Join PrimeGrid
Returning Participants
Community
Leader Boards
Results
Other
drummers-lowrise
|
Message boards :
Proth Prime Search :
Proth Prime Search (LLR) cuda?
| Author |
Message |
|---|
|
|
|
How do you get workunits for the Proth Prime Search (LLR) cuda, it won't send me any.
____________
  | |
|
 Michael Goetz Volunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 13513
ID: 53948
Credit: 237,712,514
RAC: 1,216
                           
|
How do you get workunits for the Proth Prime Search (LLR) cuda, it won't send me any.
You don't.
Although the selection is there on the website, llrCUDA isn't quite ready to go yet, so there's not actually any work available for it yet. Keep your fingers crossed and your CUDA-PPS box checked, and hopefully there will be work flowing sometime soon.
It might not be a terrible idea to put something on that page that mentions the current availability, or lack thereof.
____________
My lucky number is 75898524288+1 | |
|
|
|
|
Also a statement that it needs a Double Precision card or else it will not work. | |
|
rroonnaallddVolunteer developer
Volunteer tester
Send message
Joined: 3 Jul 09
Posts: 1213
ID: 42893
Credit: 34,634,263
RAC: 0
                
|
Also a statement that it needs a Double Precision card or else it will not work. DP isn't enough. You need also a 64bit linux/windows.
The errors on 32bit systems are not solved and i have no clue, when and if this will ever be done.
____________
Best wishes. Knowledge is power. by jjwhalen
 | |
|
JohnHonorary cruncher
Send message
Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0
                 
|
|
llrCUDA is in limbo currently. We've been unsuccessful in building a native BOINC version. Therefore, running it through the wrapper is the only other option. Unfortunately, the wrapper needs to be modified to accept llrCUDA, and we have not been successful in accomplishing that either. :(
____________
| |
|
|
|
|
primegrid_llr_wrapper (6.09, 6.10, 6.11, 6.12 and 6.12 avx) not working (don't attach thread) with llrcuda - WU get errors after 2-3 second. Work directory contains cudart.dll, cudart64_32_16.dll and cufft64_32_16.dll. Test cmd's llrcuda.exe -q"k*b^n+c" -d, llravx.exe -q"k*b^n+c" -d and cllr.exe -q"k*b^n+c" -d is working. Win2008 x64. | |
|
rroonnaallddVolunteer developer
Volunteer tester
Send message
Joined: 3 Jul 09
Posts: 1213
ID: 42893
Credit: 34,634,263
RAC: 0
|
primegrid_llr_wrapper (6.09, 6.10, 6.11, 6.12 and 6.12 avx) not working (don't attach thread) with llrcuda - WU get errors after 2-3 second. Work directory contains cudart.dll, cudart64_32_16.dll and cufft64_32_16.dll. Test cmd's llrcuda.exe -q"k*b^n+c" -d, llravx.exe -q"k*b^n+c" -d and cllr.exe -q"k*b^n+c" -d is working. Win2008 x64.
Please have a look in the result #338545769 or the running wu's #245153685 and #244142974.
It works but you cannot use 'llrCUDA' from PRPnet.
We have no native app and therefore also no public app...
____________
Best wishes. Knowledge is power. by jjwhalen
| |
|
JohnHonorary cruncher
Send message
Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0
|
llrCUDA is in limbo currently. We've been unsuccessful in building a native BOINC version. Therefore, running it through the wrapper is the only other option. Unfortunately, the wrapper needs to be modified to accept llrCUDA, and we have not been successful in accomplishing that either. :(
Another two months have passed and sadly no further developments. If anyone is interested in building a native BOINC llrCUDA application, we sure could use the help.
____________
| |
|
|
|
|
Hello,
Is there any progress in developing cuda app for PPS???
Regards,
Grzegorz Roman Granowski
____________
| |
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 13513
ID: 53948
Credit: 237,712,514
RAC: 1,216
|
Hello,
Is there any progress in developing cuda app for PPS???
Regards,
Grzegorz Roman Granowski
None that I'm aware of. I did have it running with a slightly modified BOINC wrapper under Windows. The only necessary change was to deal with selecting the correct GPU in computers with more than one GPU. That turned out to be far simpler than modifying llrCUDA to be a native BOINC app. That is to say, it ran without obvious error for a few tests. That's a long way from saying "It works reliably on all platforms," but I don't think any of the stability issues would be unsolvable.
The big problem, as far as I'm concerned, was that it just wasn't fast enough. On my hardware, llrCUDA was only about twice as fast as LLR. When you consider that I can run four LLR tasks on my quad core CPU, the CPU is twice as fast in total throughput as llrCUDA. My GPU isn't the fastest out there (GTX 460), but neither is my CPU (C2Q Q6600). Factor in newer CPUs with AVX and even a faster GPU isn't going to turn in impressive speeds compared to a CPU.
I'm not sure if anyone is actively working on llrCUDA.
____________
My lucky number is 75898524288+1 | |
|
|
|
|
OK, thanks for info, therefore it is better to have cuda application working, even if it isn;t fast enough, then not having it at all. I would gladly have my few GTX560 run for llr app. than wornikg on sieving constantely. So, if you have app working my you should consider executing it to broader public.
Regards,
Grzegorz Roman Granowski | |
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 13513
ID: 53948
Credit: 237,712,514
RAC: 1,216
|
OK, thanks for info, therefore it is better to have cuda application working, even if it isn;t fast enough, then not having it at all. I would gladly have my few GTX560 run for llr app. than wornikg on sieving constantely. So, if you have app working my you should consider executing it to broader public.
Regards,
Grzegorz Roman Granowski
That's not my call, and as I mentioned, it hasn't been tested extensively and there may be problems that need to be fixed. The performance (or lack thereof) of llrCUDA doesn't provide a lot of incentive to do the necessary testing and debugging at this point. Improving performance would be a much higher priority.
If you are looking to find some prime numbers with your GPU, you could try running GeneferCUDA either on PRPNet or BOINC.
____________
My lucky number is 75898524288+1 | |
|
|
|
|
Ok, thanks for info. | |
|
|
|
|
I've tested it back in January and it's absolutely useless on my system, if you'll pardon my harshness. The testrun was on PRPNet PPSELow and my main rig: intel i5 2500k @ 3.3GHz and GTX 570.
CPU needed ~37 seconds for one test (with AVX).
GPU needed ~140 seconds and a full core at it's disposal.
Haven't tried it on larger numbers though and I think I read somewhere that it would work better on larger numbers.
____________
PrimeGrid Challenge Overall standings --- Last update: From Pi to Paddy (2016)
 | |
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 13513
ID: 53948
Credit: 237,712,514
RAC: 1,216
|
Haven't tried it on larger numbers though and I think I read somewhere that it would work better on larger numbers.
I tried it on some large numbers (SoB), and it was more efficient, but only a little bit. It wasn't worth running the GPU for the performance gain it was providing.
____________
My lucky number is 75898524288+1 | |
|
Message boards :
Proth Prime Search :
Proth Prime Search (LLR) cuda? |
