| Author |
Message |
|---|
|
|
|
My 460 is overclocked to 1759 on the shaders and has been running the cw sieves just fine. I had been running the gfn262's but can't remember if it my 460 had been over clocked at the time. Now I get maxerr results for all the gfn ports and the smaller wu's i get maxerr immediately. The longer ones take a while.
So, what's the deal? Is there a newer version of genefercuda i need to use. on 4.3.0 now I think running 32 bit or is it because of the oc on the 460 or something else i'm doing wrong.
the temps on the 460 are steady between 67-69 c so i don't think it's a heat issue. |
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1249
ID: 12001
Credit: 18,565,548
RAC: 0
 
|
My 460 is overclocked to 1759 on the shaders and has been running the cw sieves just fine. I had been running the gfn262's but can't remember if it my 460 had been over clocked at the time. Now I get maxerr results for all the gfn ports and the smaller wu's i get maxerr immediately. The longer ones take a while.
So, what's the deal? Is there a newer version of genefercuda i need to use. on 4.3.0 now I think running 32 bit or is it because of the oc on the 460 or something else i'm doing wrong.
the temps on the 460 are steady between 67-69 c so i don't think it's a heat issue.
It most likely means that you have reached the limit for b on that version of genefer. |
|
|
|
|
|
On all the ports? so what do I need to upgrade to get them running again or should I ask :) |
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1249
ID: 12001
Credit: 18,565,548
RAC: 0
|
On all the ports? so what do I need to upgrade to get them running again or should I ask :)
I don't know which version genefer is getting the roundoff error. You should have all four versions of genefer configured in your prpclient.ini file. The client will iterate through them (in order of fastest to slowest as the slowest supports the largest b values) in an attempt to find one that can do the PRP test. If all fail, then it will fallback and use PFGW (x86) or phrot (non-x86) for the PRP test. |
|
|
|
|
|
Sorry was not very clear. I am getting the error when using the genefercuda.exe - I only want to run it on my gpu so if I can't get that working I will have to run some other job and I only have a 32 bit machine. |
|
|
|
|
|
I am also getting the roundoff errors on my two GTX 460 cards when using the genefercuda app (version 4.3.0 beta).
I had no issues running PPSieve and CWSieve wu's on both cards and I am now running llrcuda wu's on on both cards... also with no issues.
I made no hardware changes to my system and other applications are running without errors... |
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1249
ID: 12001
Credit: 18,565,548
RAC: 0
|
|
If you only want to run CUDA apps, then GFN is not for you since the limits for b have been exceeded on all ports for genefercuda.
Since genefercuda and llrcuda use fundamentally different FFTs, they have completely different limitations. I suspect that llrcuda has far fewer limitations than genefercuda. |
|
|
John Honorary cruncher
Send message
Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0
                
|
If you only want to run CUDA apps, then GFN is not for you since the limits for b have been exceeded on all ports for genefercuda.
Only N=32768 and N=65536 have reached any of the b limits. From the PSA PRPNet Projects thread:
Top b limits are listed as GeneferCUDA, Genefx64, Genefer, Genefer80
N=32768
Top limits: 1.87M, 1.5M, 2.1M, 65M+
Currently at b=3,906,752
Available for Genefer80 and pfgw
N=65536
Top limits: 1.53M, 1.2M, 1.7M, 51M+
Currently at b=2,477,788
Available for Genefer80 and pfgw
N=262144
Top limits: 815K, 865K, 1,2M, 35M+
Currently at b=256,538
Available for GeneferCUDA, Genefx64, Genefer, Genefer80, and pfgw
N=524288
Top limits: 710K, 735K, 975K, 29M+
Currently at b=44,378
Available for GeneferCUDA, Genefx64, Genefer, Genefer80, and pfgw
____________
|
|
|
|
|
|
So my getting maxerr on the gfn262 port means i've got an issue with my 460 and maybe that's it's over clocked for the gfn even if it works for primegrid? |
|
|
Ken_g6Volunteer developer
Send message
Joined: 4 Jul 06
Posts: 929
ID: 3110
Credit: 237,014,205
RAC: 37,098
                         
|
|
CWPsieve (and PPSieve, but especially CWPsieve) CUDA are not meant for stability testing. Although false positives (incorrect factors) can't get through, false negatives (missed factors) can and probably do sometimes. I would expect a few WUs to error out, but there could be so few that you would dismiss it as noise; or you might not even have run enough WUs to have had one error out yet!
So, yes, I expect that your GPU is a little unstable.
____________
|
|
|
|
|
CWPsieve (and PPSieve, but especially CWPsieve) CUDA are not meant for stability testing. Although false positives (incorrect factors) can't get through, false negatives (missed factors) can and probably do sometimes. I would expect a few WUs to error out, but there could be so few that you would dismiss it as noise; or you might not even have run enough WUs to have had one error out yet!
So, yes, I expect that your GPU is a little unstable.
Thanks Ken I believe you might be correct. I dropped the oc down to 1707 last night and it's been running gfn262 all night. So there's a magic line somewhere on my 460. Although, even oc'd at 1759 I never had any invalid cw or pps.
Rick
____________
@AggieThePew
 |
|
|
|
|
|
Read this :)
http://www.primegrid.com/forum_thread.php?id=1215&nowrap=true#35849
Lennart |
|
|
Scott BrownVolunteer moderator
Project administrator
Volunteer tester
Project scientist
Send message
Joined: 17 Oct 05
Posts: 2329
ID: 1178
Credit: 15,637,281,698
RAC: 10,139,899
                                           
|
Read this :)
http://www.primegrid.com/forum_thread.php?id=1215&nowrap=true#35849
Lennart
A bit more notice of this move would have been nice...now my only GFN machine (at my office) will be idle all weekend... :(
____________
141941*2^4299438-1 is prime!
 |
|
|
|
|
Read this :)
http://www.primegrid.com/forum_thread.php?id=1215&nowrap=true#35849
Lennart
A bit more notice of this move would have been nice...now my only GFN machine (at my office) will be idle all weekend... :(
Sorry I have to do this move quick or I have to pay 500€ just for 2 days.
Lennart |
|
|
