Colab question

[KEP]

Originally Posted by chalsall

You have never run the GPU72 TF Notebook. Not a single assignment is associated with your AKey, so I don't understand what you're talking about.

No, because even when entering the GPU72_tf notebook, I got an "GPU usage limitasion" after typing in my access key and pushing play. I have not tested last night, but if I only have about 40 minutes of GPU acces every 24 hours there is really no point in running the notebook, since it will take litterally days to complete the assignment... anyway, I'll give the GPU72_tf notebook another shot, once I clean out the GIMPs assignments and seem to clear my GPU usage limitasion.

I think I'm starting to get the hang of it, but if I have to hit a special moment in a very short window of opportunity, with at least 24 hours between each start, I'm not sure it is going to be a thing for me, due to real life commitments.

Thanks for getting back to me

[KEP]

Originally Posted by chalsall

You have never run the GPU72 TF Notebook. Not a single assignment is associated with your AKey, so I don't understand what you're talking about.

I got the notebook working. Just accessed for GPU72 a Tesla P100, so now the interesting part is, if or when I reach a limitation

[chalsall]

Originally Posted by KEP

Just accessed for GPU72 a Tesla P100, so now the interesting part is, if or when I reach a limitation

If you haven't already, go to the drop-down menu in the upper-right hand area of the window, normally labeled "CPU" with a couple of bar-graphs. Choose "Connect to hosted runtime".

The session will reconnect, and then this should change to "Busy". This "locks in" the session until time allotment termination (anywhere from two to twelve hours). If you don't do this, it will time-out after between half an hour to an hour without human UI interaction.

[KEP]

Originally Posted by chalsall

The session will reconnect, and then this should change to "Busy". This "locks in" the session until time allotment termination (anywhere from two to twelve hours). If you don't do this, it will time-out after between half an hour to an hour without human UI interaction.

That was exactly what happened when I did my manual results, for mersenne.org - I hope, that it will be more like 12 hours before the allotment termination - at least for the time being that we are behind on several of the wavefronts. Thanks for the hint

Very great work you have done with the whole infrastructure of GPU72

[kriesel]

Originally Posted by KEP

I hope, that it will be more like 12 hours before the allotment termination

Google colaboratory free is 12 hours max for a cpu; 10 hours max for a gpu. My impression is free sessions get booted when there are paying customers requesting the resources in use by free sessions.
Google colab pro (paid) goes up to 24 hour session length, or so I've heard; no actual experience with that myself.
I've seen no difference, outside gputo72, using free Colaboratory, whether "busy" is used or not, except it generally causes a disconnect and means I have to connect to Google Drive an additional time.

[kladner]

Guess I've used up all my chips at the moment. Even my paid account is running CPU only.

[KEP]

Originally Posted by kriesel

Google colaboratory free is 12 hours max for a cpu; 10 hours max for a gpu. My impression is free sessions get booted when there are paying customers requesting the resources in use by free sessions.
Google colab pro (paid) goes up to 24 hour session length, or so I've heard; no actual experience with that myself.
I've seen no difference, outside gputo72, using free Colaboratory, whether "busy" is used or not, except it generally causes a disconnect and means I have to connect to Google Drive an additional time.

Well, so far, despite use of "Connect to hosted runtime" I only got 2 hours and is now again struggeling with the GPU usage limitasion. So for now, it is P-1 using CPU only. After that completes, I'm gonna focus on getting my manual reservation for GIMPs completed and then focus will be on building a new system, with a GTX 2060 Super and a fast CPU. That way I can manually control how to use my about 2000 GHzd/d and boost my progress at CRUS. Kind of sad to be "offered" something that is such fluctuating, simply because I do not sit around on this computer and way for another 24 hour period to begin :(

[kracker]

Originally Posted by kladner

Guess I've used up all my chips at the moment. Even my paid account is running CPU only.

Could be that google's GPU's are in high demand by google cloud/compute users at the moment. Colab gets the leftovers...

[kladner]

Originally Posted by kracker

Could be that google's GPU's are in high demand by google cloud/compute users at the moment. Colab gets the leftovers...

Could be. There are many mysteries in this new field. I'm making out OK.

[chalsall]

Originally Posted by kracker

Could be that google's GPU's are in high demand by google cloud/compute users at the moment. Colab gets the leftovers...

Absolutely possible. And, we have *no* idea what the various "signals" are for GPU allotment.

I can say that while many people continue to get GPUs, they're tending to the low range of the provisioning. My own ten (10#) accounts are getting about two hours a day at the moment. I am still regularly getting T4s.

[bayanne]

I wonder whether we might get longer sessions if we opted to use K80 and P4s ...