| Author |
Message |
|---|
|
|
|
Installed the latest nvidia driver for a zotac and got an invalid cuda file error. Rolled my driver back to what I had installed but was just wondering if there is a new tpsieve-cuda-x86-windows.exe and cudart.dll file. These are both from around 7/2014
thanks |
|
|
JimB Honorary cruncher
Send message
Joined: 4 Aug 11
Posts: 920
ID: 107307
Credit: 989,573,254
RAC: 25,159
                     
|
|
No, there are no newer files. I've never had a problem with them. |
|
|
 RafaelVolunteer tester
Send message
Joined: 22 Oct 14
Posts: 918
ID: 370496
Credit: 606,444,977
RAC: 610,352
                    
|
|
Same, ever since topic started, I'm using the same files, never a problem. Can you try updating drivers again, with a clean install? |
|
|
|
|
|
I get the following after installing the latest driver
cuda error: cudaMemcpy: invalid argument
and here is my command line
tpsieve-cuda-x86-windows.exe -p6644T -P6646T -k3 -K9999 -n9M -N12M -ffppr12M_6644T-6646T.txt -M2 -d0 -t2 -q
so, I'm figuring if you are not having issues then I have something wrong.. am going to revert back to known working driver for now
thanks
|
|
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 14037
ID: 53948
Credit: 477,278,075
RAC: 296,248
                   
|
|
I've moved a series of posts here from an informational thread that should have been locked. This problem deserves its own thread anyway.
____________
My lucky number is 75898524288+1 |
|
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 14037
ID: 53948
Credit: 477,278,075
RAC: 296,248
|
|
Guys, with the new (possibly problematic) drivers, do you also have trouble running the BOINC PPS-Sieve CUDA app?
____________
My lucky number is 75898524288+1 |
|
|
RafaelVolunteer tester
Send message
Joined: 22 Oct 14
Posts: 918
ID: 370496
Credit: 606,444,977
RAC: 610,352
|
Guys, with the new (possibly problematic) drivers, do you also have trouble running the BOINC PPS-Sieve CUDA app?
Remind me to chekc it at the end of the week, once I get home.
In the meantime, I've recently did a Genefer WR CUDA, that completed 3 days ago. As well as manual sieve for GFN N=65536 and N=2097152. Neither of the 3 have had a problem with the latest drivers on my Gtx 970. Will check on PPS-Sieve Cuda and the recently released GFN Short n=21 Cuda when I can. Should I bother with OCL Cuda as well? |
|
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 14037
ID: 53948
Credit: 477,278,075
RAC: 296,248
|
In the meantime, I've recently did a Genefer WR CUDA, that completed 3 days ago. As well as manual sieve for GFN N=65536 and N=2097152. Neither of the 3 have had a problem with the latest drivers on my Gtx 970. Will check on PPS-Sieve Cuda and the recently released GFN Short n=21 Cuda when I can. Should I bother with OCL Cuda as well?
No need to check CUDA on GFN-21 since it's identical to GFN-WR. IF WR works, so will 21.
OCL CUDA isn't actually CUDA; it's an OCL app. It's only called CUDA because that has to be in the plan class name to get the BOINC client to behave properly. However, the OCL driver is part of the CUDA driver, so it's actually not a bad idea to check to see if it works.
____________
My lucky number is 75898524288+1 |
|
|
|
|
|
I tested wss and it ran fine but dont know if its the same. Will test the pps sieve. |
|
|
RafaelVolunteer tester
Send message
Joined: 22 Oct 14
Posts: 918
ID: 370496
Credit: 606,444,977
RAC: 610,352
|
In the meantime, I've recently did a Genefer WR CUDA, that completed 3 days ago. As well as manual sieve for GFN N=65536 and N=2097152. Neither of the 3 have had a problem with the latest drivers on my Gtx 970. Will check on PPS-Sieve Cuda and the recently released GFN Short n=21 Cuda when I can. Should I bother with OCL Cuda as well?
No need to check CUDA on GFN-21 since it's identical to GFN-WR. IF WR works, so will 21.
OCL CUDA isn't actually CUDA; it's an OCL app. It's only called CUDA because that has to be in the plan class name to get the BOINC client to behave properly. However, the OCL driver is part of the CUDA driver, so it's actually not a bad idea to check to see if it works.
Yes yes, I know it isn't cuda, I just wanted to use the same name as given by Boinc.
Speaking of which... wouldn't the OCL app be the same as the OCL 1/2/3 being tested? If so, there's already the few results from our beta tests at the OCL 2 thread. Although those are a lot less extensive than actual WUs being run, I guess it's already something...
That only leaves PPR12M / Factorial Manual sieve and PPS Sieve as the apps left for testing. |
|
|
|
|
|
installed the 355.82 windows 7 64 bit driver. I only installed the driver no 3d or audio or physic x. the pps sieve ran fine... in fact it ran really well.
so the only issue I've now run into is the manual sieve.
rolling back the driver so I can finish my sieve for now.
http://www.primegrid.com/result.php?resultid=648537043 |
|
|
|
|
|
So I thought I'd run a small test on the old driver vs the new driver for pps sieve and the results were quite interesting.
the old driver ran the pps sieves between 652 and 655 seconds and the new driver (355.82) runs them between 619 and 621 seconds which to me is quite a difference. |
|
|
|
|
|
So, I am the only one having issues running PPR12M with the 355.82 driver installed? Any ideas on what I am doing wrong. It's pretty much something I have done as almost everything else is working as noted by others. |
|
|
JimBHonorary cruncher
Send message
Joined: 4 Aug 11
Posts: 920
ID: 107307
Credit: 989,573,254
RAC: 25,159
|
|
Yes, you are. I just downloaded and installed the 355.82 driver and my computer is happily running PPR12M sieving. |
|
|
|
|
|
I found the issue.. here is the command line that runs
tpsieve-cuda-x86-windows.exe -p6673T -P6674T -k3 -K9999 -n9M -N12M -ffppr12M_6673T-6674T.txt -M2 -t2 -q
and here is the command line that failed
tpsieve-cuda-x86-windows.exe -p6673T -P6674T -k3 -K9999 -n9M -N12M -ffppr12M_6673T-6674T.txt -M2 -t2 -q -d0
I'd been told to add the -d0 to force it to use a gpu. I also noticed the -t2 doesn't seem to work now either as it used to tell me I was using 2 threads and now it just uses 1.
What does the -M2 switch do btw
and thanks Jim for updating the thread. I knew I was doing something wrong just had to spend a little time trying it out.
Cheers
edit: ps, sorry for the cancelled P range |
|
|
TylerProject administrator
Volunteer tester
Send message
Joined: 4 Dec 12
Posts: 1081
ID: 183129
Credit: 1,384,625,026
RAC: 7,097
          
