Google Diet Colab Notebook

[chalsall]

Originally Posted by lycorn

Thanks again for your time. Very useful info.

You're most welcome.

One other thing which just came to mind which might be relevant... From what you've written it seems like your workflow setup is (possibly) different than mine.

I have never tried running two Notebooks under the same Gmail/Google account context. I have ten different accounts, each of which has a single Notebook attached to it. These are spread across three different VPN contexts -- four appear to be here in Barbados, four appear to be sitting in a machine room in Texas, and the last two appear to be in a machine room in California.

They are all tied to the same cellphone number for the initial Gmail account setup, so Google must know they're related. But, still, I get effectively 24/7 CPU access on each account, and generally seven hours or so of GPU for each and every account.

I have to say it blows my mind a bit just how much compute Google must have, to be able to offer what we're all getting for free.

[lycorn]

I use 5 different Gmail/Google accounts, most of them unrelated. I have 2 notebooks attached to each: one for running mfaktc, the other one to run mprime. For many weeks, this hasn´t been a problem at all, and the CPU notebooks would run always for 12 hours. The GPU´s would run for variable time periods, but when stopped they would not affect the pairing CPU session.
Now (since the last 2/3 weeks or so) the differences are:

i) the GPU sessions are becoming shorter on average (although I am using GPUs less frequently than I used to do)

ii) more often than not, when the GPU session ends (sometimes after running for less than one hour) the CPU session is also disconnected. That has just happened this afternoon, during our exchange of messages: a CPU session was running for about 5 hours when it disconnected at the same time that the pairing GPU session, which disconnected after just over an hour.

iii) It is very common to see the messages about inactivity, both on the GPU and CPU sides, when these type of disconnections occur. It doesn´t happen all the time, though: this afternoon the sessions disconnected "silently" (no messages whatsoever...). When I tried to reconnect the GPU session I got a "cannot connect to GPU backend", with the additional explanation that it was due to usage limits. Given I hadn´t used the GPU for over 48 hours and that the session disconnected after just one hour, I really don´t know how to interpret it. The CPU session reconnected happily and is chugging along.

iv) if I don´t run GPU sessions, I have no problem at all getting the full 12 hours for the CPUs.

All in all, this may be a temporary issue. They could be doing changes/upgrades/maintenance to the system, who knows?
I think I will just stop using GPUs for one week or so and see how it goes.

[EdH]

I'm not doing very much. Mostly I'm testing ECM with/without GPU and other single instance type work with CADO-NFS, etc. I have gotten the "GPU inactivity" popup while in between GPU runs and I have gotten "inactivity" popups for CPU only, but none of these very often. I also rarely run for entire limits, but do on occasion.

[Chuck]

Originally Posted by chalsall

You're most welcome.

One other thing which just came to mind which might be relevant... From what you've written it seems like your workflow setup is (possibly) different than mine.

I have never tried running two Notebooks under the same Gmail/Google account context. I have ten different accounts, each of which has a single Notebook attached to it.

I have been running four notebooks under the same account for some time (paid tier).

[ATH]

Originally Posted by Chuck

I have been running four notebooks under the same account for some time (paid tier).

4 GPU notebooks on 1 Colab Pro account?

I have 2 Colab Pro accounts but only run 1 GPU notebook on each one, but they can run almost 24/7 if I remember to check and restart them quickly after they stop.

[Chuck]

Originally Posted by ATH

4 GPU notebooks on 1 Colab Pro account?

I have 2 Colab Pro accounts but only run 1 GPU notebook on each one, but they can run almost 24/7 if I remember to check and restart them quickly after they stop.

Each of the four ran for 24 hours at first, but now they cut off after about 18 hours and I have to let them "rest" for several hours before I can get a GPU again. I have never received anything other than a P100.

[Uncwilly]

Originally Posted by chalsall

Not sure if this is directly related to your workflow/use-case... But I've found it's good to click on the "Connect to hosted runtime" from the drop-down menu in the upper-right-hand side, to the immediate right of the CPU / Memory scale graphics. The Notebook then continues for the ~7 or so hours (for a GPU instance), with the graphic replaced with a "... Busy" message.

Trying to figure out how to do that in Android. Nothing yet.

[kladner]

Originally Posted by Chuck

Each of the four ran for 24 hours at first, but now they cut off after about 18 hours and I have to let them "rest" for several hours before I can get a GPU again. I have never received anything other than a P100.

I run 4 notebooks on my paid account. I currently have two free accounts, on which I have 2 notebooks each.

Right now, the paid account is running all 4. They have been up for right at an hour. I frequently set them up for overnight, and stop them sometime after I get up. They are almost always running. I habitually do a Factory reset, preferably after they stop, though if they take too long I will hit them with the FR.

Then I log out of the paid account and fire up one of the others. In the morning, I can almost always get at least one T4, and if not another, at least a P100 on a second notebook. If T4s are involved I usually stop them around 6 hours. At that point I switch to the other free account and see what I can get.
It's fun trying to game the system.

I don't really know squat about VPNs, but they sound really useful. Norton keeps pushing Secure VPN at me.

One more thing. Right now the RAM and Disk displays on all 4 NBs have been saying Busy for a while. The sessions are still running. I have sometime taken this as a warning off time running out, but these have not been running very long. Hitting on Connect to hosted makes it switch to Reconnect for a moment or 2, then it goes back to Busy. The instances are definitely still ticking over.
EDIT: The actual sequence when asking a Busy to connect is
Busy>Allocating>Reconnect>Busy and a box comes up lower left saying 'Cannot connect to the runtime.' I think it may be time to bail.

Well bail I did, and when I checked , I found that one of my secondaries was still logged in, though nothing was going on there. I think I've seen before that this can confuse or upset the Google. I logged out of all, logged into paid account, and the 4 instances are running P100s again.

[Uncwilly]

I got a full 12 hour session (non-GPU) uninterrupted. I have gotten a few sessions of late running upto 10 hours with gpu of late.

[moebius]

Originally Posted by Fan Ming

So why my attachment disappeared without notification... Did it violate any rules? Here I reupload it in zip.

Your compilation doesn't seem to work anymore.

[Fan Ming]

Originally Posted by moebius

Your compilation doesn't seem to work anymore.

The error occured at the OpenCL compilation process. Seems that the new OpenCL configuration in colab isn't compatible with old OpenCL codes in gpuowl anymore. I'll try to compile a new version of gpuowl on colab and upload it later if succeed.