Folding GPU Tweaking?

[Endgame124]

Random google searches have been turning up conflicting information, so I figured I would start a thread.

I would like to tweak out my 1080Ti (and future 1660 Super) to be best set for folding. I've read on some forum posts that Memory amount and bandwith makes almost no difference, so it should be turned very low while the stream processors should be set as high as possible. However, I've also read that turning down memory can cut performance in half and that a large memory overclock can increase folding per by 10%+.

Is there a breakdown, guide, or similar for what tweaks to make on each individual video card to get the best points per watt? For example, maybe on the 1660 Super I would be best served to dramatically down clock the memory and overclock the GPU to lower the total watts sent to the less important memory. Perhaps this would increase my PPD from stock without increasing my total board power use?

If there are no guides / recommendations, is there a tool that I can use to test different configurations without having to mangle work units in the test process?

 

[biodoc]

You can use FAHbench at the default settings to test your GPU tweaks. Unfortunately, FAHBench uses core 21 and all the new covid-19 work uses core 22. A new version of FAHBench that uses core 22 is on their to do list.

You'll need to shut down all your CPU and GPU work to use FAHBench. Heavy CPU loads will affect FAHBench tests on your GPU significantly.

 

[Assimilator1]

I would think the benchmark reporting thread should give some clues as to what differences tweaks make, I'm rather surprised to see theirs no link to it in the FaHbench page 😕.
I'll see if I can dig it up......
[update] Well it seems, their isn't a more up to date thread than the old benchmarking thread I remember 🙁, not sure why atm, I have asked.

 

[Endgame124]

I did run across this site / post Where the author reduced the power limit on the card which increased efficiency (points / watt) down to 60% board power :

NVIDIA GEFORCE GTX 1080 Folding@Home Review (Part 2)

Welcome back. In the last article, I found that the GeForce GTX 1080 is an excellent graphics card for contributing to Stanford University’s charitable distributed computing project Folding@H…

greenfoldingathome.com

I think where this could have gone to the next level would be to record the GPU clockspeed and Ram clockspeed at 60%, then increase the board power but manually downclock mem and cpu to the lowered values. Then adjust them individually to see if efficiency can be increased by focusing on GPU performance or mem performance.