GPU to 72 status...

[Runtime Error]

Originally Posted by chalsall

I've just activated your account.

Again, thanks. I had anticipated potentially spending a couple hours figuring it out, but wow that was super easy. I really appreciate how streamlined it is! haha trial factorer go burrrrr

A few of questions:

1) How can I delete a Notebook Access Key instance on GPU72, along with the associated work? I created too many and I'd like to return the assignments before they become overdue.

2) Is there a way to request to TF exponents at the 332M prize wave front? (see kriesel’s above post; that exchange was the impetus for me to join the TF effort)

3) Is there a way to request consecutive bit levels on the same exponents? e.g. take something from 73 to 81 instead incremental bit levels on different exponents? It may sound silly, but it would make me feel more ownership over and attachment to the exponents that I’m factoring, and it would add to the excitement that I feel when finding factors. (omg it spent a day factoring it and just before my session expired… BOOM!!1) My pitch: You (chalsall) have said a few times that you like to give folks agency over their contribution 😉.

4) A request for anecdotal evidence (the plural of which is data): How do you get a T4?

5) Any tips on writing a Windows-runable script that launches a Colab instance (and hits "connect" + "run") every 24 hours? I've done similar things w/ Python, but I imagine someone has a more elegant solution.

[Uncwilly]

Originally Posted by Runtime Error

4) A request for anecdotal evidence (the plural of which is data): How do you get a T4?

When you see that you have been assigned a P4 or K80, interrupt the execution. Then "factory reset runtime". Restart the code. If you get a T4, let it run. If you get a P100, you might want to let it run. Else, repeat. Wend.

And once you get one. Then do the trick of "Connect to hosted runtime"

[Uncwilly]

It looks like my MISFIT is not able to get assignments in the 332M range. I am not in front of the machine in question, nor did I have time to investigate the issue. It still has days of assignments available, so no rush.

[chalsall]

Originally Posted by Runtime Error

Again, thanks. I had anticipated potentially spending a couple hours figuring it out, but wow that was super easy. I really appreciate how streamlined it is! haha trial factorer go burrrrr

Cool. Glad to get the feedback.

Originally Posted by Runtime Error

1) How can I delete a Notebook Access Key instance on GPU72, along with the associated work? I created too many and I'd like to return the assignments before they become overdue.

I don't currently have a way to delete Access Keys, but I probably will need to add that. As well as "merging" Access Keys and CPUs, much like Primenet.

However, don't worry about taking too many assignments. After an instance is terminated, any assignments which have not been worked are returned to the pool within half an hour. Those who have had work done on them are held until you spin up another instance.

Originally Posted by Runtime Error

2) Is there a way to request to TF exponents at the 332M prize wave front? (see kriesel’s above post; that exchange was the impetus for me to join the TF effort)

There is now. I've just added two new work types to the Instances Edit Form (click on the links under the work-type columns), "LMH Trial Factoring (Breadth First)" (going to 76 bits) and "LMH Trial Factoring (Depth First)" (going to 78 bits).

I would appreciate feedback as to how deep people want to take these? 81 bits would take a while, but that's approaching optimal.

Originally Posted by Runtime Error

3) Is there a way to request consecutive bit levels on the same exponents? e.g. take something from 73 to 81 instead incremental bit levels on different exponents? It may sound silly, but it would make me feel more ownership over and attachment to the exponents that I’m factoring, and it would add to the excitement that I feel when finding factors. (omg it spent a day factoring it and just before my session expired… BOOM!!1) My pitch: You (chalsall) have said a few times that you like to give folks agency over their contribution 😉.

I hear what you're saying. But that would be a royal pain in the butt for me to have the server manage.

As a compromise, if you choose "LMH Depth First" you will consecutively get the lowest available candidate to take up to the next level. So while you won't "own" a particular candidate, you will be contributing to the first candidates to be released back to Primenet for LL assignment.

Originally Posted by Runtime Error

4) A request for anecdotal evidence (the plural of which is data): How do you get a T4?

1. *Don't* be on the paid tier.

2. Patiently do the Factory Reset cycling trick until you get lucky.

Originally Posted by Runtime Error

5) Any tips on writing a Windows-runable script that launches a Colab instance (and hits "connect" + "run") every 24 hours? I've done similar things w/ Python, but I imagine someone has a more elegant solution.

I would advise against this. For some reason, one of my accounts gets hit with the Recaptcha challenge every time I interact with it, so Google is aware some are going to be trying to automate this.

[chalsall]

Originally Posted by Uncwilly

It looks like my MISFIT is not able to get assignments in the 332M range. I am not in front of the machine in question, nor did I have time to investigate the issue. It still has days of assignments available, so no rush.

Your MISFIT was asking for work to take up to 74 bits. I have added a conditional such that it bumps up the pledge level to 76.

