PrimeGrid
Please visit donation page to help the project cover running costs for this month

Toggle Menu

Join PrimeGrid

Returning Participants

Community

Leader Boards

Results

Other

drummers-lowrise

Advanced search

Message boards : Number crunching : Recent average CPU time in project preferences is misleading

Author Message
Profile composite
Volunteer tester
Send message
Joined: 16 Feb 10
Posts: 841
ID: 55391
Credit: 784,056,583
RAC: 431,184
Discovered 2 mega primesFound 1 prime in the 2018 Tour de Primes321 LLR Turquoise: Earned 5,000,000 credits (5,477,467)Cullen LLR Gold: Earned 500,000 credits (776,297)ESP LLR Ruby: Earned 2,000,000 credits (3,433,680)Generalized Cullen/Woodall LLR Ruby: Earned 2,000,000 credits (2,093,491)PPS LLR Sapphire: Earned 20,000,000 credits (27,332,255)PSP LLR Turquoise: Earned 5,000,000 credits (6,587,988)SoB LLR Sapphire: Earned 20,000,000 credits (34,191,283)SR5 LLR Turquoise: Earned 5,000,000 credits (6,110,877)SGS LLR Ruby: Earned 2,000,000 credits (3,486,285)TRP LLR Turquoise: Earned 5,000,000 credits (7,070,795)Woodall LLR Amethyst: Earned 1,000,000 credits (1,693,614)321 Sieve (suspended) Emerald: Earned 50,000,000 credits (50,256,050)Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,571,178)Generalized Cullen/Woodall Sieve (suspended) Emerald: Earned 50,000,000 credits (50,009,610)PPS Sieve Double Silver: Earned 200,000,000 credits (385,656,505)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Jade: Earned 10,000,000 credits (10,165,888)TRP Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,071,454)AP 26/27 Turquoise: Earned 5,000,000 credits (6,616,128)GFN Emerald: Earned 50,000,000 credits (53,771,465)WW Gold: Earned 500,000 credits (932,000)PSA Double Bronze: Earned 100,000,000 credits (102,762,384)
Message 149068 - Posted: 28 Feb 2021 | 14:23:17 UTC
Last modified: 28 Feb 2021 | 14:39:32 UTC

PPSE - Recent average CPU time: 0:16:45
SGS - Recent average CPU time: 0:20:03

These times look inverted.
I see that single-threaded SGS tasks actually take around 83% of the CPU time of PPSE tasks.

Have recent FFT sizes changed sufficiently to outdate these CPU times?

Is the average CPU time of SGS tasks being unfairly weighted by multithreading?

Is the average CPU time of SGS tasks being weighted by the long tail of a few outliers with very slow CPUs?

EDIT - Another possibility:
Is SGS finding so many first primes that it tests a lot of second primes - and I just don't see this in my small sample of tasks?

Ravi Fernando
Project administrator
Volunteer tester
Project scientist
Send message
Joined: 21 Mar 19
Posts: 176
ID: 1108183
Credit: 10,207,198
RAC: 5,098
321 LLR Gold: Earned 500,000 credits (591,844)Cullen LLR Bronze: Earned 10,000 credits (82,217)ESP LLR Bronze: Earned 10,000 credits (16,570)Generalized Cullen/Woodall LLR Bronze: Earned 10,000 credits (12,551)PPS LLR Ruby: Earned 2,000,000 credits (3,098,521)PSP LLR Silver: Earned 100,000 credits (106,263)SoB LLR Silver: Earned 100,000 credits (258,849)SR5 LLR Bronze: Earned 10,000 credits (59,499)SGS LLR Silver: Earned 100,000 credits (148,878)TRP LLR Silver: Earned 100,000 credits (195,905)Woodall LLR Bronze: Earned 10,000 credits (40,424)321 Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,001,667)AP 26/27 Bronze: Earned 10,000 credits (72,774)GFN Gold: Earned 500,000 credits (502,872)WW Bronze: Earned 10,000 credits (12,000)
Message 149082 - Posted: 28 Feb 2021 | 21:14:01 UTC

I would assume this is because of TdP. Lots of heavy hitters are running PPSE to find a T5K prime, and due to the competition for firsts, this is done mostly with fast hardware. (On slower hardware, the 1st percent may be so low that PPS is better due to LLR2.) So PPSE is being run mostly on very fast CPUs right now, and SGS isn't.

