GPU crunching question

Seti@Home optimized science apps and information

Welcome, Guest. Please login or register.
Did you miss your activation email?
21 Nov 2008, 09:02:30 am

I've ran the test and got errors which look almost the same as citroja's...

The instruction at "0x696243a6" referenced memory at 0x0000045c". The memory could not be "read".

The errors and the addresses are absolutely the same for fft.exe and fft2.exe. System is: Dual Core Opteron 165 on nForce4 SLI motherboard, one Asus 6600GT 128 MB, running XP Pro SP2 and GPU drivers rev. 93.71.

I don't think memory is bad as it's been working flawlessly for the past two years and can pass 12+ hours of memtest86. Just tell me if you need more info about the system or if I can do more tests.

edit: stdout dump attached.

fft2d-xp_pro_sp2_6600gt.txt (0.57 KB - downloaded 9 times.)

fft-xp_pro_sp2_6600gt.txt (2.96 KB - downloaded 3 times.)

maybe is .NET 2.0 needed... but by rapidmind is writted nothing about it....

is true that on all machines i have tested it i have installed whole VS2005 Prof. with last SDKs and SP1 ....

Same for me, they all have VS 2003 and/or 2005 installed as well as appropriate SDKs and .NET.

May have something to do with missing runtimes, though I haven't seen .NET as required by RapidMind, either.

Time will tell

I'm waiting for the new ATIs to come out to upgrade my graphics card. RM should support those, too, in due time.

Regards,
Simon.

I have the .net 2.0 (and 1.1) framework installed, I guess that's not enough ;(

I will run some tests in a couple of days...right now I am working on the house.

I think that I may have VS2003 or 5 lying around somewhere and I can see if that changes things.

-citroja

I found out there is somehow an issue with the ms vc8 runtime... Its saying something in the system logs about it not being able to find it??? Roll Eyes

NVIDIA 6800GT / AGP 8X

RapidMind 2D FFT Benchmark
===============================================
Size: 256 x 256 = 2^8 x 2^8
Radix: 4 = 2^2
Total number of floating point operations: 5.24288e+006

Run timings, to and from host (in ms):

Average execution time: 10.5939ms
Overall average execution time: 10.5969ms
Minimum execution time: 9.46549ms
Average Mflops: 494.894
Peak Mflops: 553.894

Run timings, GPU-local (in ms):

Average execution time: 8.78433ms
Overall average execution time: 8.78745ms
Minimum execution time: 6.78877ms
Average Mflops: 596.845
Peak Mflops: 772.287

================================================

NVIDIA 7950GT / PCI-E 16x

RapidMind 2D FFT Benchmark
===============================================
Size: 256 x 256 = 2^8 x 2^8
Radix: 4 = 2^2
Total number of floating point operations: 5.24288e+006

Run timings, to and from host (in ms):

Average execution time: 9.1814ms
Overall average execution time: 9.18307ms
Minimum execution time: 8.70856ms
Average Mflops: 571.032
Peak Mflops: 602.037

Run timings, GPU-local (in ms):

Average execution time: 13.5061ms
Overall average execution time: 13.5087ms
Minimum execution time: 7.27945ms
Average Mflops: 388.186
Peak Mflops: 720.23

Nvidia GF 6600GT 586/1170, AGP 8X

RapidMind 2D FFT Benchmark
===============================================
Size: 256 x 256 = 2^8 x 2^8
Radix: 4 = 2^2
Total number of floating point operations: 5.24288e+006

Run timings, to and from host (in ms):

Average execution time: 12.1971ms
Overall average execution time: 12.2005ms
Minimum execution time: 11.4063ms
Average Mflops: 429.848
Peak Mflops: 459.649

Run timings, GPU-local (in ms):

Average execution time: 7.9736ms
Overall average execution time: 7.97726ms
Minimum execution time: 6.83655ms
Average Mflops: 657.53
Peak Mflops: 766.89

...and on an ASUS 7900GTX
I ran both tests a couple of times and the results vary quite a lot. The average Mflops and Peak Mflops vary from around 300ish-700ish.
Not sure how much I trust my mobo...the Asrock 939dual-SATA with a ULI chipset.

Got the memory error as well.

