It may not be worth doing...if the Beta works, and it WILL eventually, then they could be sending out gpu units non stop for a long time to come. Making the switching back and forth unnecessary as you would never switch back to cpu crunching with that core. MOST of us gpu crunchers crunch for OTHER gpu projects, and have been forced to do so until lately. Using an AMD gpu most projects code it to only use less than 0.05 percent of the cpu, meaning I can run all cpu cores and the gpu at the same time and not lose much. Now my Nvidia runs at 9?% of a cpu core while it is crunching. Even though they are crunching for different projects it is consistent across the projects I do crunch for. Time will tell how much of a cpu core WCG uses for both.

mikey159b...your experience from other GPU capable projects is because they were written in other languages. For example, PrimeGrid uses CAL not OpenCL. That is why it uses very little of the CPU core. (or at least from my understanding) So far, all OpenCL projects have seemed to utilize a much larger percentage of a CPU core. The trade off is that it is easier and cheaper for scientists to code in one language instead of two or three. The volunteers however lose a little efficiency on their end as far as resources go.

Yes not allowing sufficient CPU resources for GPU processing can be very inefficient. The resulting CPU overcommitment slows down both CPU applications and GPU applications. In the worst cases it can lead to the GPU being starved of CPU resources and almost stalling and this increases the chances of GPU tasks erroring or producing invalid results. As this current OpenCL application uses more CPU time than GPU time on fast ATI Cypress, Cayman or Tahiti class GPUs, allocating a core to support GPU processing is the more efficient choice.

My HD 5870 does these HCC tasks over 20 times faster than a core of my CPU. An overclocked HD 7870 may be over 60 times faster. So for me, it's more than a fair swap to use a CPU core to achieve this efficiency and increase the amount of scientific research that I can contribute.

I do understand the all cores for the cause concept that some adhere to though. However OpenCL GPU applications often use more CPU resources than applications coded in CAL, or with optimised IL coded cores. Better to accept it and move on from the myth of GPU processing not needing CPU resources. Avoiding CPU overcommitment with the use of a default ncpus value suitable for high CPU usage OpenCL GPU applications will allow greater efficiency for those contributing to GPU projects. It will also avoid the need for keen contributors to mess about with <file_name>, <open_name> mumbo jumbo.

Well these OpenCL GPU tasks have a fair amount of CPU time at the start and end. So if you allocate a CPU core to support these GPU tasks your computer will be more efficient overall as the GPU tasks will complete faster.

This is assuming that GPU work is consistently available. When GPU work is not available a CPU core would be idle. This should be handled automatically by BOINC for high CPU usage GPU applications but often isn't because variable performance potential of different contributors' hardware makes projects reluctant to do so. It's kind of a "can't please everyone" scenario.

I'm sure they will work either way but will probably work more efficiently if each GPU has a CPU core (HT or non HT) since when these WCG GPU tasks start and end they use the CPU exclusively.... hypothetically of course.