GPU client

[Archangel999]

Ah, no, I asked what time it takes to run GPU-version of SETI client on your host?
You wrote

with the x64 6.01 app ----- 127 sec
with ak v8 SSSE3.1 -----  69 sec
with ak v8 SSE4.1 ----- 45 sec

Are these numbers GPU-app run times?
What GPU app version you used?

yes it is there 6.01 app i used - setiathome_6.01_windows_intelx64.exe 127 sec
AK_v8.0_Win64_SSE4.1 - 45 sec
AK_v8.0_Win64_SSSE3.1 - 69 sec
Test wu-1

[Archangel999]

if there are some beta's upcoming i'm 100% volunteer for testing 

Best Regards
D.Draganov

[gaulois952]

I'm volunteer too 

[Devaster]

[noise]now working on ap...[/noise]

[Archangel999]

any progress

[gaulois952]

Any progress ^^ ? 

[Raistmer]

In debugging.
If bug will be found, FFT will be done by  CUDA FFT library on GPU device.
(for AP app. MB GPU dev freezed for some time AFAIK)

[Archangel999]

nice will wait for test it

[Archangel999]

hey how's the progress

Best Regards
D.Draganov

[Raistmer]

Current  GPU app (FFT, highpass filter, dechirp) works correctly but at first glance gives no performance boost (at first sight because there was no CPU load info during test, only total timings).
One of aims of GPU app as I see it to free CPU for another work. So if it does task slower than CPU version but frees CPU much, it will be total benefit.

Now we trying to extend GPU processing on FFA function too (one of the longest routines in AstroPulse). The more code will be on GPU and the less data moves between GPU and CPU remains - the faster GPU app will be.
Maybe some big architecture change of app will be needed to better suit GPU hardware. Work in progress 

[Archangel999]

wow nice thanks man keep moving !!!

Best Regards
D.Draganov

[gaulois952]

The client will be compatible with ati cards ?

[Gecko_R7]

The client will be compatible with ati cards ?

No, sorry.  Not at first. Current efforts are centered @ Nvidia using Cuda SDKs.
It's as much a learning process as a development process.  GPU & CPU are different beasts & this is a new, uncharted frontier in Seti development.
Developers are having to completely rethink how to integrate the necessary tools and their understanding of Seti code into a structure that can exploit CPU & GPU strengths & dependencies.
An ATI effort may be in the cards in the future, but there's a long way to go before a Cuda app will be sophisticated and efficient enough to merit tackling another unique architecture and SDK. 

Then again...maybe someone will join the effort that can lead this project w/ the ability, tools and skill necessary. 
This is a welcoming group. 

[zangetsu]

Ah, no, I asked what time it takes to run GPU-version of SETI client on your host?
You wrote

with the x64 6.01 app ----- 127 sec
with ak v8 SSSE3.1 -----  69 sec
with ak v8 SSE4.1 ----- 45 sec

Are these numbers GPU-app run times?
What GPU app version you used?

yes it is there 6.01 app i used - setiathome_6.01_windows_intelx64.exe 127 sec
AK_v8.0_Win64_SSE4.1 - 45 sec
AK_v8.0_Win64_SSSE3.1 - 69 sec
Test wu-1

omg wow. 
45 seconds to run one work unit? amazing! am i assuming correctly that the gpu can only do 1 at a time, or do more, but slower? can boinc even get enough work units at that speed? . i have an ati card and i wish there would be an app for it, hell for any video card for that matter:)

[Slawek]

Hi


What req. GPU APP need ?   ( OS and CPU ?)
Winxp or Win Vista version 32 bit or 64 bit?
CPU AMD or INTEL ?