fft

min_n = 4
max_n = 4
RapidMind FFT Benchmark
-----------------------------------------------
Length: 16 = 2^4
Warming up...
Run timings, to and from host (in us):
13474.2 12450.4 12100.6 12176.3 12000.8
12249.2 12216.8 12083 12250.6 12171
12129.6 12050 11978.5 12079.4 13231.3
12345.6 12037.7 12215.4 12052.5 12014
12155.4 12135.5 12087.7 12014.5 12105.3
12027.1 12308.5 12053.4 11998.3 12088.6
12097.2 12031.3 12094.2 12043.9 12319.1
12197 12095 12147.8 12140.8 12043.9
12169.3 12099.5 12098.9 12119.9 12060.9
12024.6 12077.7 12170.2 12111.2 12076.6
12091.6 12133 12135.2 12209.6 12089.4
12103.9 12028.8 11999.2 12024.3 12043
12183.6 12231.1 12174.4 12210.7 12083.3
12010.9 12070.1 11913.1 11718.9 12044.7
12047.2 12182.5 12024.3 12123.2 12045.8
12149.8 12104.2 12116.5 12075.7 12157.3
12092.8 12118.2 12050 11907 11965.1
11988.5 12042.5 12148.4 12173.2 12149.8
11928.2 11948.9 12036.9 12133.8 12017.6
11158.1 12135.8 12360.2 11971.8 12029.1
Average execution time: 12113.8us
Normalized execution time (T/N): 757.113us/sample
Normalized by complexity (T/N lg N): 189.278
Mflops (5 N lg N/T): 0.0264161
Average execution time: 12113.8us
Minimum execution time: 11158.1us
Normalized average execution time (T/N): 757.113us/sample
Normalized minimum execution time (T/N): 697.384us/sample
Average time normalized by complexity (T/N lg N): 189.278
Minimum time normalized by complexity (T/N lg N): 174.346
Average Mflops (5 N lg N/T): 0.0264161
Peak Mflops (5 N lg N/T): 0.0286786
---
Warming up...
Run timings, GPU-local (in us):
11817.5 11669.2 11650.5 11724.5 11810.3
11599.3 11591.5 11692.9 11739.3 11620
11769.5 11737.4 11725.6 11675.9 11781.8
11786.8 11600.7 11614.7 11734.8 11786.5
11787.1 11696.6 11653.3 11707.2 11798.8
11590.1 11651 11680.6 11735.4 11728.7
11698.8 11841.6 11611.9 11674.5 11639.8
11797.1 11723.4 11594.9 11648.2 11911.1
11371.3 11533.7 11715.3 11645.7 11601
11682.3 11670.6 11712.2 11644 11650.5
11862.8 11576.1 11635.7 11578.9 11748.8
11704.1 11615.8 11635.4 11584.2 11649.6
11725.3 11581.7 11577.5 11596.3 11528.4
11659.1 11597.9 11703.5 11038.6 11949.4
11815.3 11751.6 11758.9 11650.5 11527.8
11628.1 11589.3 11617.8 11594 11813.1
11807.8 11637.9 11572.2 11654.7 11650.7
11688.7 11806.7 11739.9 11730.9 11735.1
11735.1 11711.9 11772.3 11613.6 11596.8
11770.3 11682.6 11734.3 11742.4 11727.6
Average execution time: 11679.3us
Minimum execution time: 11038.6us
Normalized average execution time (T/N): 729.958us/sample
Normalized minimum execution time (T/N): 689.91us/sample
Average time normalized by complexity (T/N lg N): 182.489
Minimum time normalized by complexity (T/N lg N): 172.477
BenchFFT average Mflops (5 N lg N/T): 0.0273988
BenchFFT peak Mflops (5 N lg N/T): 0.0289893
Residuals (compare with inverse):
Average absolute: 2.4984e-008
Maximum absolute: 1.19267e-007
Average relative: -1.#IND
Maximum relative: 1.#INF
-----------------------------------------------

fft2d

RapidMind 2D FFT Benchmark
===============================================
Size: 256 x 256 = 2^8 x 2^8
Radix: 4 = 2^2
Total number of floating point operations: 5.24288e+006

Run timings, to and from host (in ms):

Average execution time: 17.5728ms
Overall average execution time: 17.5754ms
Minimum execution time: 15.4032ms
Average Mflops: 298.352
Peak Mflops: 340.376

Run timings, GPU-local (in ms):

Average execution time: 9.77248ms
Overall average execution time: 9.77488ms
Minimum execution time: 7.6717ms
Average Mflops: 536.494
Peak Mflops: 683.406

nvidia CUDA is now free !!!!

i have seen a seti client development runnig on gpu with CUDA ....(see seti forums ...)

Is CUDA faster than RapidMind??

-citroja

Quote from: Devaster on 19 Feb 2007, 01:04:48 am

nvidia CUDA is now free !!!!

i have seen a seti client development runnig on gpu with CUDA ....(see seti forums ...)

Han's Dorn is? or was trying to develop a CUDA-based client @ Nov-Dec. There hasn't been any updates in over a month from him so not sure where this is at. The site/forum he posted at has been down for a couple of weeks now.

Quote from: Gecko_R7 on 19 Feb 2007, 12:51:35 pm

Quote from: Devaster on 19 Feb 2007, 01:04:48 am

nvidia CUDA is now free !!!!

i have seen a seti client development runnig on gpu with CUDA ....(see seti forums ...)

I haven't seen him in the S@H forums recently either...he coould be busy with work or away...

-citroja

Quote from: citroja on 19 Feb 2007, 08:51:56 am

Is CUDA faster than RapidMind??

-citroja

i dont know because CUDA is for G80 only ....
but i think yes, but i cannot test it i havent a G80

Quote from: Devaster on 19 Feb 2007, 03:27:34 pm

Quote from: citroja on 19 Feb 2007, 08:51:56 am

Is CUDA faster than RapidMind??

-citroja

i dont know because CUDA is for G80 only ....
but i think yes, but i cannot test it i havent a G80

too bad...i don't have on either. But I will get one shortly...I think that it would be more important to get it running for older models since there are more of them.

let me know if more testing needs to be done.

-citroja

Seti@Home optimized science apps and information \| Powered by Enigma 2.0 (RC1). © 2003-2008, LSP Dev Team. All Rights Reserved.	Seti@Home optimized science apps and information Forums \| Powered by SMF. © 2005, Simple Machines LLC. All Rights Reserved.