Page 1 of 1
Is there still a class/call for >24 core > 1TB mem?
Posted: Mon Mar 23, 2020 6:21 pm
by komradebob
When looking through the docs, I saw mention of ways to indicate a machine had many (>24 iirc) cores or large memory footprint but it looked depricated/said the experiment ended a few years ago. Is there still a way to do this and/or a need for it? I've had informal talks with some folks and we may be able to throw a whole bunch of idle lab/cloud machines at this (many thousands of cores) for some months, and I want to make sure they are properly classified.
If we can get some WU, that is.
Re: Is there still a class/call for >24 core > 1TB mem?
Posted: Mon Mar 23, 2020 6:26 pm
by bruce
Old docs refer to a deprecated category of CPU based WUs that needed lots of cores and lots of RAM. In that era, the GPU core couldn't handle those super-big projects so they all had to be run on the CPU. FAHCore_2* has been enhanced many times since then and those specific projects are no longer active.
Re: Is there still a class/call for >24 core > 1TB mem?
Posted: Mon Mar 23, 2020 6:31 pm
by Jesse_V
You might be able to pick up coronavirus work with a 12 or 24-core CPU slot. Most of the projects prefer an even number of cores, so if you had a 32-core machine, you could make two 16-core CPU slots and pick up work that way. That would be very helpful.
Do you have any good GPUs in your lab/cloud machines? Those can be enormously powerful for F@h too.
Re: Is there still a class/call for >24 core > 1TB mem?
Posted: Mon Mar 23, 2020 6:35 pm
by bruce
Right.
BTW, FAH doesn't need the 1TB of RAM but 16 + 16 cores are good.
Re: Is there still a class/call for >24 core > 1TB mem?
Posted: Mon Mar 23, 2020 6:41 pm
by treckin
I have been getting pretty steady cpu:18 jobs as of this morning, some new projects without descriptions are issuing them.
Re: Is there still a class/call for >24 core > 1TB mem?
Posted: Mon Mar 23, 2020 6:42 pm
by Jesse_V
To any sysadmin managing this lab/cloud instance, I do also want to mention that it's possible for a single FAHControl instance to connect to and remotely managed the F@h software on other machines. You can add the subnet and a password to each of the lab machines, then put that password and each of the IP addresses into the primary FAHControl instance to build remote control. That way it'll be easier to watch and control them if you want to make adjustments.
It's not necessary for the contributions, but if you care about points, it might also be a good idea to set up a user, team, and get a passkey for these machines so that they can earn more points under a named organization. That way you can put your name on it.
Re: Is there still a class/call for >24 core > 1TB mem?
Posted: Mon Mar 23, 2020 6:44 pm
by bruce
unfortunately writing a new description for a new project does get overlooked with everything else going on in the Labs.
Tell me the project number(s) and I'll nag them, but everything new is COVAID19
Re: Is there still a class/call for >24 core > 1TB mem?
Posted: Mon Mar 23, 2020 6:51 pm
by treckin
bruce wrote: unfortunately writing a new description for a new project does get overlooked with everything else going on in the Labs.
Tell me the project number(s) and I'll nag them, but everything new is COVAID19
11779 off the top of my head, but I have been checking and some of the researchers have been adding pages to their projects, including one cancer researcher that I noticed was issuing COVID WUs under a new project.
Cheers,
Re: Is there still a class/call for >24 core > 1TB mem?
Posted: Mon Mar 23, 2020 7:15 pm
by bruce
Just posting the URL to the missing description makes it easy. e.g-
https://apps.foldingathome.org/project?p=11779
Re: Is there still a class/call for >24 core > 1TB mem?
Posted: Mon Mar 23, 2020 7:56 pm
by komradebob
We are notoriously short on GPUs unfortunately as that doesn't really match up with the workloads we run. But have a ton of 48 core systems with 384GB-1.5TB and some 192 core 3TB systems.
Re: Is there still a class/call for >24 core > 1TB mem?
Posted: Mon Mar 23, 2020 7:57 pm
by komradebob
BTW, I've gotten no new work for my 4, 8, 16,48 core machines for over 24 hours.
Re: Is there still a class/call for >24 core > 1TB mem?
Posted: Mon Mar 23, 2020 8:05 pm
by Jesse_V
There's just been overwhelming demand due to the 10x flood of new users. The teams are getting more projects online and hopefully the servers can keep up with demand soon.
One option is opting in for "advanced" units that are in the final stages of testing before they move out to the full network. There's a small chance that the workunits are unstable and its prudent to keep an eye on things and keep us posted if there's any crashes, but at least there's less competition. Its your call. If you want to do this, you can add the configuration to the CPU slot. The option name is "client-type" with "advanced" as the value. Projects moving from Beta to Advanced are announced here: viewforum.php?f=24
Re: Is there still a class/call for >24 core > 1TB mem?
Posted: Tue Mar 24, 2020 4:35 pm
by treckin
bruce wrote: unfortunately writing a new description for a new project does get overlooked with everything else going on in the Labs.
Tell me the project number(s) and I'll nag them, but everything new is COVAID19
11778 is issuing WUs to my GPU but has no description on the project page:
https://apps.foldingathome.org/project?p=11778