GPU to 72 status...

[Chuck]

Originally Posted by kladner

But, I just signed off my paid account, with 4 P100s, called up a free account and got 2 T4s. This is approximately equal to 3 P100s. I usually stop this kind of run after 5-6 hours, and line up the paid account to run overnight.
3 free accounts and one paid let's me keep as much running as I have time to set up. I can always find at least one that will run with GPUs.

I also started using my free account with two notebooks after my daily 18 hour limit is reached on the paid account. However, I never get T4s and I don't want to bother with disconnecting and reconnecting in order to get them. Sometimes I even get a slow K80 but it's free...

[kladner]

Originally Posted by Chuck

I also started using my free account with two notebooks after my daily 18 hour limit is reached on the paid account. However, I never get T4s and I don't want to bother with disconnecting and reconnecting in order to get them. Sometimes I even get a slow K80 but it's free...

I guess I'm pretty snotty about K80s and P4s, and throw them back. I'll settle for P100s, though I can't forget that they are about 2/3 of a T4 for about 5 times the power consumption. This is a concern only with the free accounts. Paid always runs P100s. Getting the higher cards can take a bit of patience.

[Runtime Error]

Hi,

I signed up for GPUto72 maybe an hour ago, but I haven't received the activation email yet (checked spam folder). I'm fairly certain that I typed my email correctly. I also tried again with a secondary email and different username, but no luck (and I apologize to the admins for spamming it). It took me to the page where it says an email was sent, and if I try again with the same username it says that the username is taken, so it seems the account was created successfully.

Is the activation email script working? Perhaps I am just being impatient, but I am excited to join the Colab effort. (And I have time this weekend to play with it.) Thank you!!!

[Uncwilly]

Originally Posted by Runtime Error

Hi,

I signed up for GPUto72 maybe an hour ago, but I haven't received the activation email yet (checked spam folder).

Did you try logging in?

[chalsall]

Originally Posted by Runtime Error

I signed up for GPUto72 maybe an hour ago, but I haven't received the activation email yet (checked spam folder).

Hmmm... I didn't receive an email either; perhaps my provider's email subsystem is having issues today. I'll drill-down.

However, I've just activated your (first) account. Username starts with a capital "C". Please PM me if you have any further issues.

[Runtime Error]

Originally Posted by Uncwilly

Did you try logging in?

Yes I did, but I just tried again and it works! I suppose I should have been more patient. Thank you!

(Still no email but I guess that doesn't matter now.)

[Runtime Error]

Originally Posted by chalsall

Hmmm... I didn't receive an email either; perhaps my provider's email subsystem is having issues today. I'll drill-down.

However, I've just activated your (first) account. Username starts with a capital "C". Please PM me if you have any further issues.

(I must've been typing my reply above as you penned this)

It works! Thank you for activating it!!!

[kladner]

Originally Posted by Runtime Error

Hi,

I signed up for GPUto72 maybe an hour ago, but I haven't received the activation email yet (checked spam folder). .......

There may be a human element involved. Chalsall (who created and runs GPU72) may not have enough coffee to focus, yet.


OOPS! Time lapse. Belated response.

[kriesel]

Originally Posted by chalsall

Feedback appreciated.

If TF capacity is now ample, consider catching up on TF level to optimal, on already-primality-tested-once exponents. Both in 100Mdigit territory and elsewhere.

[S485122]

Originally Posted by kriesel

If TF capacity is now ample, consider catching up on TF level to optimal, on already-primality-tested-once exponents. Both in 100Mdigit territory and elsewhere.

Once one test is done the optima change, they are not the same as for an exponent that had no primality test done.

At the moment the different types of work types are done separately and on different hardware instead of consecutively as was the case a long time ago : a user used to get an assignment that implied first TF, then PM-1 and finally LL, now one gets a specific work type. This means that the optima should be base on the overall throughput of GIMPS in the different work types and not on the individual machine doing the test. Except if one does the three work types on the same exponent with the same hardware, the optimum for card X or Y is not really relevant.

Jacob

[kriesel]

Originally Posted by S485122

Once one test is done the optima change, they are not the same as for an exponent that had no primality test done.

At the moment the different types of work types are done separately and on different hardware instead of consecutively as was the case a long time ago : a user used to get an assignment that implied first TF, then PM-1 and finally LL, now one gets a specific work type. This means that the optima should be base on the overall throughput of GIMPS in the different work types and not on the individual machine doing the test. Except if one does the three work types on the same exponent with the same hardware, the optimum for card X or Y is not really relevant.

Jacob

Right. One bit less TF is justified, for example, if a primality test has already been done.
The GPU72 target listed at mersenne.ca is 81 bits in the 100Mdigit range for 2 tests saved, but there are many exponents there only up to 77-79 bits. https://www.mersenne.org/report_fact...=1&tftobits=72
Some of which have inadequate or no P-1 done yet. Some of which have a primality test started, or even completed. One example is
https://www.mersenne.org/report_expo...2203309&full=1
Runtime Error mentioned this issue to me, and gave an exponent example. https://mersenneforum.org/showpost.p...50&postcount=5
I easily found a factor with two more bits of TF. Xebecer's 4944. GhzD of PRP was wasted. https://www.mersenne.org/report_exponent/?exp_lo=332356909&full=1 But another 4944. GhzD of PRP DC won't be.

(For those curious about the tradeoff calculations, https://www.mersenne.org/various/math.php is a good start. For more, see
P-1 bounds determination https://www.mersenneforum.org/showpo...4&postcount=17
TF & P-1 optimization/tradeoff with each other and primality testing http://www.mersenneforum.org/showpos...7&postcount=12
What's a good P-1 factoring strategy? Best? https://www.mersenneforum.org/showpo...9&postcount=20
Or just go with one less TF bit level, and always use full PrimeNet P-1 bounds.)