Folding Forum

Hello guys! Let's make the good list of overclocking for folding! I'm folding with 6850 card with clocks 800,1000 and trying 900,900 now because these settings caused after 10 or more hours that it got frozen. Anybody has good settings for it? And I'm interested in 6950 settings, too

I have stable settings for my other GPUs:
R9 270 - 1100,1500
7790 - 1200,1500
I think that R9 could be overclocked a bit higher, because it seems to be slower than 7790 but I don't want to go overboard...

What is the general idea? Clocks similar to scrypt mining or what algoritm?

If you've got Windows, then FAHBench is the best benchmark since it's actually based on the instruction mix contained in the FAHCore. [I wish they'd compile it for Linux.]

I've seen Windows more than 10 years ago for the last time. There's no problem to compile the application myself, you mean this one https://github.com/pandegroup/fahbench?
But... benchmarks usually only measure the computation power, does this one suggest best clocks settings, too?

It measures what FAH accomplishes. The fact that FAH performance depends almost exclusively on computational power may be indicative of something.

Normally, the FahCore does adjust memory clock down during computation (saving thermal headroom) and up when it's primarily moving data between RAM and VRAM does mean that Stanford has done some performance tradeoffs.

alicea wrote:I've seen Windows more than 10 years ago for the last time. There's no problem to compile the application myself, you mean this one https://github.com/pandegroup/fahbench?
But... benchmarks usually only measure the computation power, does this one suggest best clocks settings, too?

It does not make suggestions, but it will tell you if the current clocks are working or show an error if not. Raise the clocks until it fails. That will give you an idea of what the real client and work units will handle. Ymmv.

7im wrote:but it will tell you if the current clocks are working or show an error if not. Raise the clocks until it fails. That will give you an idea of what the real client and work units will handle. Ymmv.

Yeah, Im trying this way, but without errors shown, the GPU freezes after several hours for example yesterday evening I've set 750,1100 and it worked perfectly until today 9:30 morning. Now I've lowered to 700,1100 and is working without any problems for over 12hrs.

bruce wrote:Normally, the FahCore does adjust memory clock down during computation (saving thermal headroom) and up when it's primarily moving data between RAM and VRAM does mean that Stanford has done some performance tradeoffs.

And that's why there's not much sense to manipulate with memory, but rather with core, isn't it?