance numbers are important for people in my industry who run cryptographic applications. They suggest we should buy AMD, not Intel, at least when running RSA.

brxndxn is right: the only thing that matters for most people is bang-for-buck. When you look at a price range, say $400-$450, in some areas the Opteron 2350 ($389) is faster and in others it's the Xeon E5345 ($455). Here are some notes I keep about reviews:

Benchmarks where the Opteron 2350 is faster:
- SPECjbb2005 http://anandtech.com/IT/showdoc.aspx?i=3091&p=5 or http://techreport.com/articles.x/13176/4
- MySQL http://anandtech.com/IT/showdoc.aspx?i=3091&p=9
- Winrar http://anandtech.com/IT/showdoc.aspx?i=3091&p=10
- Power consumption http://techreport.com/articles.x/13176/10
- LS-DYNA (actually 2347 vs. X5355, which is in favor of Intel) http://www.topcrunch.org/bench...ils.sfe?query=2&id=740 http://www.topcrunch.org/bench...ils.sfe?query=2&id=690
- SPECfp_rate2006 peak http://blogs.zdnet.com/Ou/?p=735
- Valve's Map Compilation Benchmark, perf/watt with 12+ GB RAM (the 2347HE also beats the E5345) http://www.theinquirer.net/default.aspx?article=42434
- OpenSSL 2048-bit RSA sign ops (31% faster clock-for-clock) http://www.tecchannel.de/serve...en/1729228/index9.html
- OpenSSL 2048-bit RSA verify ops (34% faster clock-for-clock) http://www.tecchannel.de/serve...en/1729228/index9.html

Benchmarks where the Xeon E5345 is faster:
- LINPACK (Intel optimized) http://anandtech.com/IT/showdoc.aspx?i=3091&p=6
- 3DS Max 9 Arch. HD http://anandtech.com/IT/showdoc.aspx?i=3091&p=8
- Fritz Chess http://anandtech.com/IT/showdoc.aspx?i=3091&p=11
- Valve VRAD map build time http://techreport.com/articles.x/13176/4
- Cinebench http://techreport.com/articles.x/13176/5
- POV-Ray http://techreport.com/articles.x/13176/5
- MyriMatch http://techreport.com/articles.x/13176/6
- STARS Euler3d http://techreport.com/articles.x/13176/6
- Folding@Home http://techreport.com/articles.x/13176/7
- The Panorama Factory http://techreport.com/articles.x/13176/8
- picCOLOR http://techreport.com/articles.x/13176/8
- Windows Media Encoder http://techreport.com/articles.x/13176/9
- SiSoft Sandra Mandelbrot http://techreport.com/articles.x/13176/9
- Valve's Map Compilation Benchmark, execution time http://www.theinquirer.net/default.aspx?article=42434

The bottom line is: Barcelona is competitive against Clovertown in the entry-level and mid-range 2P market (where 70-80% of the 2P sales happen).

 

[Phynaz]

Originally posted by: dmens

Originally posted by: Viditor
The reviews were all done on the B1 stepping, but it's the BA stepping that's actually shipping...and BA is supposedly at least 5% faster than B1.

im going to call shenanigans on that one. hard to believe release candidates will be let out in the wild with that much of a performance gap with the "real release".

I'm with you. I don't know of any chip ever made that had any significant performance tuning between release candidates. RC's are for bug fixes, not performance tuning.

 

[Viditor]

Originally posted by: Phynaz

Originally posted by: dmens

Originally posted by: Viditor
The reviews were all done on the B1 stepping, but it's the BA stepping that's actually shipping...and BA is supposedly at least 5% faster than B1.

im going to call shenanigans on that one. hard to believe release candidates will be let out in the wild with that much of a performance gap with the "real release".

I'm with you. I don't know of any chip ever made that had any significant performance tuning between release candidates. RC's are for bug fixes, not performance tuning.

If you re-read Kris's blog, you'll see that the performance enhancement was actually a bug fix...

"These processors, manufactured after work-week 30 (WW30 for those who work in the corporate world) include errata fixes not present in the chips reviewed on September 10th"

The fixes resulted in a net 5%+ gain in performance...

 

[Phynaz]

Originally posted by: Viditor

Originally posted by: Phynaz

Originally posted by: dmens

Originally posted by: Viditor
The reviews were all done on the B1 stepping, but it's the BA stepping that's actually shipping...and BA is supposedly at least 5% faster than B1.

im going to call shenanigans on that one. hard to believe release candidates will be let out in the wild with that much of a performance gap with the "real release".

I'm with you. I don't know of any chip ever made that had any significant performance tuning between release candidates. RC's are for bug fixes, not performance tuning.

If you re-read Kris's blog, you'll see that the performance enhancement was actually a bug fix...

"These processors, manufactured after work-week 30 (WW30 for those who work in the corporate world) include errata fixes not present in the chips reviewed on September 10th"

The fixes resulted in a net 5%+ gain in performance...