[chalsall]

Originally Posted by Uncwilly

When you see that you have been assigned a P4 or K80, interrupt the execution. Then "factory reset runtime". Restart the code. If you get a T4, let it run. If you get a P100, you might want to let it run. Else, repeat. Wend.

Personally I won't settle for anything less than a P100. Not worth my (free) time allotment. (I can't believe I just wrote that... )

To speed things up, you don't even have to interrupt the executing Notebook. Just Factory Reset with your mouse, and then "Cntl-F9" to run all the Sections (usually, but annoyingly not /aways/, in order from the top).

It can take up to five such cycles, but you will /eventually/ get something better than a P4 (~98% of the time).

[Runtime Error]

Thank you for the helpful replies!

I've currently got two accounts using two notebooks each, running on the same machine (using different browsers). IP address doesn't seem to matter, nor does phone number on the associated account. It seems easy enough to get P100s, and I'll probably stick with them since T4s seem rare, although I did manage to get one!

Originally Posted by chalsall

1. *Don't* be on the paid tier.

2. Patiently do the Factory Reset cycling trick until you get lucky.

Wait, they give paid users lower quality equipment?

[chalsall]

Originally Posted by Runtime Error

Wait, they give paid users lower quality equipment?

Yup. We don't understand it either...

BTW, for those really serious about LMH work... I've added another work-type: Nominal. This used to be called "Depth", and goes to 78 bits. What is now called "Depth" goes to 81 bits, which is where we'll release back to Primenet at.

But, be aware this takes a long time! 17.5 hours on a P100:

20200512_001400 ( 3:54): Starting trial factoring M332193xxx from 2^80 to 2^81 (737.12 GHz-days)
20200512_001404 ( 3:54):  Exponent  TF Level  % Done     ETA   GHzD/D  Itr Time |   Class #,   Seq # |    #FCs | SieveRate |  SieveP
20200512_001408 ( 3:54): 332193xxx  80 to 81    0.1%  17h31m  1008.85   65.759s |    0/4620,   1/960 | 393.85G | 5989.4M/s |   82485
20200512_001611 ( 3:56): 332193xxx  80 to 81    0.3%  17h29m  1007.84   65.825s |   12/4620,   3/960 | 393.85G | 5983.4M/s |   82485
20200512_001718 ( 3:57): 332193xxx  80 to 81    0.4%  17h28m  1007.90   65.821s |   24/4620,   4/960 | 393.85G | 5983.7M/s |   82485
20200512_001823 ( 3:58): 332193xxx  80 to 81    0.5%  17h28m  1007.27   65.862s |   25/4620,   5/960 | 393.85G | 5980.0M/s |   82485
20200512_001930 ( 3:59): 332193xxx  80 to 81    0.6%  17h26m  1007.61   65.840s |   28/4620,   6/960 | 393.85G | 5982.0M/s |   82485
20200512_002035 ( 4:00): 332193xxx  80 to 81    0.7%  17h27m  1005.67   65.967s |   33/4620,   7/960 | 393.85G | 5970.5M/s |   82485
20200512_002142 ( 4:01): 332193xxx  80 to 81    0.8%  17h25m  1006.49   65.913s |   37/4620,   8/960 | 393.85G | 5975.4M/s |   82485
20200512_002247 ( 4:03): 332193xxx  80 to 81    0.9%  17h25m  1005.47   65.980s |   40/4620,   9/960 | 393.85G | 5969.3M/s |   82485

It will be interested to see how long a P-1 run is going to take on Colab instances... I'll try that tomorrow.

[Uncwilly]

Originally Posted by chalsall

Your MISFIT was asking for work to take up to 74 bits. I have added a conditional such that it bumps up the pledge level to 76.

I fixed my MISFIT. I just need to work to get it to keep less in queue. And on another machine MISFIT does not seem to fetch. It has let the queue empty.

[kladner]

I recently changed an 8 core i7 machine from running 2 P-1 workers with 4 cores, to 4 workers with 2 cores. In My GPU72 assignments and completed pages I only see the machine name and (2), (3), (4). When I noticed and checked the machine, which is doing nothing else at the moment, and found all 4 workers chugging away. I had it call PrimeNet just to make sure that was working.
I wonder if M_ (1)'s work is being credited.

[Runtime Error]

Originally Posted by chalsall

Yup. We don't understand it either...

BTW, for those really serious about LMH work... I've added another work-type: Nominal. This used to be called "Depth", and goes to 78 bits. What is now called "Depth" goes to 81 bits, which is where we'll release back to Primenet at.

But, be aware this takes a long time! 17.5 hours on a P100

Thanks for adding this. I've been trying some of the 81 bit level jobs. It seems that a notebook instance likes to first start a new exponent upon launch, and if it finishes, it will move on to any partially completed jobs. But unless it gets a T4, it will not finish the job within 12 hours. I currently have a handful of partially completed 81-bit jobs, and this evening's notebooks just started fresh exponents. Do you have any advice? Thank you!