CUDA Factorial Benchmark from TOC.ru

**NaMcO** · 10-04-2009, 07:13 AM

Here's what i got,

**NapalmV5** · 10-04-2009, 08:04 AM

Originally Posted by HeUeR

nice Napalm nice

brrr, i cant wait till i will move on some quad core cpu, my cpu is so week, that i cant get normals score with my 8800GTX

thanks.. do you get any hang time after it reaches 99.6-9% ??

i took the 8800 ultra out and installed it for you to see what 8800 gtx does @ quad

i clocked my ultra to your gtx clocks

Originally Posted by HeUeR

250000 bench:

GPU: 27.903s

q9650 @ 4.5ghz/ddr2 @ 500mhz
this includes ~8 sec hang time.. actual time = 14 seconds

only a few times i get it right ^

cooled by arctic accelero/70F ambient also/cap+vmod ^

**HeUeR** · 10-04-2009, 08:36 AM

What do you mean with that hang time, didnt understand i think. You mean soft freeze for some sec, or ... !?

6 sec difference between same cards @ dual core vs quad core

Quad is doing what he needs to do

Il pm you, i want to check some more benches with setup: same card freq @ dual vs quad

**NapalmV5** · 10-04-2009, 08:52 AM

yeh freeze.. after it reaches ~99.9% it hangs/freezes for ~8 sec sometimes more

@ 1620 shader it reaches 99.9% in 14 seconds just like @ 2106 shader it reaches 99.9% in 11 seconds ^

if you dont get the freeze/hang time then its probably my 8800

lol more benches ?? one bench before i pull it out ok ?

either way get a quad asap.. everything will get faster i guarantee you

**HeUeR** · 10-04-2009, 09:21 AM

Napalm check private messages, i sent you pm with 2 more benches what i want to see and very big tnx you for spending time for me. And yea i know quad core will blast my vga much more higher

**NapalmV5** · 10-04-2009, 09:36 AM

finally got it.. my 8800s been thru a lot.. mods wars sickness

about pm/benches - alright you got it

**HeUeR** · 10-04-2009, 09:40 AM

how did ya get bench done without freez?

**NapalmV5** · 10-04-2009, 10:25 AM

i ran and reran the bench until it posted the right time i had the same troubles @ 2106 shader

i know whats wrong with my 8800 i didnt think it would affect the bench since it doesnt affect games/other benches

when ill get it fixed if results are any diff ill repost..

**NapalmV5** · 10-05-2009, 03:11 PM

subzero

**Talonman** · 10-05-2009, 08:11 PM

Originally Posted by Talonman

Were you guys also having your video RAM, not free up after each run in v0.3.6 too, or was that just me?

Bump!

**OverFoxtrot** · 10-07-2009, 02:24 AM

Hi All. I was return from Moscow

Yours result is super! And i have MultiGPU alpha version of this benchmark. Do you want try to test it?

**[XC] riptide** · 10-07-2009, 04:08 AM

Originally Posted by OverFoxtrot

Hi All. I was return from Moscow

Yours result is super! And i have MultiGPU alpha version of this benchmark. Do you want try to test it?

What a silly question. Of Course they want to test it!! This is Xtreme Systems.

**Talonman** · 10-07-2009, 04:11 AM

YES!!!

Post up a download link.

**HeUeR** · 10-07-2009, 04:38 AM

yup, give us the new version, i will make little bit popular this soft in my country oc forum

**Talonman** · 10-07-2009, 05:11 AM

Rest assured my friends at EVGA will also get the word!

nVidia already knows my stance...
http://forums.nvidia.com/index.php?showtopic=105567

"For future CUDA Superhero Challenges, I would rather not see a problem assigned.

I think the creative programs that might be submitted, will be more exciting for us nVidia fans to look at. They would also make for a better advertising video too.

This CUDA app is my current favorite: The CUDA Factorial Benchmark
http://www.xtremesystems.org/forums/...d.php?t=233280
(It is also ready to be downloaded and run as is.) We need more apps like that. I downloaded your .cu file, but have no idea how to run it.

The CUDA Factorial Benchmark lets us see how much faster our GPU's are, over our CPU. That is exactly the kind of app we need to help educate the non-believers in the GPU Revolution.

