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 : Generalized Fermat Prime Search : Have OCL1 reached any paramter limits?

Author Message
shinozkProject donor
Send message
Joined: 15 Sep 12
Posts: 20
ID: 172579
Credit: 3,276,927,336
RAC: 4,680,993
321 LLR Bronze: Earned 10,000 credits (73,438)PSP LLR Turquoise: Earned 5,000,000 credits (8,546,573)SoB LLR Double Silver: Earned 200,000,000 credits (346,491,920)SR5 LLR Ruby: Earned 2,000,000 credits (3,685,924)SGS LLR Silver: Earned 100,000 credits (102,618)TRP LLR Jade: Earned 10,000,000 credits (19,352,438)PPS Sieve Gold: Earned 500,000 credits (788,814)AP 26/27 Silver: Earned 100,000 credits (206,193)GFN Double Ruby: Earned 2,000,000,000 credits (2,851,568,577)WW Sapphire: Earned 20,000,000 credits (46,152,000)
Message 155187 - Posted: 24 Apr 2022 | 3:55:15 UTC

My GFN21 tasks have many maxErr recently.
maxErr exceeded for 801998^2097152+1, 0.5000 > 0.4500
(http://www.primegrid.com/result.php?resultid=1336857496)

I dont think it means a hardware failar, since all tasks with many maxErr dont became invalid.
It occors on the stock clock.

Wingmen tasks show that it seems occor only in OCL1 of GFN21.
The wingman tasks, Tesla, have the same errors.
http://www.primegrid.com/result.php?resultid=1281962873

Have OCL1 reached any paramter limits in GFN21, such as B limits?

Profile JeppeSNProject donor
Avatar
Send message
Joined: 5 Apr 14
Posts: 1731
ID: 306875
Credit: 46,031,372
RAC: 243,392
Found 1 prime in the 2020 Tour de Primes321 LLR Gold: Earned 500,000 credits (593,283)Cullen LLR Gold: Earned 500,000 credits (611,298)ESP LLR Silver: Earned 100,000 credits (174,818)Generalized Cullen/Woodall LLR Silver: Earned 100,000 credits (112,799)PPS LLR Jade: Earned 10,000,000 credits (17,086,131)PSP LLR Silver: Earned 100,000 credits (428,457)SoB LLR Silver: Earned 100,000 credits (466,812)SR5 LLR Silver: Earned 100,000 credits (210,142)SGS LLR Silver: Earned 100,000 credits (112,277)TRP LLR Silver: Earned 100,000 credits (342,501)Woodall LLR Silver: Earned 100,000 credits (109,455)321 Sieve (suspended) Silver: Earned 100,000 credits (175,037)PPS Sieve Bronze: Earned 10,000 credits (10,113)AP 26/27 Bronze: Earned 10,000 credits (12,129)GFN Ruby: Earned 2,000,000 credits (4,228,147)WW Jade: Earned 10,000,000 credits (13,744,000)PSA Turquoise: Earned 5,000,000 credits (7,614,290)
Message 155189 - Posted: 24 Apr 2022 | 8:53:24 UTC

Yes, I guess it is expected and normal.

You can see some "fuzzy" limits in the top post of the old (but pinned) thread Genefer "B" limits. For GFN21 and OCL, I see 660,000 in the table; your b is 801,998.

/JeppeSN

shinozkProject donor
Send message
Joined: 15 Sep 12
Posts: 20
ID: 172579
Credit: 3,276,927,336
RAC: 4,680,993
321 LLR Bronze: Earned 10,000 credits (73,438)PSP LLR Turquoise: Earned 5,000,000 credits (8,546,573)SoB LLR Double Silver: Earned 200,000,000 credits (346,491,920)SR5 LLR Ruby: Earned 2,000,000 credits (3,685,924)SGS LLR Silver: Earned 100,000 credits (102,618)TRP LLR Jade: Earned 10,000,000 credits (19,352,438)PPS Sieve Gold: Earned 500,000 credits (788,814)AP 26/27 Silver: Earned 100,000 credits (206,193)GFN Double Ruby: Earned 2,000,000,000 credits (2,851,568,577)WW Sapphire: Earned 20,000,000 credits (46,152,000)
Message 155192 - Posted: 24 Apr 2022 | 13:44:57 UTC

Thanks.

Yves Gallot
Volunteer developer
Project scientist
Send message
Joined: 19 Aug 12
Posts: 715
ID: 164101
Credit: 305,196,637
RAC: 1,681
GFN Double Silver: Earned 200,000,000 credits (305,196,637)
Message 155202 - Posted: 24 Apr 2022 | 21:29:32 UTC - in response to Message 155187.

My GFN21 tasks have many maxErr recently. maxErr exceeded for 801998^2097152+1, 0.5000 > 0.4500
I dont think it means a hardware failar, since all tasks with many maxErr dont became invalid.
It occors on the stock clock.
Have OCL1 reached any paramter limits in GFN21, such as B limits?

Yes, OCL is now "out of range" for GFN21. The "fuzzy" range was crossed and the error is now systematic. A new version of genefer will be generated to invalidate OCL if b >= 780000.
The same sort of errors occured with the CPU version that was updated recently.
This is the limit of the "standard" algorithm where b is a 64-bit flloat. Above this limit, a Number Theoretic Transform is the most efficient method on GPU and an Irrational Base Transform on CPU. These transforms are already used for n < 21.
Note that on GPU, NTT (OCL4 & OCL5) were already faster than OCL on Geforce GPUs. Then the error occurs only on Tesla and AMD GPUs.

When the GFN projects started on PrimeGrid in 2011, 6 GFN-18, 4 GFN-19 and no GFN-20 were found with the 64-bit flloat transform. The implementation of some new transforms was necessary to extend these lists. Today we know that no GFN-21 exists in the range that can be tested with the basic method. GFN-22 is the only project that can still use the 64-bit flloat transform.

Post to thread

Message boards : Generalized Fermat Prime Search : Have OCL1 reached any paramter limits?

[Return to PrimeGrid main page]
DNS Powered by DNSEXIT.COM
Copyright © 2005 - 2022 Rytis Slatkevičius (contact) and PrimeGrid community. Server load 1.40, 1.51, 1.64
Generated 25 Sep 2022 | 5:41:43 UTC