folding@home number of clients vs numbers of computers/cores/threads?

If you are looking inside F@H, every slot is a client.

If you computer changes IP address, those are new clients.

So if you have two PCs running MacOS you have 2 clients unless they got new IP addresses. MacOS does not support GPUs.

If you add a Win10 box with a i5, you add a client.

If it has an Nvidia Card, that will be a another client.


Every entity that is turning in results is a client.

I Mac Mini
2 Mac Mini
3 Win10 i5
4 Nvidia GPU

Four clients until an IP address changes.

I have a client (Core 2 Duo T6600 at 2.2 Ghz) that produces less than 1000 Points Per Day. I a have client (GTX 1060 6gb) that produces over 400,000 Points Per Day, they are both one client.

F@H claims I have 17 clients within 7 days and 23 clients within 50 days. I have 11 clients, but IP numbers have changed over time.

If I win the lottery and buy a a 256 thread Epyc PC running Linux, it is still one client. (In Windows, I would need a slot for every 32 threads so I would have more clients but less Points Per Day)

https://www.youtube.com/watch?v=KjsSvjA5TuE

DrBoar said:

I was about to get a GTX 1650 DDR6 for my Dell computer (as a modest gaming rig) But now it is on hold as I have to see how the new 1650 variants pan out.

Click to expand...

My GTX 1650 is a Low Profile card and that usually means less performance, but mine is getting 300,000 PPD 24/7

DrBoar said:

I did not know that the GPU score so much higher than CPUs. Too bad my old ATI 4770 can not do anything useful.

Click to expand...

An AMD Epyc with 256 threads should do as well as any Graphics Card, it is just that most CPUs do not have as many threads as a Graphics Card.

https://www.techpowerup.com/gpu-specs/r ... -4770.c290

Your ATI does not support OpenCL 1.2 which is a requirement, if it did, it might not meet deadlines. It is about 3/5 as fast as a GT 1030. It was new in 2009.

But now it is on hold as I have to see how the new 1650 variants pan out.

Click to expand...

What exact difference are you looking for? General performance or specific to F@H?

https://www.tomshardware.com/news/nvidi ... 0-variants

I suspect as far as F@H goes they'll be exactly the same or so close it won't matter, unless there's something truly unexpected inside TU106 that gets enabled. OTOH if you have time to wait, it doesn't hurt to wait to see benchmarks...

We have not heard a rumor of a new Ampere card under 180 watts (RTX 3060 ti)

The GTX 1650 may remain the most powerful card available under 75 watts.

The good news is that it became 20% more productive yesterday with the release of a CUDA F@H core_22 0.0.13.

It looks like 360k PPD.

Older Nvidia cards may increase less.

https://foldingathome.org/2020/09/28/foldingathome-gets-cuda-support/

I was wondering why my 1070 Ti suddenly had a ~40% jump in points-per-day lately; this switch to CUDA from OpenCL explains it, especially with the gains that the "moonshot" work units seem to be getting!

Looking at the chart, it looks like Pascal cards are seeing some of the most benefit from this; I wonder if the OpenCL scheduler isn't taking full advantage of the sort-of-asynchronous compute capabilities that they offer, but CUDA does? It's also a bit interesting to see that the 980 Ti and 750 Ti see substantial performance improvements, but the other Maxwell cards (960, 970, 980) don't.

MadMac_5 said:

I wonder if the OpenCL scheduler isn't taking full advantage of the sort-of-asynchronous compute capabilities that they offer, but CUDA does?

Click to expand...

Nvidia wrote OpenCL 1.2 support in 2015, there is no hint they have updated it. Meanwhile CUDA has been being updated with each new generation of cards. A modern CUDA driver may have 5 years of improvements over Nvidia's OpenCL driver.

Nvidia is at 1.2, the newest OpenCL is at 3.0 So all Nvidia cards act like Kepler in OpenCL.


(My theory is that at some point, Nvidia decided OpenCL was not selling Nvidia cards. This is just a theory; I have no facts.)