I want the version of the CUDA Factorial Benchmark that uses all 3 of my GPU's, instead of just 1/2 of my 295. I have been told they are working on it.

If anybody knows of a CUDA program that I can download, that will use both sides of my 295, and my 280 dedicated PhysX processor all at once to solve a problem, please post up a link!
I want that program real bad, whatever it is.

I had the incorrect idea that new apps were going to be submitted by CUDA programmers, that would be up for download, and that we could also run. That was bad on my part.
I assumed that was what the CUDA Superhero Challenge was all about."

Off topic: I can't post in the GPU section yet, so i have to get my posts in where I can!

It is a well known fact that PhysX can run on the CPU... Just much slower. Nvidia did optimized PhysX to run on the GPU for added performance, as expected. It is also well known the today's GPU's offer much more performance than the ageia PPU ever did.

We have all seen this video, of 8 cores making 3,500 boxes swirling in a tornado running on a CPU. This is nothing more than an honorable mention for what a top of the line CPU can do, but not anything that directly challenges the PhysX performance that we can get from a single GPU.
http://www.viddler.com/explore/HardOCP/videos/36/
That's OK, and providing you have 8 cores to calculate on. I wish they would have used a Q6600 or a dual core CPU for that test.

The real deal...
http://www.nvidia.com/object/gpu_tec...ml#livewebcast

In the video section under: Recording of the keynotes and general sessions...
'Opening keynote with Jen-Hsun Huang'
About 1/2 way through his presentation, when speaking on Visual Computing:

An example is posted about analyzing the New Orleans levee break.
# of Particles: 163,000
# of steps per particle: 1,000,000
CPU (Core 2, @ 2.5GHz): 24 Days to calculate...
(1) GTX 260: 4 Hours to calculate...
Total Speedup in performance = 144 X as fast.

Looks like running on the GPU was the correct choice for them.

In the 'Day 3 Keynote: Richard Kerris, CTO, Lucasfilm'
In the first 1/2 of the video, there is the Industrial Light & Magic, 2009 example.
Projuct GPU: Benefit from the use of GPU simulator, animation and rendering techniques for production, when they can provide a distinct advantage over the CPU based processors.

A research example is posted of a cube comprised of over 12,000 rigid bodies, calculated in 1 hour and 13 minutes using 1 GPU.

The next example is a colored cube comprised of over 100,000 rigid bodies, it took 3 hours and 20 minutes to calculate.

Keeping in mind that the most they were ever previously able to send, was 14,000 rigid bodies spanning multiple machines. The GPU is giving them much more performance than the CPU ever could, and they are enjoying a productivity increase, along with a substantial cost savings. It would be a tough sell to them, that the CPU is the correct choice now.

In the new FERMI presentation, it was specifically mentioned how PhysX will enjoy a speedup running on the new GPU's. Any concern about performance running PhysX on the GPU will be even smaller of an issue than it is now.

You can bet that when the Havok Physic engine is successfully ported over to OpenCL, and running on GPU's, the Havok boys will be looking for a performance gain. In fact, I dare say WAY more performance! If not, their effort was just in vein.
Most nVidia guys won't want Havok if we have to only run it on out CPU's. We already have been spoiled by PhysX. You can bet that the more demanding Havok effects will run much faster on the GPU.

I just love how GPU's are being put to more serious processing tasks than just playing games.

I do believe it to be the way of the future, and the next big thing.

The CUDA Factorial Benchmark helps to show this, and a big reason why it is one of my favorite CUDA apps.
Keep up the fine job!

**NapalmV5** · 10-07-2009, 06:23 AM

Originally Posted by Talonman

Bump!

the issue youre having @ post #38 is due to heat: throttle clocks kick in

more cooling or lower clocks and youll get constant results

as for running out of gpu ram.. i had the same @ 8800 ultra

no reboot i just had to restart the app..

edit: i just ran the bench on the 280.. cooled by arctic accelero/vmodded

i dont get the "Not enough memory on GPU device"

@ 1728 constant/correct results

though @ 1782 shader i get erratic results like you do get on your 280

**OverFoxtrot** · 10-07-2009, 09:14 AM

CUDA Factorial Benchmark 0.3.7 alpha
It is necessary to restart the program after each start. Wait for results

Thanks to all for support

P.S Somebody fix title of theme(wrong site name)?

