FPU banchmarks

[JSt0rm]

ok I've done some more research on my audio card (from other post in cpu) and it is possible to run a amd cpu/nforce chipset combo. They tested Athlon XP Thoroughbred/Barton CPU and nforce2 but I see no reason why a nforce3 and amd 64 would not work.

The question at this point is who does floating point the best. I have always thought intel/p4 was the king in this area but I would like to rethink this in light of my other thought that amd was impossible.

Does anybody have benchmarks of this feature? I already know that the 64 will beat that pants off my 2.4c when it comes to doom3 and such but in reality I should be buying cpus for work reasons and not for play ;p

 

[Markfw]

Do you mean Sandra arithmetic mark ? Bench yours, and I will try my Athlon64 at 2.1, and My Opterons at 2.2 and let you know.

 

[JSt0rm]

Ok this is the newest sisoft sandra software. I tried running in diadnostics mode and that test was only 1 point higher in each test then this restarted xp load with a couple windows open so this will do. This will be interesting to see.




SiSoftware Sandra

Benchmark Results
Dhrystone ALU : 7216MIPS
Whetstone FPU : 3001MFLOPS
Whetstone iSSE2 : 5282MFLOPS

Performance Test Status
Run ID : LUCID on Thursday, September 09, 2004 at 4:16:45 PM
NUMA Support : No
SMP Test : No
Total Test Threads : 2
SMT Test : Yes
Number of Test SMT (per CPU) : 2
Dynamic MP/MT Load Balance : No
Processor Affinity : P0 L1
Number of Runs : 64000 / 640

Processor
Model : Intel(R) Pentium(R) 4 CPU 2.40GHz
Speed : 2.40GHz
Performance Rating : PR3192 (estimated)
Type : Standard
L2 On-board Cache : 512kB ECC Synchronous ATC (8-way sectored, 64 byte line size)

Chipset 1
Model : Micro-Star International Co Ltd (MSI) 82865G/PE/P, 82848P DRAM Controller / Host-Hub Interface
Front Side Bus Speed : 4x 200MHz (800MHz data rate)

Features
SSE2 Technology : Yes
SSE3 Technology : No
HTT - Hyper-Threading Technology : Yes

 

[jhu]

what kind of fpu test?

 

[JSt0rm]

the sandra arithmetic cpu test

 

[Vee]

Originally posted by: JSt0rm01
ok I've done some more research on my audio card (from other post in cpu) and it is possible to run a amd cpu/nforce chipset combo. They tested Athlon XP Thoroughbred/Barton CPU and nforce2 but I see no reason why a nforce3 and amd 64 would not work.

The question at this point is who does floating point the best. I have always thought intel/p4 was the king in this area but I would like to rethink this in light of my other thought that amd was impossible.

Does anybody have benchmarks of this feature? I already know that the 64 will beat that pants off my 2.4c when it comes to doom3 and such but in reality I should be buying cpus for work reasons and not for play ;p

There no longer is such a simple thing as FPU. The important thing is how the code is organized and compiled. I would suspect your audio use of FP, can and have been done as SSE2 packed math. This is the ONLY kind of FP the Intel/P4 is competitive on! In case of all other kinds of FP math, Intel is totally outclassed by all AMDs, Durons, XPs, A64s, Semprons. But AMD socket A does not support SSE2, so if the code does not also include optimizations for 3DNow+, as an alternative for AMD, any socket A CPU would do poorly in comparisons with P4 on SSE2. A64s finally, support everything.

SSE2 vectorized FP loops is the only kind of FP that is benchmarked these days. This means that these benchmarks (Sandra, PCMark, SysMark etc.) are highly misleading about true FP performance. In your case, I would guess it's somewhat relevant though.

 

[Markfw]

SiSoftware Sandra

Test Status
NUMA Support : No
SMP Test : No
Total Test Threads : 1
SMT Test : No
Dynamic MP/MT Load Balance : No
Processor Affinity : No
Number of Runs : 64000 / 640

Processor
Model : AMD Athlon(tm) 64 Processor 3000+
Speed : 2.10GHz
Model Number : 3000 (estimated)
Performance Rating : PR3150 (estimated)
Type : Standard
L2 On-board Cache : 512kB ECC Synchronous Write-Back (16-way, 64 byte line size)

Chipset 1
Model : ASUSTeK Computer Inc Apollo K8HTB CPU to PCI Bridge
Front Side Bus Speed : 1x 210MHz (210MHz data rate)

Chipset 2
Model : Advanced Micro Devices (AMD) Athlon 64 / Opteron HyperTransport Technology Configuration
Front Side Bus Speed : 1x 210MHz (210MHz data rate)

Features
SSE2 Technology : Yes
SSE3 Technology : No
HTT - Hyper-Threading Technology : No

I will update with the Opteron in a minute

Edit: where are the results, hang on
8746 Dry mips, 3317/4333 mflops whet

How do you get it do put the output for you ?

