mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > Software

Reply
 
Thread Tools
Old 2003-07-30, 15:28   #1
naclosagc
 

11·227 Posts
Default AMD Opteron

Is there any performace advantage from running on an Opteron? Would the software have to be rewritten or recompiled to get an advanatge from the 64 bitness? Are there plans to do this? Thanks.

Andy
  Reply With Quote
Old 2003-07-30, 15:40   #2
ET_
Banned
 
ET_'s Avatar
 
"Luigi"
Aug 2002
Team Italia

479110 Posts
Default

Not yet to all your questions :-)

Luigi
ET_ is offline   Reply With Quote
Old 2003-07-30, 21:23   #3
crash893
 
crash893's Avatar
 
Sep 2002

23×37 Posts
Default

as far as ive heard

no

becuase the "64 bitness" doenst really affect prime that much beucase it relies on floating point opperations

it may help indirectly becuase other things that do take advantage of the 64 bit cpu will reqire less of its time ( like the os and other apps) leaving more time for prime
crash893 is offline   Reply With Quote
Old 2003-07-31, 02:25   #4
QuintLeo
 
QuintLeo's Avatar
 
Oct 2002
Lost in the hills of Iowa

26×7 Posts
Default

George does not have access to an Opteron to do any enhancements for it.

Based on existing benchmark testing I've seen around the web, the existing SSE2 code DOES work well with an Opteron, but is still somewhat slower than a P-IV due to relative clock rates, though fairly close *per* clock despite not having specifically Opteron-optimised code anywhere.

George has indicated that there might be some small gains to make from optimizing the cache usage, but doesn't think it would make a BIG difference.



Same answers for the Athlon64, since it's not yet released and AMD doesn't appear to be sending samples (yet) to anyone other than big manufacturers (esp. motherboard makers).
QuintLeo is offline   Reply With Quote
Old 2003-07-31, 13:13   #5
dsouza123
 
dsouza123's Avatar
 
Sep 2002

2·331 Posts
Default

Running an Opteron vs an Athlon on a 32-bit OS, there should be a few advantages.

SSE2 code will work.

Even though the extra general purpose registers available to 64-bit code aren't available, internally there should be more resources/temporary registers to work with.
Each assembly level instruction is broken down into many smaller internal instructions which take advantage of all available hardware resources ( hidden registers, running multiple internal instructions in parallel, etc ).
There is available extra unused hardware, needed for the extra general purpose registers, which should allow better register management/ register renaming.

The L2 cache is larger, an Athlon has 256k, (exception some old slower 900 Mhz and less had 512k). The Opteron can have 512k or 1024k.
So less cache misses. Also the cache bus width is double.
dsouza123 is offline   Reply With Quote
Old 2003-07-31, 16:05   #6
sdbardwick
 
sdbardwick's Avatar
 
Aug 2002
North San Diego County

2·11·31 Posts
Default

There is an Opteron SMP article up at AMDMB.COM that has some Prime95 benchmarks. I don't know what version they tested with, but the benchmark results only go up to 1792K.

Selected benchmark figures (I chose the highest numbers listed for each FFT size):
[code:1]
256K - 16.628
512K - 35.543
1024K - 78.184
1792K - 149.181[/code:1]

Which puts the Opteron 244 (1.8GHz) at about equal to a 1.6Ghz-ish P4.

While they cast Prime95 in an interesting light - "Prime 95 is a benchmark used to find Mersenne Prime numbers" - at least they provided a link to www.mersenne.org! :)
sdbardwick is offline   Reply With Quote
Old 2003-08-01, 00:36   #7
db597
 
db597's Avatar
 
Jan 2003

7·29 Posts
Default

If I'm not mistaken, the FPU of the Opteron has 3 modes. 1 is the normal mode, another is a hybrid mode and the last is a completely new way of doing things.

I'd expect that the classic FPU mode will be little better than an Athlon (perhaps extra cache, SSE, more FSB/memory bandwidth, and lower latency will help a little).

But what about the other 2 modes? Won't these avoid the antiquated floating point stacks of the x86 architecture? Understandably, this will require a lot of re-coding as a downside.
db597 is offline   Reply With Quote
Old 2003-08-01, 02:25   #8
dsouza123
 
dsouza123's Avatar
 
Sep 2002

2·331 Posts
Default

I believe you are mistaken about the FPU in the Opteron.
It is the same as the Athlon.

