TalkChess.com

A lot of people who read this forum have real nice GPUs.
You can go to the Folding At Home project and donate CPU/GPU time for protein folding.
If you chose "Any" for the project, they now have Corona virus 19 as a target. You cannot specifically choose that option, though I guess it will get extra effort due to the crisis.
One of my machines is dedicated to that task

I should warn you that it will make chess tasks difficult on that machine while the folder is running.
Even though it does not use a lot of RAM, and even if you turn off the CPU version so that it is only supposed to use your GPUs, it will still make your chess program page fault like mad. I guess it has a lot of virtual memory or something. The GPU only version uses the OpenCL driver which uses the CPUs for some tasks, so you will get one thread used for each GPU you allow them to use, but it seems to have a large impact on other things the machine is trying to do.

All in all, if it saves a few million people it seems like a good idea to me. And even if it does not solve the virus chemistry in time, it seems valuable to at least have tried.

Nice idea, Dann!

Apart from Folding At Home, does anybody know if the medical research for a vaccine implies
embarassingly parallel computation of some kind? If so, it would be interesting to suggest that
the EU authorities write and distribute a PC client to the millions of people who are confined
at home for the next few months.

Stéphane

The code for folding at home is public.
It uses my GPUs rather badly (I told it to use 2 GPUs and it uses about 1/3 of 1 GPU at max).
It is using OpenCL in my Nvidia 2080 TI Supers, when native Cuda code would be much faster.
I guess that if someone skilled at GPU programming got involved it would at least quadruple the throughput.

I am getting 50% GPU usage now, on both GPUs for the system where utilization was low, so something seems to be going better than before.
Maybe we could make a team called "TalkChess" or something like that.

Dann Corbit wrote: ↑Tue Mar 17, 2020 5:18 pm It is using OpenCL in my Nvidia 2080 TI Supers

How can you tell?

Thanks for reminding me about FAH.

grahamj wrote: ↑Wed Mar 18, 2020 9:22 pm

Dann Corbit wrote: ↑Tue Mar 17, 2020 5:18 pm It is using OpenCL in my Nvidia 2080 TI Supers

How can you tell?

Thanks for reminding me about FAH.

I posted some messages on the FAH message board

Dann Corbit wrote: ↑Tue Mar 17, 2020 11:40 pm I am getting 50% GPU usage now, on both GPUs for the system where utilization was low, so something seems to be going better than before.
Maybe we could make a team called "TalkChess" or something like that.

I think utilization is protein dependent. When I started participating 15 years ago, before GPU code, it used SSEx instructions for optimizations. But there were different algorithms for types of folding. Some were really slow while the fastest took only a 5 days or so. When GPU algorithms started in the late 2000s, things sped up about five-fold (on my 2009 hardware).

I suspect the method of folding dictates how much the GPU can be utilized. Just guessing though since I've not participated in 10 years.

Do NOT be discouraged if you do not have a GPU computer. FAH accepts CPU based computers. I have 3 CPU-based boxes connected online to them - all are Intel 5's, 2 have 4 cpu and 1 has 6 cpu.

The setup asks for you to choose between low/medium/high. High uses all cpus, Medium uses 1 less cpu than high, and low I don't know about. I suggest using Medium for 2 reasons- first, it lets you have full control of 1 cpu to do your everyday chores, and second, there is less chance of an overheating problem than if you use high. There is a possibility that your machine will be in almost constant use.

The way it works is that you first have to wait in a queue for a project to be assigned. Sometimes the wait is more than an hour. After your job finishes (usually about 1 hour), you go back in the queue. Btw, the 3 boxes I have online with FAH all have the same IP, and I did not notice any problem with that situation.

Apparently a group of scientists used Summit to come up with chemicals that may stop covid-19.

https://chemrxiv.org/articles/Repurposi ... 11871402/3

I do not know if it is protein folding that they did but it seems scientists have access to the fastest supercomputers which I am not sure can be matched by folding@home. Earlier, DeepMind also put forward their own results with AlphaFold.
I find it exciting that you can cure diseases with just computation. I guess in a year we will know who will take the top prize and how
valuable computational methods were to finding a vaccine for COVID-19