I don't need to re-read hearsay, thanks.

 

[zpdixon42]

OneEng: the OpenSSL RSA implementation is based on the internal BN (big number) library. It exclusively makes use of integer operations. Some wild guesses as to why K10 is faster in this integer-based benchmark:
- K10 has a 64-kB L1 data cache, vs. only 32 kB in Intel's Core microarchitecture
- K10 has a branch prediction unit able to store up to 24 return address, vs. only 16 addresses in Core
- K10 can fetch up to 32 bytes of instructions, vs. only 16 bytes in Core
- K10 has a 12-stage integer execution pipeline, vs. 14-stage in Core (the more stages you have, the higher will be the penalty for mispredicted branches, various Pentium 4 models had 20- and 31-stage integer pipelines !)

 

[dmens]

Originally posted by: zpdixon42
OneEng: the OpenSSL RSA implementation is based on the internal BN (big number) library. It exclusively makes use of integer operations. Some wild guesses as to why K10 is faster in this integer-based benchmark:
- K10 has a 64-kB L1 data cache, vs. only 32 kB in Intel's Core microarchitecture
- K10 has a branch prediction unit able to store up to 24 return address, vs. only 16 addresses in Core
- K10 can fetch up to 32 bytes of instructions, vs. only 16 bytes in Core
- K10 has a 12-stage integer execution pipeline, vs. 14-stage in Core (the more stages you have, the higher will be the penalty for mispredicted branches, various Pentium 4 models had 20- and 31-stage integer pipelines !)

uh, yeah, that'd explain why c2d is kicking ass in just about every other integer benchmark.

 

[dmens]

Originally posted by: Viditor
If you re-read Kris's blog, you'll see that the performance enhancement was actually a bug fix...

"These processors, manufactured after work-week 30 (WW30 for those who work in the corporate world) include errata fixes not present in the chips reviewed on September 10th"

The fixes resulted in a net 5%+ gain in performance...

sorry i find that almost impossible. a 5% performance deviation from modeling should have been found almost immediately and fixed well before qualification.

that is unless this 5% gain is a frequency fix, but that of course means the performance of the supposed "BA" stepping will be the same as the ones reviewed when clocked at the same frequency.

 

[zpdixon42]

dmens wrote: uh, yeah, that'd explain why c2d is kicking ass in just about every other integer benchmark.

I have contributed some optimized assembly code to OpenSSL, I know what I am talking about. Its RSA implementation does NOT use the FPU at all. Heck look at its source code, this is an opensource project. The world is not black & white. It's not like Core has to win every integer benchmark and K10 has to win every floating point benchmark. There are dozens of architectural differences (see list in my previous post) which might advantage, in some cases, a processor you wouldn't expect to perform better than its competitor.

 

[SniperDaws]

id hate to get into an argument with you lot, Ewwwww proper geeks, i wouldnt have a leg to stand on, you'd blind me with science and win every time. all i know is the K10 isnt going to be the big deal everyone is expecting it to be.

 

[Viditor]

Originally posted by: SniperDaws
id hate to get into an argument with you lot, Ewwwww proper geeks, i wouldnt have a leg to stand on, you'd blind me with science and win every time. all i know is the K10 isnt going to be the big deal everyone is expecting it to be.

And you know this because the Magic 8 Ball told you so?

 

[AlabamaCajun]

Originally posted by: zpdixon42

dmens wrote: uh, yeah, that'd explain why c2d is kicking ass in just about every other integer benchmark.

I have contributed some optimized assembly code to OpenSSL, I know what I am talking about. Its RSA implementation does NOT use the FPU at all. Heck look at its source code, this is an opensource project. The world is not black & white. It's not like Core has to win every integer benchmark and K10 has to win every floating point benchmark. There are dozens of architectural differences (see list in my previous post) which might advantage, in some cases, a processor you wouldn't expect to perform better than its competitor.

Absolutely correct! All these benchmarks floating around are running different levels of optimizations and some are specific to one brand.

 

[AlabamaCajun]

Originally posted by: SniperDaws
id hate to get into an argument with you lot, Ewwwww proper geeks, i wouldnt have a leg to stand on, you'd blind me with science and win every time. all i know is the K10 isnt going to be the big deal everyone is expecting it to be.

Give K10 some time. It will move ahead until a XEON Neha gets it's pads in a socket. I'm not saying K10 will wipe all the floors with every xeon but it's coming forth with enought to hold a good market. Intel is pushing hard to keep this difference from becoming a problem again. Most of it is at the expense of faster and hotter main boards to support those high FSBs. AMD still holds an advantage with the onboard FSB meaning I can change CPUs without needing a new mobo. As an overclocker you will want the new mobos for K10 next year for power settings but for servers runnng socket F, you're good to go until you need full support of HT3.

"She blinded me with Science" Thomas Dolby!