mersenneforum.org

mersenneforum.org (https://www.mersenneforum.org/index.php)
-   PrimeNet (https://www.mersenneforum.org/forumdisplay.php?f=11)
-   -   "First-time assignments" (https://www.mersenneforum.org/showthread.php?t=20646)

Rodrigo 2015-11-08 20:53

"First-time assignments"
 
I have a laptop with an i7-3630QM processor set up to do "first-time assignments".

Up until recently, this meant that it was doing first-time LL tests exclusively, as I'd intended for it. But in the past few weeks, the laptop has started to do DCs in the 34xxxxxx range even though I haven't touched the settings. I have four workers going and two of them are doing this now.

Has there been a change in the PrimeNet assignment rules (or anywhere else) to account for this change?

Thanks for any info on this!

S485122 2015-11-08 22:21

I think, recently it was decided that all computers having first time LL work would get a few DC now and then to test the computer for reliability. You might find this in a thread. I do not remember if it is in one of the threads about double-checks or in the new assignment rules thread.

Jacob

kladner 2015-11-08 22:31

[QUOTE=S485122;415443]I think, recently it was decided that all computers having first time LL work would get a few DC now and then to test the computer for reliability. You might find this in a thread. I do not remember if it is in one of the threads about double-checks or in the new assignment rules thread.

Jacob[/QUOTE]

I remember George saying recently that he had upped the DC percentage from 10% to 20%, but I thought this was in regard to What Makes Sense. I tried to track the comment down in several threads, but did not find it.

chalsall 2015-11-08 23:27

[QUOTE=kladner;415448]I remember George saying recently that he had upped the DC percentage from 10% to 20%, but I thought this was in regard to What Makes Sense. I tried to track the comment down in several threads, but did not find it.[/QUOTE]

That was in the [URL="http://mersenneforum.org/showpost.php?p=413846&postcount=452"]Strategic Double Checks[/URL] threat.

It makes sense. Never trust the sanity of anyone or anything.

Rodrigo 2015-11-08 23:42

Thanks, guys.

I just double-checked (so to speak) and the laptop is indeed set to first-time tests, rather than to "what makes the most sense." Maybe something, somewhere, is overriding that setting? :unsure:

Rodrigo

chalsall 2015-11-09 00:00

[QUOTE=Rodrigo;415458]I just double-checked (so to speak) and the laptop is indeed set to first-time tests, rather than to "what makes the most sense." Maybe something, somewhere, is overriding that setting? :unsure:[/QUOTE]

OK, I can't speak to this authoritatively, but IMO doing LL tests on a laptop is a really stupid idea.

Constrained thermodynamics and power systems on almost all laptops.

Where does this help anywhere unless you wish to crash and burn?

That's a serious question.

Chuck 2015-11-09 02:02

[QUOTE=chalsall;415460]OK, I can't speak to this authoritatively, but IMO doing LL tests on a laptop is a really stupid idea.

Constrained thermodynamics and power systems on almost all laptops.

Where does this help anywhere unless you wish to crash and burn?

That's a serious question.[/QUOTE]

I have two laptops that have been running Prime95 around the clock for almost three years. A Sandy and an Ivy. The Ivy I use only three of the four cores to keep the temperature safe. They are sitting on fan bases. Half first-time tests and half double checks. I only look at them every few days. The lids are partially closed to keep the screens dark.

Madpoo 2015-11-09 04:12

[QUOTE=Rodrigo;415458]Thanks, guys.

I just double-checked (so to speak) and the laptop is indeed set to first-time tests, rather than to "what makes the most sense." Maybe something, somewhere, is overriding that setting? :unsure:

Rodrigo[/QUOTE]

There was a weird little thing where accounts created after a certain date (April of 2014 I think it was) were set to get double-checks by default, although Prime95 would default to "whatever makes sense". It only did double-checks when new worker threads were created in some particular way.

We fixed that and now new worker threads will use the Prime95 default (whatever makes sense). I guess just to make sure, you should check the setting for each worker and make sure they're set to the work type you think. Change it to something else and back and let it talk to the server, just to make sure things are in sync?

Rodrigo 2015-11-09 06:42

[QUOTE=Madpoo;415477]There was a weird little thing where accounts created after a certain date (April of 2014 I think it was) were set to get double-checks by default, although Prime95 would default to "whatever makes sense". It only did double-checks when new worker threads were created in some particular way.

We fixed that and now new worker threads will use the Prime95 default (whatever makes sense). I guess just to make sure, you should check the setting for each worker and make sure they're set to the work type you think. Change it to something else and back and let it talk to the server, just to make sure things are in sync?[/QUOTE]
All right, I changed all four workers individually to "world record sized numbers to test," then changed them each back to first-time tests and observed the notification about communicating with PrimeNet.

