Tom's Hardware > Forum > CPU & Components > CPUs > Tom's openSSL benchmark results.

Tom's openSSL benchmark results.

Forum CPU & Components : CPUs - Tom's openSSL benchmark results.

Tom's Hardware: Over 1.4 million members in 6 different countries available to answer all your high-tech questions. Sign up now! Its free!
Ask the community Add a reply
Word :    Username :           
 
deanjo   09-30-2008 at 08:09:07 PM

I'm sorry but your benchmarks here:

http://www.tomshardware.com/charts [...] L,844.html

are so out of the realm of reality it's not even funny. When it comes to openSSL results the AMD's quite literally stomp on any intels. This has been proven over and over. As one of the contributors to the Phoronix Test Suite which you use to run your linux benchmarks, real world testing shows that the AMD's are far superior in this test and makes me wonder how you supposedly came up with such unrealistic results. Either you "interpolated" results based on other tests or you compiled the test on a intel based machine and then just swapped the harddrive over booted and re-ran the test without recompiling the test resulting in no optimizations for the AMD processors. These are the only ways that the results can be explained. By not recompiling the suites to take advantage of the feature set of the processors you nullify one of the key points of the Phoronix Test Suite which is to take advantage of processor specific optimizations. Having noticed the results in the openSSL tests I can't help but wonder that all the linux tests are now null and void because of shoddy benchmarking practices.

Dean Hilkewich
Phoronix Test Suite contributor

Add a reply
Sponsored Links
Register or log in to remove.
rtfm   09-30-2008 at 09:54:20 PM
- 0 +

Wouldn't suprise me, Toms has been on a slippery slope for a while now.....

Reply to rtfm
strangestranger   10-01-2008 at 12:25:26 AM
- 0 +

some sites of course have never made it up the slope in the first place.

------------------------------ I'm a git, deal with it.

Antec 1200,PC Power & Cooling 750,Gigabyte DS4-x48,Intel Q9550@3.4 W/Xigmatek S1283,8GB OCZ DDR2 800,ATI 4870X2,X-FI>CA 640C amp>Tannoy R300/Senn 595's
 Reply to strangestranger
wisecracker   10-01-2008 at 01:16:32 AM
- 0 +

(trying to think of something positive to say ...)

At least they are trying to update the CPU Charts - in the past it's been a great resource.

Reply to wisecracker
deanjo   10-01-2008 at 07:50:33 AM
- 0 +

Updating with false and/or inaccurate results is not a helpful resource but increases the chance of somebody buying on inaccurate information and being disappointed with their purchase.

Reply to deanjo
kg4icg   10-01-2008 at 08:09:56 AM
- 0 +

Just 1 little fact you overlooked. AMD can't use DDR3 memory while the X48 motherboards that the Intel chips were put on can. Look at the memory specs before you jump to conclusions.


Message edited by kg4icg on 10-01-2008 at 08:13:05 AM
------------------------------ Did I hit you with a Mack Truck?

 

Reply to kg4icg

apaige   10-01-2008 at 12:57:50 PM
- 0 +

Deanjo: I'm a (modest) PTS contributor myself. The PTS benchmarks are *not* optimized, not for Intel, not for AMD, not for any CPU. It compiles apps with no optimization switches and doesn't handle CFLAGS. I don't know how Tom's Hardware got those results, because even if they were compiled on the Intel system and subsequently run on the AMD system, it doesn't explain the numbers, which are very low, even for Intel CPUs.

 

kg4icg: even with DDR2 RAM, the AMD numbers should be nearly 4 times higher. Even the Intel numbers are way too low. BTW, check out the memory bandwidth bench: the first half ofthe chart is entirely green. Even with DDR2 vs. DDR3 RAM, all AMD CPUs score better than the fastest Intel rig. http://www.tomshardware.com/charts [...] h,807.html


Message edited by apaige on 10-01-2008 at 01:02:19 PM
Reply to apaige
deanjo   10-01-2008 at 05:59:29 PM
- 0 +

apaige, this is not true, upon recompiling of the the tests architecture flags are used especially in the openSSL where it has been optimized for the various architectures. For example just running ./configure on a source that has optimizations will results in:

Quote :


./configure
Detected operating system: Linux
Detected host architecture: x86_64
Checking for cc version ... 4.3, ok
Checking for host cc ... cc
Checking for cross compilation ... no
Checking for CPU vendor ... AuthenticAMD (16:2:3)
Checking for CPU type ... AMD Phenom(tm) 9850 Quad-Core Processor
Checking for kernel support of mmx ... yes
Checking for kernel support of mmxext ... yes
Checking for kernel support of 3dnow ... yes
Checking for kernel support of 3dnowext ... yes
Checking for kernel support of sse ... yes
Checking for kernel support of sse2 ... yes
Checking for kernel support of cmov ... yes
Checking for mtrr support ... yes
Checking for GCC & CPU optimization abilities ... native



Which then results in the binary being compiled with:
-Wall -Wno-switch -Wpointer-arith -Wredundant-decls -O4 -march=native -mtune=native -pipe -ffast-math -fomit-frame-pointer

With the march and mtune flags set to native this selects the CPU to tune for at compilation time by determining the processor type of the compiling machine. Using -mtune=native will produce code optimized for the local machine under the constraints of the selected instruction set. Using -march=native will enable all instruction subsets supported by the local machine (hence the result might not run on different machines).

