Originally Posted by Carniflex
Mine is running at kinda low GPU utilization. Have 6770 @ 950 core / 1375 memory (its under water) and GPU utilization is around 20 .. up to 30% occasionally. I have 1 Gb of vRAM - I saw earlier in this thread a note that there are some RAM requirements for the GFX card so ... I think I would do ok running 2 of these for a bit better GPU utilization ? But 3 would already go above requirement? What was the vRAM requirement anyway I dont think I have seen it mentioned explicitly.
Another interesting note is that after OC'ing the CPU higher the GPU threads use considerably less CPU. With 1055T at aprox 3.5 GHz the CPU utilization was around ~10% but now at 4060MHz it's hovering around 3% of CPU utilization. I pushet the HT link also up to 2900 when increasing the OC though.
And how do I get multiple GPU WU's to run on the same gfx card ? It's somewhere in the options ?
Okay, first of all, a 1055T running over 3.5GHz is possibly a problem. Hell, even under water, running any Thuban CPU can be a problem past 15% overclock while crunching BOINC tasks. It may only glitch once per week, but that is more than enough... I also have a 1055T and it hates running past 3.5GHz without really high voltages that really heat things up and waste a lot of power for a lousy few percent overclock.
Having said that, the ATI/AMD version of Einstein uses OpenCL and anything OpenCL is slow and less efficient. In fact, Collatz is now running OpenCL and the times have doubled when compared to the CUDA version. Also, these apps use a ton more CPU power than more native CUDA/CAL applications. Soo, your 20-30% is about in line for a 6770 with 800 stream processors. POEM can also be low in the utilization department on ATI/AMD cards... It all depends on the math they use and how it's used...
To run more than ONE task at a time, simply go to the Einstein project preferences and select 0.5 GPU utilization. If you have more than one system, you might assign this to WORK or HOME instead of default to test. I run 2 WU's at the same time on the nvidia cards always.
In any case, many GPU projects favor one type or the another just because of the processing they do... it's normal. It won't happen instantly if you have a lot of WU's cached... but should start running muliple WU's on the next download...