Guess there's nothing further to do now until new assignments come in and we see what they are.

Thanks for the idea.

Rodrigo

VBCurtis 2015-11-09 08:10

[QUOTE=chalsall;415460]OK, I can't speak to this authoritatively, but IMO doing LL tests on a laptop is a really stupid idea.

Constrained thermodynamics and power systems on almost all laptops.

Where does this help anywhere unless you wish to crash and burn?

That's a serious question.[/QUOTE]

I ran a laptop i7-quad (2600QM, I think) for 5 years 24/7 with 3-4 threads of LLR (the JV version of P95) and 4-5 threads of ECM or NFS sieving. The laptop did fail, but it was due to the power plug physically disconnecting from the board, rather than any heat issue. If the laptop is well-made and well-cooled (i7 laptops are more likely than most to qualify for both), I don't see why running Prime is a bad plan.

Oh- I *did* set windows to max out at 90% of max CPU power, so it was 37W instead of 54W I believe that made max speed 1.7GHz instead of 1.8, well worth the 30% power reduction. CPU-Z temp readings were usually in the low 60s; when they hit 70, I took the plastic cover off and de-dusted (twice in 5 years).

I also ran CUDA-LLR on the GPU for 12 months straight, and lots of NFS poly select/etc after I realized I'd likely burn up the card with LLR. My kill-a-watt said the power supply drew 95W from the wall with GPU running full tilt and 8 CPU threads going. Nice laptop!

LaurV 2015-11-09 08:44

OTOH, getting one "invasive DC" in 10 assignments, even when you take "first LL tests only", totally makes sense for me, to check the sanity of your hardware from time to time. [B]I salute that initiative[/B] and have nothing against it! Saves me the trouble to do that by hand! Eventually, DC assignment should be repeated (i.e. another exponent, but again a DC) if the first "invasive DC" fails to match the residue (it can be a problem with your hardware, or the initial first-time residue was indeed bad).

Rodrigo 2015-11-09 15:22

One of the workers is on its second consecutive DC. That doesn't sound like a "from time to time" reality check.

In fact, getting that second straight DC was the event that triggered my posting here -- because it could no longer be viewed as merely an occasional deviation from the setting.

Rodrigo

kladner 2015-11-09 17:28

I think this has been covered, but have you made sure that the PrimeNet CPU>Computer Properties page agrees with Prime95 Test>Worker Windows? This should be part of the "change-and-change-back" procedure Aaron mentioned above.

This was important for my setup, after I decided to treat the FX-8350 as properly a quad-core instead of an octo-core. With each integer pair and its respective FPU set as a worker, P95 sees four [U]Worker Windows[/U], but PrimeNet still sees eight [U]Worker Threads. [/U]On P95 this shows as-
[LIST=1][*]DC[*]LL[*]DC[*]DC[/LIST]On PrimeNet it looks like-
[LIST=1][*]DC[*]LL[*]DC[*]DC[*]DC[*]DC[*]DC[*]DC[/LIST]In this case, I don't think the settings for THREADS 5-8 have any effect.

Sorry if I am repeating, but this situation seems really persistent, so I am going all Help Desk on you. :smile:

Rodrigo 2015-11-09 22:37

To the Help Desk: :wink:

Just checked the Computer Properties page in PrimeNet as you suggested. There are four worker threads, all set to do "LL first test" work; and the percentage of "DC instead of LL" is at zero.

Is there anything else I should be looking for?

Thanks very much! :tu:

Rodrigo

kladner 2015-11-10 04:00

Hi Rodrigo,

I am afraid that is the best I can do. It seems you have things covered, but the assignments contradict that.

Madpoo 2015-11-10 04:28

[QUOTE=kladner;415641]Hi Rodrigo,

I am afraid that is the best I can do. It seems you have things covered, but the assignments contradict that.[/QUOTE]

Yeah, the DC assignments it got may have just been a result of those workers being set to "get whatever makes sense" and getting a 20% chance of DC assignments.

They should be okay now and we'll know for sure next time they request more work.

All in all though, a few double checks here and there is a good thing. It helps with the DC work and also gives us an idea of how reliable the system is long before anyone starts double-checking your work. :smile:

Rodrigo 2015-11-10 05:57

Yeah, I'm OK with the occasional DCs for that reason. What [B]is[/B] puzzling, though, is that I never did have that computer set to do "whatever makes the most sense," it was always set to do first-time LL tests.

Anyway, I changed the settings back and forth as instructed and hopefully that'll take care of it. I'll report back if the next assignment or two are still not LLs.

Rodrigo


All times are UTC. The time now is 10:05.

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