Let's buy GIMPS an Opteron!

[Xyzzy]

Looks like P-1 time!

  PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  Command
 3112 gw        39  19  359m 359m  736 R 99.9 72.5 691:50.32 ./mprime

[Dresdenboy]

Originally Posted by GP2

I wonder if there is any other way to tell what revision it is, other than looking at the markings on the CPU, which is impractical after thermal compound and heat sink have been applied. Do the CPUID programs out there handle this yet?

I researched that topic a bit. The CPUID information is enough to find out about the revision. That should be available in most CPUID tools and also in /proc/cpuinfo.

Here's what I found out:
All Opterons have following values:
family 15
model 5

Revision B or earlier show stepping 0 or 1
Revision C shows stepping 8

(Athlon 64 1.8GHz shows family 15, model 4, stepping 8)

CPU-Z additionally shows a revision string, which is "SH7-B3" for some rev B CPUs. I also read "B0" somewhere. So it could be that "B" in this string really means revision B and that there are different B steppings too.

Some really new benchmark results (http://pcweb.mycom.co.jp/benchmarkla.../25/index.html) show that there is not just a little difference but a difference in the two digit percent range. I don't show SSE2 differences in Sandra again (since it's an synthetic benchmark and can suffer overproportionally from a few wrongly (or not at all) optimized instructions). Instead this one (especially MPEG-2 and DIVX) is more interesting - and a real world benchmark (taken from http://pcweb.mycom.co.jp/benchmarkla.../25/page5.html):
http://pcweb.mycom.co.jp/benchmarkla...images/g15.png

Remember: the clock difference between 240 (1.4GHz) and 144 (1.8GHz) is only 28.6%!

The 240 could be a CPU which was sold months ago while the 144 is really new.

Regards,
DDB

[Xyzzy]

mv@opteron:/proc> cat cpuinfo
processor       : 0
vendor_id       : AuthenticAMD
cpu family      : 15
model           : 5
model name      : AMD Opteron(tm) Processor 140
stepping        : 1
cpu MHz         : 1396.059
cache size      : 1024 KB
fpu             : yes
fpu_exception   : yes
cpuid level     : 1
wp              : yes
flags           : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 syscall nx mmxext lm 3dnowext 3dnow
bogomips        : 2785.28
TLB size        : 1088 4K pages
clflush size    : 64
address sizes   : 40 bits physical, 48 bits virtual
power management: ts ttp

[Xyzzy]

$925.13 in donations, $835.32 spent, and $89.81 left...

I just got around to ordering the CD-ROM... It was $16.99 plus $5 for delivery...

http://www.newegg.com/app/viewProduct.asp?description=27-101-204&depa=1

[Bok]

Back in town today, this is my cpuinfo

Opteron64:~ # cat /proc/cpuinfo
processor : 0
vendor_id : AuthenticAMD
cpu family : 15
model : 5
model name : AMD Engineering Sample
stepping : 8
cpu MHz : 1800.028
cache size : 1024 KB
fpu : yes
fpu_exception : yes
cpuid level : 1
wp : yes
flags : fpu vme de pse tsc msr pae mce cx8 sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 syscall nx mmxext lm 3dnowext 3dnow
bogomips : 3591.37
TLB size : 1088 4K pages
clflush size : 64
address sizes : 40 bits physical, 48 bits virtual
power management: ts ttp


so I guess it is a 'C'

I'll start running some tests if I can find the right software.

[Dresdenboy]

Originally Posted by Bok

cpu family : 15
model : 5
model name : AMD Engineering Sample
stepping : 8

Yep, looks like a "C" :)

Originally Posted by Bok

I'll start running some tests if I can find the right software.

You could run Sandra and TMPGEnc on it. Both show a significant advantage for revision C and are available as test/demo versions. And everything else that uses SSE2 (and also SSE) could be tested - as long as it correctly detects the presence.

ScienceMark 2.0 (www.sciencemark.org) has some matrix multiply benchmarks (SGEMM, DGEMM) optimized for different architectures.

And most interesting: Prime95/mprime on a revision C! :) Plus the x87-only results (CpuSupportsSSE2=0).

[Bok]

Originally Posted by Prime95

To bok, who may have a revision C Opteron. If it is revision C run p95tst, do Advanced/Time 20000000, and post the strange timer results in results.txt.

I'm not sure whether I've run the right thing or not as I do not get a results.txt

I just ran mprime (what is p95tst ??) these are the results. Please let me know if this should be something different.

Main Menu

1. Test/Primenet
2. Test/User Information
3. Test/Vacation or Holiday
4. Test/Status
5. Test/Continue
6. Test/Exit
7. Advanced/Test
8. Advanced/Time
9. Advanced/P-1
10. Advanced/ECM
11. Advanced/Priority
12. Advanced/Manual Communication
13. Advanced/Unreserve Exponent
14. Advanced/Quit Gimps
15. Options/CPU
16. Options/Preferences
17. Options/Torture Test
18. Options/Benchmark
19. Help/About
20. Help/About PrimeNet Server
Your choice: 8

Exponent to time (10000000): 20000000
Number of Iterations (10):

Accept the answers above? (Y):
p: 20000000. Time: 76.813 ms.
p: 20000000. Time: 77.903 ms.
p: 20000000. Time: 78.012 ms.
p: 20000000. Time: 76.832 ms.
p: 20000000. Time: 76.813 ms.
p: 20000000. Time: 76.771 ms.
p: 20000000. Time: 76.808 ms.
p: 20000000. Time: 76.771 ms.
p: 20000000. Time: 76.889 ms.
p: 20000000. Time: 77.663 ms.
Iterations: 10. Total time: 0.771 sec.
Estimated time to complete this exponent: 17 days, 20 hours, 29 minutes.

Hit enter to continue:


Bok

[Dresdenboy]

@bok:
You can find p95tst here: http://www.mersenne.org/gimps/p95tst.zip.

[Bok]

yuk,

I'm only running linux I'm afraid, that's a win executable.

I've got a drive with XP (32bit) installed on it for the opteron, but as I'm at work (ssh'd ) I can't swap...

any linux tests I can do ??

Bok

[Dresdenboy]

There are some small and quickly available benchmarks (like bytemark: http://www.tux.org/~mayer/linux/bmark.html) but they are compiler dependend.

Later you could run some tests on WinXP. It's not that urgent to know the results :)

I'm still at work, will go home soon (GMT +1). Then I'll have a look at some SSE2 stuff on Opteron.

[gbvalor]

BTW, would anyone like to volunteer to get some mlucas and glucas timings?

I've just made one in the opteron (our opteron) :)


These are the results for latest snapshot of Glucas v2.9.1
[code:1]
milsec/iter (user time)
FFT(k) round check
on / off
--- ---------
512 52/ 50
576 60/ 59
640 68/ 64
768 79/ 76
896 99/ 97
1024 108/105
1152 125/122
1280 140/135
1536 167/163
1792 206/205
2048 230/223
[/code:1]

The binary is made using SSE2 and the system compiler GCC 3.3.

You can try with a Revision C chip downloading my latest home snapshot

ftp://ftp.oxixares.com/glucas/glucas-2.9.1.tar.gz

Then a usual configure and make. To test, do a selftest

./Glucas -s p

And you will see the timings in 'selftest.out' file.

Guillermo