|
What does the -M2 switch do btw
I believe the -M2 switch is so that it combines -1 and +1 forms into the same sieve.
____________
275*2^3585539+1 is prime!!! (1079358 digits)
 Proud member of Aggie the Pew
 |
|
|
JimBHonorary cruncher
Send message
Joined: 4 Aug 11
Posts: 920
ID: 107307
Credit: 989,573,254
RAC: 25,159
|
|
-t2 would try to use two different GPUs simultaneously IIRC. We've found that one or both of the GPUs skips factors when run like that, so we ask that you not do it. One reason that I have stats charts for each kind of sieving is so that I can spot any reservation where the factor density is unusually low.
tpsieve-cuda will always want to run on a GPU, never on a CPU. -d is the device number for if you have more than one GPU.
Yes, -M2 is the argument that specifies to sieve both +1 (Sierpinski) and -1 (Riesel) forms. It takes much the same time as sieving only +1 and we make our Riesel sieves available to the Riesel Prime Search users at mersenneforum.
Edit: And one of your recent ranges is suspiciously low, Rick. 6644T-6646T looks wrong. When one of my GPUs is free on Saturday I'll sieve a bit of that myself. If I find even one factor that you didn't, then it'll have to be redone. I say one factor because I don't intend to resieve the entire thing myself. Any two runs should be identical once sorted. I'll let you know privately.
http://www.primegrid.com/sieving/ppr12m/ |
|
|
|
|
-t2 would try to use two different GPUs simultaneously IIRC. We've found that one or both of the GPUs skips factors when run like that, so we ask that you not do it. One reason that I have stats charts for each kind of sieving is so that I can spot any reservation where the factor density is unusually low.
tpsieve-cuda will always want to run on a GPU, never on a CPU. -d is the device number for if you have more than one GPU.
Yes, -M2 is the argument that specifies to sieve both +1 (Sierpinski) and -1 (Riesel) forms. It takes much the same time as sieving only +1 and we make our Riesel sieves available to the Riesel Prime Search users at mersenneforum.
Edit: And one of your recent ranges is suspiciously low, Rick. 6644T-6646T looks wrong. When one of my GPUs is free on Saturday I'll sieve a bit of that myself. If I find even one factor that you didn't, then it'll have to be redone. I say one factor because I don't intend to resieve the entire thing myself. Any two runs should be identical once sorted. I'll let you know privately.
http://www.primegrid.com/sieving/ppr12m/
I dont even remember which system I ran it on. If I need to I will be happy to re-run it. It's possible it was on my one system that has the new gpu. So, -t2 isn't how many threads to use? I've used it for almost the whole time and it does increase thru put or so I thought but will see what happens without it on my next run.
the -d0 switch definitely will NOT run on the 355.82 driver on my 970 system so I just leave it out, get a message that gpu 1 is not valid and it starts running.
And my next question is, if -t? should not be used, how do we tell it to use more than 1 thread?
Cheers |
|
|
|
|
-t2 would try to use two different GPUs simultaneously IIRC. We've found that one or both of the GPUs skips factors when run like that, so we ask that you not do it. One reason that I have stats charts for each kind of sieving is so that I can spot any reservation where the factor density is unusually low.
That's super disappointing to hear, up until now I have exclusively run tpsieve with -t2.
____________
 |
|
|
TylerProject administrator
Volunteer tester
Send message
Joined: 4 Dec 12
Posts: 1081
ID: 183129
Credit: 1,384,625,026
RAC: 7,097
|
the -d0 switch definitely will NOT run on the 355.82 driver on my 970 system so I just leave it out, get a message that gpu 1 is not valid and it starts running.
I'm now getting the "Cuda error: cudaMemcpy: invalid argument" error when running tpsieve CUDA on my GTX 760 with the 359.06 driver.. Stumbled upon this thread again and removed the -d0 switch and it works again.. Interesting.
____________
275*2^3585539+1 is prime!!! (1079358 digits)
Proud member of Aggie the Pew
|
|
|
