Colab question

[moebius]

Originally Posted by Aramis Wyler

I don't know if something is wrong with the colab hookup,

And I don't know if it is related to your problem, but I have had the following phenomenon since 2 days:

I can no longer download my gpuowl savefiles to my computer with the Colab File-browser (in Colab File-browser the download progress is only fake displayed).
I work around this by simply downloading the intermediate results via the pull-down menu File-> Search in Google Drive.

[Aramis Wyler]

Just making a note here that this appears to have been resolved.

[kriesel]

Originally Posted by chalsall

Lastly, for some odd reason, one of the accounts always asks me to prove that I'm not a robot ("Recapta"). Annoying, but I figure my giving Google a bit of computer vision training data isn't a bad trade. And I've trained my own wet neural net to be really good at recognizing Buses, Cross-walks, Bicycles, Stairs and Chimneys...

One of 3 of my accounts is suspected of robothood, and you left out at least cars, hills, traffic lights, and motorcycles.

[firejuggler]

hrmmm
can't get it to work.

[Chuck]

It's a problem with the GPU72 site certificate.

[kladner]

I guess this as good a place as any to note that I just started up four notebooks on a paid account. They came in with Tesla V100 GPUs, which are reporting around 3800 GHz-d/d.
w00t!
First I've seen of this. The other account which had run for about 12 hours was still on P100s.
$6,909.00 on Amazon.
14-15 min for a 104.9M 74-75
28-30 min for 104.9M 75-76

[ATH]

I just got 1 V100 on 1 of my instances: 4.4 hours for 54M DC compared to 7.1 hours on a P100.


4.4 hours for a 54M exponent is amazing to think about. 33M exponent took me 211 days in 2003, that is several orders of magnitude...

[chalsall]

Originally Posted by kladner

They came in with Tesla V100 GPUs, which are reporting around 3800 GHz-d/d. w00t!

Wow! Sweet!!!

It will be interesting to see if these are also made available to the free accounts. It always struck me as a bit strange that the paid accounts were never given T4s, while the free accounts were getting them more often than the P100s.

[kriesel]

Originally Posted by Chuck

How can you find out how much power they are using?

nvidia-smi

as in https://www.mersenneforum.org/showpo...73&postcount=8 for example.

[kladner]

Originally Posted by Chuck

How can you find out how much power they are using?

https://www.nvidia.com/en-gb/data-center/tesla-v100/
It appears that they draw 300 W.


With regard to chalsall's mention of T4s, they draw 70 or 75 W. In comparison to P100s, they deliver about 50% more for 28% of the power. I wonder what the differences are. T4s don't seem to have the Tesla label. It's just strange to me that they can get so much work for so little power.

[Chuck]

Originally Posted by kladner

I guess this as good a place as any to note that I just started up four notebooks on a paid account. They came in with Tesla V100 GPUs, which are reporting around 3800 GHz-d/d.
w00t!
First I've seen of this. The other account which had run for about 12 hours was still on P100s.

I didn't realize until now that I had these V100 GPUs running; that explains the big spike in my GPU72 statistics graph. (I also have four running on a paid account)