mutually exclusive units
28 Oct 2010 6:24:53 UTC
Topic 195412
(moderation:
When my computer starts running a cuda program, it stops a regular Einstein program and vice versa. That is, it switches between a non-cuda and a cuda program. This does not seem efficient to me. SETI simply starts running one additional cuda program without stopping any other SETI program.
mutually exclusive units
)
Do you have Boinc set to not use the computer while it is 'in use'? Open the Boinc Manager, down by the clock, and go to Advanced, Preferences and then click on the processor usage tab and see if the box is checked that says "use GPU while computer is in use". If it is not checked that is why I think you seeing the switching you are seeing. Boinc processes GPU units on a first in first out basis only, so if you have a bunch of Seti gpu units in your cache and then get some Einstein gpu units, the Seti units will get processed before any Einstein gpu units get processed. This is ONLY for gpu units, the cpu units are handled differently. Also the current Einstein gpu units are mostly processed by the cpu, the next version of the Einstein gpu client will be more gpu than cpu. So for instance on a quad core you will not see 4 cpu units crunching and 1 gpu units crunching if you are only crunching for Einstein, like you do for the other projects. On Einstein you will see 3 cpu units and 1 gpu units crunching at the same time on a quad core.
Whereas a Seti CUDA task will
)
Whereas a Seti CUDA task will use the GPU 99% and 1 % CPU, an Einstein CUDA task will use about 20% GPU and 80% CPU still. That's what you see. The only thing run on the GPU in the Einstein ABP2 CUDA tasks is the FFT algorithm. The rest of the calculations are still done on the CPU. A next version of the ABP program will possibly change this.
In the mean time, read the ABP2 GPU thread. It has a lot more information.