Opteron:
SiSoftware Sandra

Test Status
NUMA Support : No
SMP Test : Yes
Total Test Threads : 2
SMT Test : No
Dynamic MP/MT Load Balance : No
Processor Affinity : No
Number of Runs : 64000 / 640

Processor
Model : 2x AMD Opteron(tm) Processor 248
Speed : 2.19GHz
Model Number : 3490 (estimated)
Performance Rating : PR6580 (estimated)
Type : Standard
L2 On-board Cache : 1024kB ECC Synchronous Write-Back (16-way, 64 byte line size)

Chipset 1
Model : Advanced Micro Devices (AMD) Athlon 64 / Opteron HyperTransport Technology Configuration
Front Side Bus Speed : 2x 796MHz (1592MHz data rate)

Chipset 2
Model : Advanced Micro Devices (AMD) Athlon 64 / Opteron HyperTransport Technology Configuration
Front Side Bus Speed : 2x 796MHz (1592MHz data rate)

Features
SSE2 Technology : Yes
SSE3 Technology : No
HTT - Hyper-Threading Technology : No

17908 dry mips, 6865/9054 mflops on the Opteron. I guess they do kill the P4 !
I just figured out that the 9054 is the isse2 mark

 

[JSt0rm]

copy to clipboard button and paste

 

[Markfw]

I did, and you see the results, except I had to add the benchmark results myself at the bottom !!!???@#%#$%#@% Stupid Sandra

 

[JSt0rm]

the dual cpu solution is more then I need at this time and the amd 3k isnt showing me some huge gains but its not less so thats a good thing.

 

[JSt0rm]

btw thanks for running those tests ;p

 

[jhu]

Originally posted by: JSt0rm01
the sandra arithmetic cpu test

ok, if that's pretty much the only fpu intensive application you'll be using, then you'll probably want a p4. if you're going to be doing any other serious work such as simulating cloud formations on your cluster, you'll want either itanium2 or power5.

 

[JSt0rm]

No no I use cubase sx and I'm not doing anything that crazy however I can bring a cpu down to a crawl considering the work needs to be done in as close to realtime as I can get (setting up delays per channel on my digital mixer is time consuming and not fun). I work at 3ms to 6ms latency depending and my card can support 1.5ms latency.

So the focus is keeping the cpu at 50% or less and getting the latency as low as possible. Of course the size of projects matters but the 2.4 has gotten me pretty far with some heavy xp tweaks.

I will keep away from the land of hobby weather simulations

 

[jhu]

oh, well if it supports sse2, then you may still want to get a p4

 

[zephyrprime]

Check out here: http://www.spec.org/cpu2000/results/
Sandra's not so great. I'm surprised so many people use it but I suppose that's because it's freely available.

And don't discount the SSE performance. SSE is very important becuase it's the recommended replacement for the old x86 FPU even for non-packed applications.

 

[zephyrprime]

Originally posted by: JSt0rm01
No no I use cubase sx and I'm not doing anything that crazy however I can bring a cpu down to a crawl considering the work needs to be done in as close to realtime as I can get (setting up delays per channel on my digital mixer is time consuming and not fun). I work at 3ms to 6ms latency depending and my card can support 1.5ms latency.

So the focus is keeping the cpu at 50% or less and getting the latency as low as possible. Of course the size of projects matters but the 2.4 has gotten me pretty far with some heavy xp tweaks.

I will keep away from the land of hobby weather simulations

Since your application is done in realtime and sensitive to hiccups, you may want to use the Pentium with HyperThreading. HT can do a lot to ensure greater responsiveness.

 

[JSt0rm]

is there a free version? I mean that program would cost me $500 I would hope its better then some freeware. ;p

 

[JSt0rm]

Yah I have HT enabled. I might just upgrade my cpu to a 3.2 prescott. Its a $200 upgrade so its not a huge bite. Its just odd that the sse and sse2 instruction set isnt really supported very well by the amd solutions.

I would still love to have a a64 for gaming and its performance wasn't horrid but I think that a 3.0c or e would outperform the 3k in sse1 and sse2. But why is this considering both cpus support it and the a64 makes better use of its clock cycles?

 

[jhu]

Originally posted by: zephyrprime
Check out here: http://www.spec.org/cpu2000/results/
Sandra's not so great. I'm surprised so many people use it but I suppose that's because it's freely available.

And don't discount the SSE performance. SSE is very important becuase it's the recommended replacement for the old x86 FPU even for non-packed applications.

well, that's only because intel implemented it that way. if they'd made the 'fxch' free on the p4 as in earlier processors it wouldn't really matter. besides, sse can't completely replace the fpu for many things.