Colab question - Page 10

kladner · 2020-05-01, 16:54

Just wondering, is it possible for users to get a display of completed work from Colab? Alternatively, could the GPU72 Completed Assignments page be enhanced by adding "sort by computer"? On that subject, it would be useful if the sortable columns could reverse the order on repeated clicks.
Just some idle wishes. Not of any great importance.

chalsall · 2020-05-01, 17:38

Quote:

Originally Posted by kladner

Just some idle wishes. Not of any great importance.

Of which I share... There are all kinds of cool data to expose on this. The good news is it's all being captured by the database, I just need the cycles to query and expose (graphs, tables, etc) the datasets.

In the immediate term... If you're using the Colab Autosubmit feature, all your results now appear on your Primenet computer report under the "GPU72_TF" virtual machine. So you can look there every day to see what Colab is contributing to your (and our) efforts.

P.S. This is one of the reasons its useful to have different AKeys (read: Instance Name) for each account. Then you'll be able to see when in the day it came online, and how long it survived...

kladner · 2020-05-01, 18:22

I have gradually gotten it together to have identifiable names for the paid account, plus the three free ones I use to work around usage limits. I usually let the paid run 4 notebooks overnight, and then mess with free accounts to try to fill the off time. It's gratifying when I can snag 2 T4 instances, which are about equivalent to 3 P100s. Sometimes the magic works, sometimes it doesn't. If not, I can usually get 2 p100s going. I don't consider P4s and K80s worth using up time allotments, so if that's all I can get I give up and wait a while.

In Overall System Progress, quarterly report, I see substantial bumps in output starting around 3/27. Trial factoring depth by day shows a spike for 74 bits, and a bigger one just after for 75 bits. Would these be evidence of Colab use picking up?

chalsall · 2020-05-01, 18:35

Quote:

Originally Posted by kladner

In Overall System Progress, quarterly report, I see substantial bumps in output starting around 3/27. Trial factoring depth by day shows a spike for 74 bits, and a bigger one just after for 75 bits. Would these be evidence of Colab use picking up?

Part of it. And, Oliver has been doing his thing recently. Also, our mysterious Anonymous DCTF'er has moved to LLTF (for a while, I hope). This weekly report gives a good indication of how things are currently.

kladner · 2020-05-02, 00:14

I forgot about Anon. It looks like they may soon pass Yellyfish, but Oliver won't have anyone stepping on his heels any time soon. That is, unless someone acquires a rack or two of T4s.
EDIT:
I counted Colab TFs for 5/1/20. (roughly, by eye) There are about 108, at around 39 GHz-d/d, so about 4200 GHz-d.

I haven't done the same for P-1, yet.
DOG! I love it when I can snag 2 T4s on a free run!

RichD · 2020-05-04, 02:03

I've recently been getting rogue assignments on my CoLab runs. Namely 97225243 which is a TF to 75 where my "key" was asking for TF to 77. But nothing (namely this one) shows on my assigned or completed.

chalsall · 2020-05-04, 03:39

Quote:

Originally Posted by RichD

I've recently been getting rogue assignments on my CoLab runs. Namely 97225243 which is a TF to 75 where my "key" was asking for TF to 77. But nothing (namely this one) shows on my assigned or completed.

Is there any chance you accidentally didn't put in your AKey?

Someone did a Colab run today as Anonymous (read: the AKey field wasn't filled in in the form). The candidate you listed above was worked by this instance.

I see two of your instances running right now, doing work to 77, as set in your instance configuration. Know that you'll sometimes be given lower work, stepping up to 77 ("Depth" is currently focused on high P-1'ed candidates).

kladner · 2020-05-04, 05:41

In my ongoing adoration of T4s, I just found that they run on 70 W. GTX1060 is pulling 125 W as I write.
EDIT: And doing ~600GHz-d/d vs 1700 or better on the T4.
EDIT: (T4) Which costs over $2000.
EDIT: Wow! P100 draws 250 W.

RichD · 2020-05-04, 11:16

Quote:

Originally Posted by chalsall

Is there any chance you accidentally didn't put in your AKey

I had to reconnect it (time expired?) and later noticed the AKey was missing. I stopped it and restarted (w/ AKey) and got a couple new assignments that never showed up in my list. I guess all is fine.

chalsall · 2020-05-04, 12:36

Quote:

Originally Posted by kladner

EDIT: (T4) Which costs over $2000.

Yeah, quite amazing what Google is ***giving*** us. A T4 instance on AWS costs $7.68 USD a day.

The only thing I can guess is Google has had great success with its own custom AI silicon, so basically have a bunch of GPUs already deployed which they no longer need.

chalsall · 2020-05-04, 12:39

Quote:

Originally Posted by RichD

I stopped it and restarted (w/ AKey) and got a couple new assignments that never showed up in my list. I guess all is fine.

OK... I'll move the eight (8#) Anon runs from yesterday over to your account. I'm afraid there's nothing I can do about the submitted results on Primenet.

2020-05-01, 18:22	#102
kladner "Kieren" Jul 2011 In My Own Galaxy! 2·3·1,693 Posts	I have gradually gotten it together to have identifiable names for the paid account, plus the three free ones I use to work around usage limits. I usually let the paid run 4 notebooks overnight, and then mess with free accounts to try to fill the off time. It's gratifying when I can snag 2 T4 instances, which are about equivalent to 3 P100s. Sometimes the magic works, sometimes it doesn't. If not, I can usually get 2 p100s going. I don't consider P4s and K80s worth using up time allotments, so if that's all I can get I give up and wait a while. In Overall System Progress, quarterly report, I see substantial bumps in output starting around 3/27. Trial factoring depth by day shows a spike for 74 bits, and a bigger one just after for 75 bits. Would these be evidence of Colab use picking up? Last fiddled with by kladner on 2020-05-01 at 18:22 Reason: line break removal

2020-05-02, 00:14	#104
kladner "Kieren" Jul 2011 In My Own Galaxy! 2×3×1,693 Posts	I forgot about Anon. It looks like they may soon pass Yellyfish, but Oliver won't have anyone stepping on his heels any time soon. That is, unless someone acquires a rack or two of T4s. EDIT: I counted Colab TFs for 5/1/20. (roughly, by eye) There are about 108, at around 39 GHz-d/d, so about 4200 GHz-d. I haven't done the same for P-1, yet. DOG! I love it when I can snag 2 T4s on a free run! Last fiddled with by kladner on 2020-05-02 at 05:23

2020-05-04, 05:41	#107
kladner "Kieren" Jul 2011 In My Own Galaxy! 27AE₁₆ Posts	In my ongoing adoration of T4s, I just found that they run on 70 W. GTX1060 is pulling 125 W as I write. EDIT: And doing ~600GHz-d/d vs 1700 or better on the T4. EDIT: (T4) Which costs over $2000. EDIT: Wow! P100 draws 250 W. Last fiddled with by kladner on 2020-05-04 at 06:11

2020-05-01, 16:54	#100
kladner "Kieren" Jul 2011 In My Own Galaxy! 10011110101110₂ Posts	Just wondering, is it possible for users to get a display of completed work from Colab? Alternatively, could the GPU72 Completed Assignments page be enhanced by adding "sort by computer"? On that subject, it would be useful if the sortable columns could reverse the order on repeated clicks. Just some idle wishes. Not of any great importance.

2020-05-04, 02:03	#105
RichD Sep 2008 Kansas 3×1,129 Posts	I've recently been getting rogue assignments on my CoLab runs. Namely 97225243 which is a TF to 75 where my "key" was asking for TF to 77. But nothing (namely this one) shows on my assigned or completed.