12700 may not meet the timeout

[arisu]

I'm folding on a GTX 770M (Kepler), and P12700 R8 C147 G47 has an ETA of exactly one day, but the timeout is almost exactly 23 hours.

It's only barely going to miss the timeout at this rate. So two questions. Should I just dump it so it's immediately reassigned to someone else, or just let it pass the timeout and get assigned to someone else anyway in 23 hours (it will meet the deadline but that's kind of pointless)? And also, is there a way to be blacklisted from that particular project if it's just barely not able to make it on this GPU, besides just using iptables to block the work server?

[muziqaz]

You can dump it, to not waste your own time, and get another project which might be suitable for that GPU.

Species 3 and below will be disabled for this project, once eggs are smashed and distributed among family members

[arisu]

I suppose I can run it for like 5 minutes on my RTX 4090 Mobile then move it back. That should give it enough of a head start that it finishes on time. And by the time it is done I imagine species 3 will be disabled for it. Is that allowed or is that considered cheating anyway?

[muziqaz]

I suppose I can run it for like 5 minutes on my RTX 4090 Mobile then move it back. That should give it enough of a head start that it finishes on time. And by the time it is done I imagine species 3 will be disabled for it. Is that allowed or is that considered cheating anyway?

v8 does not do that

[arisu]

It does if you move it to another device and edit /etc/machine-id so that it doesn't know it's been moved. Although the devices have to be similar enough that they have the same cores, and if one is using OpenCL, you can't resume it with CUDA (it'll delete the checkpoint if that happens).

It would also be possible to manually run the core file on the work directory but I know that would violate the EULA if the results were sent back, since it's not being executed by the vanilla client.

[muziqaz]

It does if you move it to another device and edit /etc/machine-id so that it doesn't know it's been moved. Although the devices have to be similar enough that they have the same cores, and if one is using OpenCL, you can't resume it with CUDA (it'll delete the checkpoint if that happens).

It would also be possible to manually run the core file on the work directory but I know that would violate the EULA if the results were sent back, since it's not being executed by the vanilla client.

Let's assume that it doesn't, and leave it at that before starting giving ideas to people who have no clue what's going on
There is a request on GitHub to make V8 a bit more robust in regards to WUs in work folder. At the moment it is extremely picky, even within the same PC

[arisu]

Tell me when you've changed the minimum species. I'll turn my GPU on after that so it doesn't grab another 12700.

[muziqaz]

Constraints have been adjusted

[arisu]

Thank you.

And same issue with 14959 btw, which is a pretty big project for an old mobile GPU (>100k atom membrane simulation with 10 forces). Even on the now-species-7 Radeon 780M, it has a TPF of 7 minutes but 17 minutes on the GTX 770M. The GTX is just barely unable to make the timeout without help from a faster system.

The 1260x projects and 17655 are working fine (17655 is the largest in the 1765x series with 70k atoms). Those two projects are getting 500k PPD and 77k PPD, respectively.