It is zero meaning for the GPU project per se, except now that it's running about 7-8 times faster [really the science processes about 18 times faster on the GPU side], the door has opened to add way more work from others in the same field of research. As was the case with CFSW [Sustainable water], the question may even have been asked if there was scientific value to re-running with greater resolution, depth of investigation. Guess not.

Even higher over here, the GPU+CPU work units process ~100 times faster than if I was doing CPU only with the same work units. There is much merit to GPGPU computing. Although I wonder if it could be programmed in a better way, much like POEM developers programmed their work units, which seem to be much more efficient despite using less GPU resources. That is that some instructions that the GPU can execute actually execute significantly faster on the CPU, and other instructions execute significantly faster on the GPU, so they programmed their app with the best of both worlds in mind. The downside is that it is very memory intense, but the upside is that it's significantly faster than if they offloaded almost everything to the GPU.

The transfer rate is entirely dependent on the speed of the processing power of your GPU and CPU. The faster your GPU, the more intense the memory requirement is to utilize it. With 1800MHz CL9 dual channel DDR3, I can run the GPU at 500MHz core instead of 1200MHz core, and only have 82% GPU usage. Therefore, there is a lot of potential left untapped, but even this speed is enough to process an insane degree of computations. 700,000 BOINC PPD at the above rate. I imagine if you had triple the memory transfer rate (Quad Channel 2700MHz CL9 or Triple Channel 3600MHz CL9, or something insane like 5400MHz Dual Channel CL9), you could get up to 2 million PPD off a single 7950. If you have more than one GPU, you are SOL.

I notice that your chart went from NA to 112 days for HCC1.
At the same time something changed because it takes twice the time to run CPU and GPU+CPU work units.
Did they double the size to 4 images per work unit?

Now that a lot of people have been doing HCC GPU for a while, I am wondering if it would be useful to start a thread for people to post GPU cards + MEM, CPU etc along with the % they allocate per WU and the approximate time per result they achieve.

I, for one, am starting to introduce more GPU cards to the mix and would appreciate being able to refer to a guide of other people's results to set up ratios to get the best bang for my buck and save me the time of 'tweaking' things just for myself.

Does this seem like a reasonable suggestion?