Einstein does CUDA

kolala · 27 Nov 2009 at 10:06

http://einstein.phys.uwm.edu/forum_thread.php?id=7683

We have finally begun to automatically deliver CUDA work & applications (plan class "ABP1cuda23") to machines that satisfy the following requirements:

- enabled NVIDIA GPU work in Einstein@home preferences
- NVidia GPU with at least 450MB of free memory
- Display Driver version 190.38 (&up), i.e. CUDA 2.3 capability
- BOINC Core Client version 6.10 (&up)

CUDA Beta App testers should drain their work cache and switch back to the normal project work.

BM

Oh well that should keep the GPU warm while SETI is borked and out of WU

kolala · 11 Dec 2009 at 01:13

Its improving... http://einstein.phys.uwm.edu/forum_thread.php?id=7683&nowrap=true#101108

Our tests indicate that the upcoming ABP2 GPU tasks typically take ~0.6 hours per WU. This is still "just" a factor of 2-3 faster than the ABP2 CPU version, but after the ABP2 release we are going to concentrate on improving the GPU code.

kolala · 13 Dec 2009 at 14:37

I've got version 2.3 dlls in the einstein folder but I don't remember putting them there. The files also have slightly different date attributes from the ones I put in the SETI folder so I think its likely that it gets the new versions by itself.

kolala · 20 Jan 2010 at 02:32

Its got another big bit better
http://einstein.phys.uwm.edu/forum_thread.php?id=7104&nowrap=true#101646

ABP2 will perform the same search on the same data as ABP1 currently does. But due to code optimizations and a smaller template bank it does this much faster; also ABP2 results don't validate against ABP1 ones, so we set up a new 'application'. We expect the ABP2 WUs to take ~1/5 of the time of ABP1 WUs.