On a Phenom using native results in an instruction set support for MMX, SSE, SSE2, SSE3, SSE4A, 3dNOW!, enhanced 3dNOW!, ABM and 64-bit instruction set extensions.

On a Core2 it results in an instruction set of MMX, SSE, SSE2, SSE3 and SSSE3 instruction set support,



kg4icg, DDR3 has no bearing on this test and many others. Very few apps are memory bandwidth limited.


Message edited by deanjo on 10-01-2008 at 06:36:36 PM
Reply to deanjo
apaige   10-01-2008 at 07:24:11 PM
- 0 +

Deanjo: point taken, but an Intel-optimized build would still run much faster. And the fact that even the Intel scores are way too low suggests it's a different kind of issue.

Reply to apaige
deanjo   10-01-2008 at 08:51:49 PM
- 0 +

Not necessarily, there are a few more things to consider. First of all just because the proper flags for architecture is used, the application also has to utilize those instruction sets. Many applications don't but applications such as mplayer, GMP, and openSSL do have instruction set specific calls. GCC also has heavy AMD processor specific optimizations. GMP for example when using a generic or intel binary on a AMD processor will result in speeds of 1/2 of what a build done on the AMD system compiled library will do. As well any apps that use yasm for their assembly will show great improvements in speed with their AMD specific code that is autodetected upon the calling of YASM.


Message edited by deanjo on 10-01-2008 at 08:59:15 PM
Reply to deanjo
deanjo   10-17-2008 at 04:49:05 AM
- 0 +

Will someone from Tom's please reply? Because your testing methods in linux are severly flawed.

Processor: AMD Phenom 9850 Quad-Core Processor @ 2.50GHz (Total Cores: 4), Motherboard: ASUSTeK M3N-HT DELUXE, Chipset: Unknown, System Memory: 3965MB, Disk Space: 20GB, Graphics: GeForce 8800 GT 512MB (600/900MHz), Screen Resolution: 1600x1200
http://www.phoronix.net/phoronix-test-suite/remote-graph/graph-remote.php?g=BAR_GRAPH&t=OpenSSL&s=RSA%204096-bit%20Performance&n=0.9.8h&u=Signs%20Per%20Second&i=Toms%20Way;Proper%20Rebuild;&v=68.50;188.22;&p=HIB&x=1.2.2

Full page here http://global.phoronix-test-suite. [...] 7652-29891


Message edited by deanjo on 10-17-2008 at 04:52:13 AM
Reply to deanjo
deanjo   10-17-2008 at 09:22:13 AM
- 0 +

BTW Phoronix Test Suite has now been updated to track motherboard, processor, OS, and compiler version when a test is installed, and if any one ends up getting changed, force re-install forcing thus forcing a recompile of that test to avoid shoddy benchmarking like Tom's is practicing here. It's a hardware benchmark not a OS benchmark.


Message edited by deanjo on 10-17-2008 at 09:25:05 AM
Reply to deanjo
cangelini   10-17-2008 at 09:39:21 PM
- 0 +

Deanjo,

We're looking into this--thanks for bringing it up.

FWIW, in the future, firing off a feedback email will catch more eyes with editorial, as we don't often go through all of the forums.

Again though, thanks!

Message quoted 1 times
Reply to cangelini
deanjo   10-19-2008 at 12:56:21 AM
- 0 +

cangelini wrote :


FWIW, in the future, firing off a feedback email will catch more eyes with editorial, as we don't often go through all of the forums.
!



I would have, unfortunately I cannot seem to find the appropriate email address to do so.

Reply to deanjo
Ask the community Add a reply
Tom's Hardware > Forum > CPU & Components > CPUs > Tom's openSSL benchmark results.
Go to:

There are 1135 identified and unidentified users. To see the list of identified users, Click here.

Forum map
Bestofmedia Forum ©
2000-2009 Bestofmedia Group
Page generated in 0.347 seconds
Please mind

You are about to answer a thread that has been inactive for more than 6 months.
If you still wish to proceed, please ensure that your posting is original and does not duplicate or overlap any prior responses to this thread.

Add a reply Cancel
Sponsored links
  • Ask the community now
  • Publish
Ad
Related Topics
See all relatives topics
They won a badge
Join us in greeting them