Profile composite
Volunteer tester
Send message
Joined: 16 Feb 10
Posts: 841
ID: 55391
Credit: 784,056,583
RAC: 431,184
Discovered 2 mega primesFound 1 prime in the 2018 Tour de Primes321 LLR Turquoise: Earned 5,000,000 credits (5,477,467)Cullen LLR Gold: Earned 500,000 credits (776,297)ESP LLR Ruby: Earned 2,000,000 credits (3,433,680)Generalized Cullen/Woodall LLR Ruby: Earned 2,000,000 credits (2,093,491)PPS LLR Sapphire: Earned 20,000,000 credits (27,332,255)PSP LLR Turquoise: Earned 5,000,000 credits (6,587,988)SoB LLR Sapphire: Earned 20,000,000 credits (34,191,283)SR5 LLR Turquoise: Earned 5,000,000 credits (6,110,877)SGS LLR Ruby: Earned 2,000,000 credits (3,486,285)TRP LLR Turquoise: Earned 5,000,000 credits (7,070,795)Woodall LLR Amethyst: Earned 1,000,000 credits (1,693,614)321 Sieve (suspended) Emerald: Earned 50,000,000 credits (50,256,050)Cullen/Woodall Sieve (suspended) Turquoise: Earned 5,000,000 credits (5,571,178)Generalized Cullen/Woodall Sieve (suspended) Emerald: Earned 50,000,000 credits (50,009,610)PPS Sieve Double Silver: Earned 200,000,000 credits (385,656,505)Sierpinski (ESP/PSP/SoB) Sieve (suspended) Jade: Earned 10,000,000 credits (10,165,888)TRP Sieve (suspended) Sapphire: Earned 20,000,000 credits (20,071,454)AP 26/27 Turquoise: Earned 5,000,000 credits (6,616,128)GFN Emerald: Earned 50,000,000 credits (53,771,465)WW Gold: Earned 500,000 credits (932,000)PSA Double Bronze: Earned 100,000,000 credits (102,762,384)
Message 149083 - Posted: 28 Feb 2021 | 21:48:39 UTC - in response to Message 149082.
Last modified: 28 Feb 2021 | 21:49:41 UTC

I would assume this is because of TdP. Lots of heavy hitters are running PPSE to find a T5K prime, and due to the competition for firsts, this is done mostly with fast hardware. (On slower hardware, the 1st percent may be so low that PPS is better due to LLR2.) So PPSE is being run mostly on very fast CPUs right now, and SGS isn't.

You are suggesting then that the correct answer is:
"... the average CPU time of SGS tasks being weighted by ... slow CPUs ..."

And, in an effort to get more firsts, many people are applying multithreading to PPSE, especially on machines that don't have AVX-512. So it's the opposite situation of "... the average CPU time of SGS tasks ... weighted by multithreading" - because it's actually for PPSE tasks in which added CPU time is being expended in larger quantities for shorter turnaround time.

The point I'd like to make is that the average CPU time is technically accurate but not quite the right metric for the purpose at hand. People are trying to decide which subprojects to select using the "average CPU time" as guide to expected relative runtime, because we all know CPUs come in all kinds of speeds. The metric we should have is not the average CPU time, but the expected time for an "average CPU". Then at least the relative magnitudes of the subproject runtimes is correct. You can bet it raised an eyebrow or two when PPSE was seen to take 25% longer to compute than SGS instead of the other way around (as suggested by the CPU times on the preferences page).

Post to thread

Message boards : Number crunching : Recent average CPU time in project preferences is misleading

[Return to PrimeGrid main page]
DNS Powered by DNSEXIT.COM
Copyright © 2005 - 2021 Rytis Slatkevičius (contact) and PrimeGrid community. Server load 1.30, 2.11, 2.50
Generated 21 Sep 2021 | 20:13:57 UTC