The CPU/general instruction set is what has 3 modes.
32-Bit mode uses the same instruction set as an Athlon ( plus SSE2 )
Needs 32-bit OS and 32-bit programs. (use existing programs)

64-Bit mode uses the new instruction set, has extra registers, the registers are 64-bit. ( has SSE2 )
Needs 64-bit OS and 64-bit programs. (recompile or new 64-bit programs)

mixed 32/64-bit can run programs using either instruction set. (has SSE2)
Needs 64-bit OS ( not sure ) and 32-bit or 64-bit programs. (32-bit are existing programs)
dsouza123 is offline   Reply With Quote
Old 2003-08-01, 13:31   #9
Dresdenboy
 
Dresdenboy's Avatar
 
Apr 2003
Berlin, Germany

192 Posts
Default

Quote:
Originally Posted by dsouza123
Running an Opteron vs an Athlon on a 32-bit OS, there should be a few advantages.

SSE2 code will work.

Even though the extra general purpose registers available to 64-bit code aren't available, internally there should be more resources/temporary registers to work with.
Each assembly level instruction is broken down into many smaller internal instructions which take advantage of all available hardware resources ( hidden registers, running multiple internal instructions in parallel, etc ).
There is available extra unused hardware, needed for the extra general purpose registers, which should allow better register management/ register renaming.

The L2 cache is larger, an Athlon has 256k, (exception some old slower 900 Mhz and less had 512k). The Opteron can have 512k or 1024k.
So less cache misses. Also the cache bus width is double.
x87 code is nearly as fast as SSE2 with packed doubles because SSE2 for 2 double precision values in most cases translates into 2 internal Ops. The register file (88 regs) is still the same. And Opteron + Athlon are really fast on x87 code (unlike P4). At least SSE2 doesn't need to expicitly calculate with first register but the needed FXCH in x87 code don't eat any resources besides one of three instruction issue slots.

The number of internal registers depends on the max. possible load of operations. The Opteron has two new pipeline stages to arrange incoming code in a better way for decode and issueing. Currently there are some code alignment and instruction reordering optimizations necessary (by hand or by compiler) to create an optimal instruction stream.

The bigger L2 cache won't help much since mem interface is much faster (max. 6.4GB/s vs. 3.2GB/s with 40-60% higher latency on Athlon) and the cache hitrate of prime95 is already very good (over 97% on my AXP with 256kB L2 AFAIR).

But there are surely possible optimizations by changing code which is currently optimized for P4. There are some things which P4 does fast and Opteron and vice versa.
Dresdenboy is offline   Reply With Quote
Old 2003-08-02, 17:37   #10
trif
 
trif's Avatar
 
Aug 2002

2×101 Posts
Default

The P4 was a dog until George managed to get one in order to do development on it. So I think it is too early to count the Opteron/Athlon64 out.
trif is offline   Reply With Quote
Old 2003-08-05, 12:03   #11
Dresdenboy
 
Dresdenboy's Avatar
 
Apr 2003
Berlin, Germany

192 Posts
Default

Quote:
Originally Posted by trif
The P4 was a dog until George managed to get one in order to do development on it. So I think it is too early to count the Opteron/Athlon64 out.
There was big potential for doing such optimizations since normal FPU code (x87) is very slow on P4 and SSE2 is really fast.

On Opteron x87 is really fast and SSE2 not really faster ;) (it can't do more adds and muls per cycle than x87). But SSE2 has other advantages then - you can use the registers freely for calculation (no need to use the first as one operand) and in 64bit mode you have 16 SSE2 regs instead of 8.
Dresdenboy is offline   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Let's buy GIMPS an Opteron! Xyzzy Lounge 264 2006-08-17 12:39
AMD Athlon 64 vs AMD Opteron for ecm thomasn Factoring 6 2004-11-08 13:25
Opteron web server... Xyzzy Lounge 14 2003-11-05 23:07
Opteron Bottleneck?? Prime95 Hardware 31 2003-09-17 06:54
What will an AMD Opteron be classified as ? dsouza123 Software 4 2003-08-02 14:29

All times are UTC. The time now is 16:51.

Mon Jan 18 16:51:35 UTC 2021 up 46 days, 13:02, 0 users, load averages: 1.93, 2.07, 1.90

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.