wondering if moonshot sprint 5 has lower # of WU

[Knish]

Compared to sprints 1-4, when I spot check my clients many of them are not running 1342x during sprint 5. I checked the logs of one and it got only 14 moonshot WUs out of 42. It's probably very similar for the other V100's running. Do we keep draining the moonshot servers dry? That's pretty cool if true!

[Neil-B]

There may be more other covid projects running .. not sure how priorities are worked across the same preference

[PantherX]

With the same preference, the Project weight is used which is set by the researchers depending on various factors.

[Knish]

hmm, 4 out of 5 of the tesla GPUs are currently folding on cancer projects 16918

[psaam0001]

I've seen only a few Moonshot WU's being worked on by systems (my Win 10 system has done one or two to my knowledge).

Though, I've been putting time into getting the most out of my legacy desktop systems (just did a 2-core to 4-core Socket 775 Core2 upgrade today w/new CPU cooling fan; and recently added ASUS GT 710's where possible to get more GPU WU's processed).

Those will be the last hardware upgrades for my legacy (pre-Socket AM4) systems.... Starting in January 2021, I'm going to start saving for a real "COVID Buster".... Got my wish list set up--just got to save the money.

Paul

[bruce]

... and recently added ASUS GT 710's where possible to get more GPU WU's processed

Be careful about the GT710. It's a Kepler so support is more difficult. The GT730 is at least a Fermi (I run two of them.) and the GT1030 is A LOT more GPU for a similar entry-level price.

Of course of you got a bin full of 710s, go ahead and use them for OpenCL. One nice thing about them is that they fit in a single half-height slot.

[Joe_H]

Be careful about the GT710. It's a Kepler so support is more difficult. The GT730 is at least a Fermi (I run two of them.) and the GT1030 is A LOT more GPU for a similar entry-level price.

Just a small correction, support is easier for the newer Kepler based nVidia GPUs than for Fermi. The GT 710 was based on Kepler, two of the three different GT 730 cards were based on Kepler, one was based on Fermi and has 1/4 the shaders and only OpenCL 1.1 support.

[bruce]

Oops. You're right.

[psaam0001]

These GT 710's were purchased to fill in those PCIe x1 slots that were doing nothing at all.

I don't have any more PCIe slots I can use at this time. They are all full.

Wish list "A" has a system configuration w/a 64 core Epyc CPU and 7 RTX 4000's on it.

Wish list "B" has a system configuration w/a 64 core Ryzen Threadripper CPU; 1 GT 710 (PCIe x1) and 3 RTX 3090's....

And Fedora will be the OS.

[Clarification: Only one of the above would be built--and only when the funds are saved up to pay for the parts in full.]

Paul

[psaam0001]

BTW: My GT 1030 is on a Moonshot WU right now.

Paul

[PantherX]

...Wish list "A" has a system configuration w/a 64 core Epyc CPU and 7 RTX 4000's on it.

Wish list "B" has a system configuration w/a 64 core Ryzen Threadripper CPU; 1 GT 710 (PCIe x1) and 3 RTX 3090's.... ...

Personally, I would suggest a slightly different approach:

System 1 - Only CPU folder 24/7/365.25
It would be a 64C/128T that just folds CPU WUs. No dGPU at all.

System 2 - Only GPU folder 24/7/365.25
It would be vary between dual to quad GPUs depending on the PCIe lanes, PSU requirement and cooling. Each GPU would be allocated 2 CPUs

Reason is that you would have two specialized systems without making a combo box and having more efficiency and less headaches when it comes to troubleshooting or upgrading.

[psaam0001]

Right.... These are just thoughts (only 1 system would be built). Though, my current candidate motherboard for Wishlist "B" has 3 PCIe x16 slots (supporting 16 lanes) and 1 single lane slot. I may need an extension cable for that third x16 slot though.

Besides, it might be 2 or 3 years before I have the funds saved up. Which means new hardware will be on the market, and new challenges will be on researchers minds.

Paul