mersenneforum.org

mersenneforum.org (https://www.mersenneforum.org/index.php)
-   PrimeNet (https://www.mersenneforum.org/forumdisplay.php?f=11)
-   -   3.5% increase in registered CPUs *this past week* (https://www.mersenneforum.org/showthread.php?t=22164)

NBtarheel_33 2017-03-30 00:57

3.5% increase in registered CPUs *this past week*
 
We have added just over 50,000 (yes, that is 5x10[SUP]4[/SUP]!) new CPUs to GIMPS in the past seven days, representing a 3.5% increase in the number of registered CPUs in *just one week*. Over 20,000 of them were added on Monday alone. The number of new users, however, in that same time period has remained relatively constant.

Is this a single new user with massive compute at their disposal, or did PrimeNet glitch and see a bunch of existing CPUs as new to the project...or did President Trump shut down NSA and direct all of their compute power to GIMPS? Wait, I know...it's Russia. It makes sense that Putin should like to know M50 before the Western imperialists get their hands on it.

One interesting effect of all of this concentrated new power is that the GIMPS maximum theoretical 30-day throughput has reached an all-time high of 1,437+ TFLOP/s, blowing away the previous record of 1,201+ TFLOP/s set in the wake of the discovery of M49 in early 2016. Actual 30-day throughput is at 334+ TFLOP/s, representing a meager 23% - as opposed to the normal 35-40% - efficiency value. If (when?) all of these new CPUs begin returning results, it should be easy to achieve 30-day sustained throughput levels of 500-550 TFLOP/s based on past efficiency values.

GP2 2017-03-30 01:35

I wonder if this is related to 29.1 somehow.

If you upgrade, does it create a new ComputerGUID ? The change from ThreadsPerTest to CoresPerTest might make the server think it's dealing with a new machine.

potonono 2017-03-30 02:49

[QUOTE]If you upgrade, does it create a new ComputerGUID ?[/QUOTE]

I believe any major windows 10 upgrades have a tendency to do that.

NBtarheel_33 2017-04-05 09:35

1 Attachment(s)
[QUOTE=GP2;455797]I wonder if this is related to 29.1 somehow.

If you upgrade, does it create a new ComputerGUID ? The change from ThreadsPerTest to CoresPerTest might make the server think it's dealing with a new machine.[/QUOTE]

The reporting of far larger than normal daily additions to the GIMPS CPU count is still occurring (see the attached graph). This behavior seems to have begun on or about March 24th, while George started the 29.1 [URL=http://www.mersenneforum.org/showthread.php?t=22141]thread[/URL] five days earlier. I also wonder if users would have already installed 29.1 on almost 100,000 computers (the number of "new" CPUs that have been added since the unusually large additions began).

It would be nice to track down the cause of this bug, whatever it may be. It is artificially inflating both the overall CPU count and the theoretical maximum throughput figures.

retina 2017-04-05 09:38

[QUOTE=NBtarheel_33;456215]It would be nice to track down the cause of this bug, whatever it may be.[/QUOTE]It might not be a bug. Who says it isn't real?[QUOTE=NBtarheel_33;456215]It is artificially inflating both the overall CPU count and the theoretical maximum throughput figures.[/QUOTE]But those values are of no consequence anyway. Only actual work done matters.

NBtarheel_33 2017-04-05 09:58

[QUOTE=GP2;455797]I wonder if this is related to 29.1 somehow.

If you upgrade, does it create a new ComputerGUID ? The change from ThreadsPerTest to CoresPerTest might make the server think it's dealing with a new machine.[/QUOTE]

FWIW, I just upgraded to 29.1 on a quad-core running Windows 8 and the existing computer name and GUID carried over without a hitch. No duplicate entries in the CPU list.

NBtarheel_33 2017-04-05 10:08

[QUOTE=retina;456216]It might not be a bug. Who says it isn't real?[/QUOTE]

'Twould be lovely if it were indeed real. But there has been no large increase in assignments or results that would be expected in the wake of 100,000 new computers joining GIMPS in such a short period of time.

[QUOTE=retina;456216]But those values are of no consequence anyway. Only actual work done matters.[/QUOTE]

Until we find a new prime, and the media looks for a concrete way to describe the size and power of our virtual supercomputer. "1.56 million CPUs with a maximum throughput of 1.75 petaflops" sounds impressive, but not if it is a significant overestimate.

Prime95 2017-04-05 21:41

It looks like GPUto72 is registering these CPUs to get LL work. Something's broken on our end or Chris' end.

Chris and I are looking for the bug.

In the meantime, don't get excited thinking a new mega-user has joined GIMPS.

Prime95 2017-04-06 19:07

Chris and I have found and fixed the problem.

48000 CPU rows have been deleted. Another 50000 rows will be cleaned up in the coming days.

Madpoo 2017-04-07 03:49

[QUOTE=Prime95;456294]Chris and I have found and fixed the problem.

48000 CPU rows have been deleted. Another 50000 rows will be cleaned up in the coming days.[/QUOTE]

And another 16K that can be removed now. At this rate George should be able to clear out the spurious extras by the end of the weekend.


All times are UTC. The time now is 12:17.

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, Jelsoft Enterprises Ltd.