Google Diet Colab Notebook

[petrw1]

Originally Posted by Ensigm

Got an almost-bewildering six hour and thirty minutes——you are hearing it right——P100 GPU runtime on the same free account today. At this point I'm no longer having the slightest idea about what Google is doing with their hardware. Maybe someone will notice this and put it to an end very soon, probably in a month or two.

I get 2 free P-100 almost every day each for about 8 hours.

[LaurV]

Do you guys do LL/PRP? (as I am only doing TF on colab, I don't get it what's the big fuss about P100, for me the best hit is T4, but that's rare and only survives 2-3 hours, the most of the time I get "no GPU")

[petrw1]

Originally Posted by LaurV

Do you guys do LL/PRP? (as I am only doing TF on colab, I don't get it what's the big fuss about P100, for me the best hit is T4, but that's rare and only survives 2-3 hours, the most of the time I get "no GPU")

I am doing GPUOwl P-1 under Colab.....P-100 are 4-8 times faster than any other GPU for this.
When I was doing TF a few months ago I too was getting 2-3 hours a day and almost never got a T4. I'm not sure why I get 3 times as much time using these same GPUs for P-1.

[Ensigm]

Originally Posted by Ensigm

Got an almost-bewildering six hour and thirty minutes——you are hearing it right——P100 GPU runtime on the same free account today.

And around 7h25min today (interestingly it was not an integer multiple of 30 min + a few minutes as usual). More and more close to @petrw1's 8 hours.

[lycorn]

How many NBs are you running concurrently?

[moebius]

Now it has caught me too, the first time that I no longer get a GPU runtime on all 3 accounts.

[moebius]

Originally Posted by moebius

Now it has caught me too, the first time that I no longer get a GPU runtime on all 3 accounts.

Works for me again, also with P100, but the running times are now a little shorter about 7-8 hours

[kriesel]

Originally Posted by LaurV

Do you guys do LL/PRP?

Yes. All GIMPS computation types are possible there on either cpu or gpu. But, latency can be a problem. Expiration can be a problem for wavefront primality first tests or higher. And CUDAPm1 did not seem to work. https://www.mersenneforum.org/showthread.php?t=24839
Occasionally an installation stops working for some reason, such as it no longer matches cudart, driver, and build versions, and then begins working again days later.
I just had cpu-only sessions of 12:11, 12:07, 12:03 on 3 simultaneous instances from one system, and 12:05 and 3:22 from another. In the next round, 3 of the 5 got gpus just now.

[bayanne]

MIKR seems to being allocated the same exponents as myself.
Is this a glitch in the system?

M29301253 from 2^70 to 2^71
M29301277 from 2^70 to 2^71

[petrw1]

Originally Posted by bayanne

MIKR seems to being allocated the same exponents as myself.
Is this a glitch in the system?

M29301253 from 2^70 to 2^71
M29301277 from 2^70 to 2^71

He's not using GPU72. He stole several of mine too. If you have anymore you might as well release them. He's now in 28.0 so won't be bothering us.

[bayanne]

Originally Posted by petrw1

He's not using GPU72. He stole several of mine too. If you have anymore you might as well release them. He's now in 28.0 so won't be bothering us.

Thanks, I have just released one that he has completed