GPU to 72 status...

[nomead]

Originally Posted by chalsall

Perhaps you're after more specific data?

I use "nvidia-smi dmon" for that more specific, real-time data.Also there's options to show even more data on each line, if this is not enough. Memory usage, PCIe traffic, etc. Of course a GPU-Z graph is nice to look at, too.

palvi2:~> nvidia-smi dmon
# gpu   pwr gtemp mtemp    sm   mem   enc   dec  mclk  pclk
# Idx     W     C     C     %     %     %     %   MHz   MHz
    0   212    65     -    96     1     0     0  6800  1830
    0   213    65     -    96     1     0     0  6800  1830
    0   218    65     -    95     1     0     0  6800  1830
    0   213    65     -    94     1     0     0  6800  1830
    0   214    65     -    96     1     0     0  6800  1830

[chalsall]

Hey all. Need to do an update to the server, which involves a reboot (because the kernel will be replaced).

Scheduled for 2310 UTC today. Hopefully this will go better than the last time (read: the server should be off-line for about two minutes, rather than twelve hours)....

[chalsall]

So, the update and reboot appeared to work well.

Nice....

[Chuck]

Oliver has really been helping a lot with DCTF lately. I am surprised no one else has mentioned it. I see he has just taken another 3,000 assignments in the 58M range.

Thanks.

[chalsall]

Originally Posted by Chuck

Oliver has really been helping a lot with DCTF lately. I am surprised no one else has mentioned it. I see he has just taken another 3,000 assignments in the 58M range. Thanks.

I think we're all just kinda used to Oliver blowing our minds! But, yeah, thanks!

[kladner]

Only for "DC already done?

[ET_]

Originally Posted by kladner

Only for "DC already done?

Yep. I reverted to LL tests for the moment.

[chalsall]

Originally Posted by kladner

Only for "DC already done?

Oh, shoot. Thanks for pointing that out.

Oliver PM'ed me on Friday, saying something along the lines of "Whoops. I think I've reserved all of the remaining DCTF work..."

I scrambled to bring in some 55M work to feed to those who use automated fetchers (so they didn't "run dry"), but forgot to update the Preview code (sadly, these are actually separate code paths; they shouldn't be, but I've been a lazy programmer).

Until I have some time (super busy on other stuff at the moment), DCTF should be considered effectively optimally completed for now (except for those doing specialized work).

Technically, anyone who continues to do DCTF'ing in 55M are still contributing to GIMPS. But unless you have a 20xx class card, it's "sub-optimal".

Thanks for all the cycles everyone!!!

[Chuck]

Switching MISFIT to "First time tests" "Let GPU72 decide"

I have some 55M DCTF already queued. I'll let them finish.

[chalsall]

Originally Posted by Chuck

Switching MISFIT to "First time tests" "Let GPU72 decide" I have some 55M DCTF already queued. I'll let them finish.

Thanks.

Just to put on the table, until and unless we get a reasonably sizeable amount more "compute v7.5" class kit in the fleet, we're effectively done with DCTF.

RIP....

[Chuck]

Is there any preference whether I do TF spanning multiple bitlevels all at once or split it into stages?