Couple noob questions

[Shill]

Hello,
Few questions to the community, moderators please feel free to comment.

First, I am curious any one know how to limit graphical processing unit (GPU) usage while using the Folding@Home (FAH) Windows client? Example: the video card is either off or on at regardless of preset level but at least with the central processing unit (CPU) it will at least adjust the usage between the light, medium, and full presets. I would like to be able to drop both CPU and GPU usage down when I am just on the computer casually doing my thing (ie light and medium presets)

Second, is there a way to stop the FAH client from accepting new WU after it completes one? I have a GTX 660 Ti and so far the few GPU WUs I have ran are usually completed in about 6-7 hours. Which is perfect when I am at leaving for work that is my one a day but when I come home it would be nice to see computer finished its FAH WU and possible have entered into a sleep or shutdown state. But more so I do not want to accept something I can not complete ie going away on vacation.

Thirdly, regarding FAH performance, specifically GPU performance. What should we be looking for exactly to optimize our capabilities? Does the FAH client use single and or double precision floating point? Does the GPUs ability to support newer APIs like OpenGL 4.4 help over older generation GPU'. I am consider getting the newer AMD R9 series since they seem to be better for compute purposes (and cheaper in general to Nvidia) but seek a community census confirmation if you will I am on the right track.

Finally, please take into account I am post under 'New Donor start here' so be gentle and kind. Help me... help the FAH community.
Thank you all in advanced. -SH

[Joe_H]

A couple methods have been tried in the past to throttle GPU usage, they did not work well. So the current method of defaulting to only doing GPU folding when the system is "Idle" was chosen. I don't know if work is still being done to see if a method of throttling the GPU client can be programmed.

Using the FInish function in FAHControl will finish work on the current WU, turn it in and then leave the folding client idle. More information on configuring the client is available from The Software link at the top of forum pages. There are links on that page to installation guides, an introduction to the client and configuration. One useful table on the Intermediate V7 FAQ covers what the various folding levels do in detail - http://folding.stanford.edu/home/faq/fa ... ate/#ntoc5.

The folding cores in current use do single precision calculations. Some older cores tried out double precision calculations, but the additional overhead was found not to be needed. As for OpenGL, that is not used by folding cores, but may be used in the display portions such as FAHViewer. Two GPU folding cores use OpenCL, not to be confused with OpenGL which is an entirely different set of commands. Core_16 for AMD-ATI GPU's is about to be retired, and Core_17 recently came out of beta and is used at the advanced client-type setting on nVidia and AMD-ATI.

[7im]

Hello Shill, welcome to the folding forum.

I can answer a few quick questions. By design, CPUs load share well, GPUs do not. At full power, both CPU and GPU will fold all the time. On Medium, the CPU will fold full time, and the GPU will fold only when the system is idle (fah screen save kicks in, or monitor shuts off in power saver mode).

Right click the slot and select the Finish option. WU will finish, upload, not download new work.

FAH is single precision only. Newer OpenGL does not help. FAH does not use it. CUDA, OpenCL are the 2 fah uses.

[Shill]

Thank you both, Joe_H and 7im.

[PantherX]

...I don't know if work is still being done to see if a method of throttling the GPU client can be programmed...

To elaborate a little, this is a limitation from the hardware vendors. Thus, F@H doesn't have any control over if and when it is implemented. Hopefully, in the future, the hardware and software will support a GPU priority system similar to the CPU one which would be pretty useful to a lot of donors.

[bruce]

To elaborate a little, this is a limitation from the hardware vendors. Thus, F@H doesn't have any control over if and when it is implemented. Hopefully, in the future, the hardware and software will support a GPU priority system similar to the CPU one which would be pretty useful to a lot of donors.

To elaborate a little more, there has been talk about drivers which allow a GPU to allocate fewer shaders than "all" to a given GPU program but so far it's just talk -- nothing outside of their labs.

To reduce the power/heat/throughput you could always underclock but if you're also using the same GPU to refresh your desktop or decode your movie, that isn't helpful.

For CPUs, your OS has had many years of development of a scheduling system that allows a high priority task to interrupt a lower priority task (and FAH uses a very low priority) but no such concepts exist in today's GPU drivers. GPU work is FirstComeFirstServed and once a block of work is started, work requested by other software waits until the first block is finished. Pause when idle prevents FAH from getting in line ahead of a screen refresh by stopping FAH until you're not using your screen.

[Shill]

Thank you additionally PantherX and bruce for the extra information.