**NapalmV5** · 10-07-2009, 09:29 AM

- erratic results fixed @ 280

- faster than 3.6

- higher shader clock

**Talonman** · 10-07-2009, 10:14 AM

Originally Posted by OverFoxtrot

CUDA Factorial Benchmark 0.3.7 alpha
It is necessary to restart the program after each start. Wait for results

Thanks to all for support

P.S Somebody fix title of theme(wrong site name)?

Thanks for the new version, and conformation that the app must be restarted after each run.

I get home in 5 more hours, and will test it out...

Love the ALL box.

**OverFoxtrot** · 10-07-2009, 10:45 AM

We have some results on our forum and now working at beta version.
If you want see real diffrence - use 750000! calculation and bigger. Now speed is small because some calculation make on CPU and need optimization.

**Talonman** · 10-07-2009, 04:42 PM

Odd...

If I use my normal 999,000, the program will not run for me with all GPU's selected.

If I use your suggested 750,000, the program runs, but the clocks drop back down from 3D Performance mode, to 2D clock speeds before the run completes...

If I use 999,000 or 750,000 and select a single GPU, the program runs just fine, and the clocks stay in 3D Performance mode for the entire run.

I don't know if this will help, but here is my event log:

Log Name: Application
Source: Windows Error Reporting
Date: 10/7/2009 9:08:12 PM
Event ID: 1001
Task Category: None
Level: Information
Keywords: Classic
User: N/A
Computer: MaximusSE
Description:
Fault bucket 1496996379, type 1
Event Name: APPCRASH
Response: None
Cab Id: 0

Problem signature:
P1: CUDA_Factorial_Benchmark.exe
P2: 1.0.3566.40857
P3: 4acbab93
P4: StackHash_f398
P5: 0.0.0.0
P6: 00000000
P7: c00000fd
P8: 75a19bb4
P9:
P10:

Attached files:
C:\Users\Scott\AppData\Local\Temp\WER6DCF.tmp.vers ion.txt

These files may be available here:
C:\Users\Scott\AppData\Local\Microsoft\Windows\WER \ReportArchive\Report0f16aabf
Event Xml:
<Event xmlns="http://schemas.microsoft.com/win/2004/08/events/event">
<System>
<Provider Name="Windows Error Reporting" />
<EventID Qualifiers="0">1001</EventID>
<Level>4</Level>
<Task>0</Task>
<Keywords>0x80000000000000</Keywords>
<TimeCreated SystemTime="2009-10-08T02:08:12.000Z" />
<EventRecordID>37229</EventRecordID>
<Channel>Application</Channel>
<Computer>MaximusSE</Computer>
<Security />
</System>
<EventData>
<Data>1496996379</Data>
<Data>1</Data>
<Data>APPCRASH</Data>
<Data>None</Data>
<Data>0</Data>
<Data>CUDA_Factorial_Benchmark.exe</Data>
<Data>1.0.3566.40857</Data>
<Data>4acbab93</Data>
<Data>StackHash_f398</Data>
<Data>0.0.0.0</Data>
<Data>00000000</Data>
<Data>c00000fd</Data>
<Data>75a19bb4</Data>
<Data>
</Data>
<Data>
</Data>
<Data>
C:\Users\Scott\AppData\Local\Temp\WER6DCF.tmp.vers ion.txt</Data>
<Data>C:\Users\Scott\AppData\Local\Microsoft\Windo ws\WER\ReportArchive\Report0f16aabf</Data>
</EventData>
</Event>

One more error message from another run:

Faulting application CUDA_Factorial_Benchmark.exe, version 1.0.3566.40857, time stamp 0x4acbab93, faulting module unknown, version 0.0.0.0, time stamp 0x00000000, exception code 0xc0000005, fault offset 0x00000000, process id 0x72c, application start time 0x01ca47c4811d9199.

**OverFoxtrot** · 10-07-2009, 11:21 PM

Talonman Thanks.
Clocks drop back down from 3D Performance mode to 2D clock speeds because CPU make calculation in the end. We will fix this in beta version.
The main thing that the program loads all videocards in system.