Folding Forum

I noticed when receiving a WU for project 13850 last night that the time from assignment to timeout is one day, whereas CPU projects within a similar amount of base credit would have a period around five days (e.g. 13807 et seq.). I can get the WU done fast enough (the points are irrelevant), but I thought I would inquire to make sure this is intentional for COVID-19 projects. For example, I might be a little more worried about getting a WU from 13834 (1 day period, 2750 points) done before it times out.

There are two time limits, the shorter one (timeout) is only the limit for bonus points. Work will still be returned up to the second limit (expiration).
The timeout has been reduced to reflect the faster average processor and the urgency of the work.

Makes sense. I do want to try to beat the timeout so the WU doesn't get needlessly handed out to someone else (I only run with one thread on my laptop so the fan doesn't run too loudly). Thanks!

Yes, the timeout can be challenging for a single-thread slot. IIRC the aim is that a full-time single thread on a typical PC will make the timeout, but older/slower PCs may struggle.

The timeouts are too short for the amount of work. My computer is currently busy with a WU for Project 14412. The timeout is only 24 hours later, while it takes 6 hours to finish. I don't want to run folding@home 24/7, only when I already use my computer, so with such a short deadline and so much work I can't guarantee to be able to finish the work in time.

I think I will just stop running folding@home altogether if there is no solution. I already removed my GPU slot because it couldn't keep up with the timeouts at all, but that's to be expected because I don't have a good GPU. My CPU though is a xeon e5-4667 v3 (16 cores at 2,97 GHz). Is this normal? I would expect a 16-core CPU from 2014 to be able to keep up given that most people still have 4 to 8 cores.

I see pretty high kernel times of around 20 - 30% while running folding@home. I normally use my CPU for other compute tasks (mostly fractal rendering) and kernel time is usually close to 0%. I would expect folding@home to be similar in the sense that it's also a number crunching process.

Dinkydau wrote:...I don't want to run folding@home 24/7, only when I already use my computer, so with such a short deadline and so much work I can't guarantee to be able to finish the work in time... My CPU though is a xeon e5-4667 v3 (16 cores at 2,97 GHz). Is this normal? I would expect a 16-core CPU from 2014 to be able to keep up given that most people still have 4 to 8 cores.

I see pretty high kernel times of around 20 - 30% while running folding@home. I normally use my CPU for other compute tasks (mostly fractal rendering) and kernel time is usually close to 0%. I would expect folding@home to be similar in the sense that it's also a number crunching process.

It could be possible that you're running additional CPU intensive tasks that is causing significant slow down? for CPU folding, if you assign 16 CPUs to the CPU Slot and you're consistently using 2 CPUs (Chrome, OS, etc.), it will drastically impact the performance since F@H will need to wait for all 16 threads to processing data before processing the next set of data (that's the percentage change from 1% to 2%).

I would suggest that you see what your daily baseline CPU usage is and then adjust your CPU slot accordingly.

Okay, that's good to know. Yes, I'm consistently using the CPU for other things such as watching videos on youtube. Because of hyperthreading the CPU actually has 32 threads. I set folding@home to use 28. Let's see if it helps.

The other thing to think about it if you allocate 28 CPUs, the work allocated to your CPU would be catered towards the beefier hardware where the expectation is that it runs 24/7 as opposed to a 8 CPU system which is a typical house-hold system.

If 28 CPUs don't work, try 8 CPUs or 12 CPUs and see if that works well in your situation.

It helps a lot. At this pace the whole WU could have finished in under 2 hours. I'll try lowering the thread count a little more to find the optimum.