Have you seen any reports of crunchers using it yet? I am not a gamer and the driver I have works, so am not an early jump on the bandwagon guy. Some of the last versions have actually made crunching slower!
I have not seen any reports as to how this driver might impact "crunchers". I too am not a gamer and believe in the "if it ain't broke then leave it alone" approach to driver upgrades. If there was a "crunch" improvement then I would certainly make the move.
I can confirm that the notebook version of 337.88 works without any problems (and no further runtime increase) with FGRP3-opencl.
My card is GT 755M on Wintendo 7 pro 64bit.
I haven't tried any other CUDA/OpenCL applications yet, though, but I assume it should work, too.
I have upgraded 2 machines - 660Ti and 750Ti. I always close BOINC before such upgrade, but this time after starting BOINC, it had problems finding the GPUs/CUDA devices. So I needed to perform a reboot. Now all seems to be OK.
I have upgraded 2 machines - 660Ti and 750Ti. I always close BOINC before such upgrade, but this time after starting BOINC, it had problems finding the GPUs/CUDA devices. So I needed to perform a reboot. Now all seems to be OK.
I too see that sometimes when I upgrade, I have gotten into just rebooting after the update is done. I do not update my drivers often, they just work.
The 337.88 driver has caused me problems on S6CasA (OpenCL) on my GTX 660 running under Win7 64-bit (it is OK on CUDA tasks). But I get groups of errors every two or three days. It is not just a single work unit, but all the ones in the buffer that go out after 0 seconds of run time. Then, the next batch that downloads will run OK. At first I thought it might be corrupted files in the BOINC data folder, so I removed the project and re-attached, but that did not fix it.
The solution was to go back to the 327.23 driver, which is known to do better on OpenCL tasks on at least some other projects (Folding, etc). And my run times decreased; I was getting 28 1/2 minutes on 337.88, but only 22 1/2 minutes on 327.23. These drivers do not support CUDA 6.0, but that is not yet needed on Einstein anyway.
Activated exception handling...
[15:38:46][4800][INFO ] Starting data processing...
[15:38:46][4800][INFO ] CUDA global memory status (initial GPU state, including context):
------> Used in total: 628 MB (1421 MB free / 2049 MB total) -> Used by this application (assuming a single GPU task): 0 MB
[15:38:46][4800][INFO ] Using CUDA device #0 "GeForce GTX 660" (0 CUDA cores / 0.00 GFLOPS)
[15:38:46][4800][INFO ] Version of installed CUDA driver: 6000
[15:38:46][4800][INFO ] Version of CUDA driver API used: 3020
[15:38:47][4800][INFO ] Checkpoint file unavailable: status.cpt (No such file or directory).
------> Starting from scratch...quote]
Where I am going with this is my GTX660 has newer and what I have been led to believe more powerful CUDAs than CUDA32, it seems to me that Einstein could get more production using them.
New NVIDIA driver available
)
Have you seen any reports of crunchers using it yet? I am not a gamer and the driver I have works, so am not an early jump on the bandwagon guy. Some of the last versions have actually made crunching slower!
I have not seen any reports
)
I have not seen any reports as to how this driver might impact "crunchers". I too am not a gamer and believe in the "if it ain't broke then leave it alone" approach to driver upgrades. If there was a "crunch" improvement then I would certainly make the move.
I can confirm that the
)
I can confirm that the notebook version of 337.88 works without any problems (and no further runtime increase) with FGRP3-opencl.
My card is GT 755M on Wintendo 7 pro 64bit.
I haven't tried any other CUDA/OpenCL applications yet, though, but I assume it should work, too.
I have upgraded 2 machines -
)
I have upgraded 2 machines - 660Ti and 750Ti. I always close BOINC before such upgrade, but this time after starting BOINC, it had problems finding the GPUs/CUDA devices. So I needed to perform a reboot. Now all seems to be OK.
-----
RE: I have upgraded 2
)
I too see that sometimes when I upgrade, I have gotten into just rebooting after the update is done. I do not update my drivers often, they just work.
The 337.88 driver has caused
)
The 337.88 driver has caused me problems on S6CasA (OpenCL) on my GTX 660 running under Win7 64-bit (it is OK on CUDA tasks). But I get groups of errors every two or three days. It is not just a single work unit, but all the ones in the buffer that go out after 0 seconds of run time. Then, the next batch that downloads will run OK. At first I thought it might be corrupted files in the BOINC data folder, so I removed the project and re-attached, but that did not fix it.
The solution was to go back to the 327.23 driver, which is known to do better on OpenCL tasks on at least some other projects (Folding, etc). And my run times decreased; I was getting 28 1/2 minutes on 337.88, but only 22 1/2 minutes on 327.23. These drivers do not support CUDA 6.0, but that is not yet needed on Einstein anyway.
Does anyone know which CUDAs
)
Does anyone know which CUDAs Einstein uses?
RE: Does anyone know which
)
http://einsteinathome.org/task/440532175
Where I am going with this is
)
Where I am going with this is my GTX660 has newer and what I have been led to believe more powerful CUDAs than CUDA32, it seems to me that Einstein could get more production using them.
They were working on a CUDA
)
They were working on a CUDA 5.5 version a while ago, but it disappeared into a black hole (as it were).