Rechecking CUDALucas-only LL results

kriesel · 2022-11-14, 23:39

These are the 74 exponents returned by George's database query for exponents verified using CUDALucas only. They are candidates for independent verification of seed 4 final LL res64 only, using different software; gpuowl, prime95, or mlucas. Perhaps even cllucas, although I would avoid it due to speed and reliability and mainly because it was derived directly from CUDALucas.
Note, these can not be registered with the PrimeNet server.
I interpret ATH's posts as he intended to rerun one, has done so, and the rest are fair game for others, no toes mashed.

So far, 73 matched, no mismatches yet, 1 to go as of mid 2022-12-04 (US CST), of which 1 is not yet queued. Good progress and encouraging results!
I've added [CUDALucas original fft-length estimate] markers based on a pseudorandomly selected fft file from a Quadro 2000 CUDALucas benchmarking run from long ago. [length] divides those preceding, that could have used it, from those following, for which it is expected to be too small. Units are K words. This is only an estimate of how many and which lengths the preceding CUDALucas runs might have involved, since different GPU models benchmarking often select different fft lengths to record in the fft file, or fastest fft lengths may be nonreproducible from one benchmark run to another on the same GPU, or users may force specific lengths. We don't have data on which GPU models were used on specific previous CUDALucas LL runs. Following the length is (estimated count). The point is to estimate coverage of CUDALucas fft lengths.
I count 17 populated estimated lengths (out of 88 in the applicable range in the selected fft benchmark file), of which 16 have at least one completed matched result below, and 16 have all completed. Note 20000K and all other fft lengths suitable for 100Mdigit Mersennes are absent from the list below.

When claiming or updating one or more for this additional checking with anything except CUDALucas, please state which exponent(s), and for each, what application(s) will be used, including what version(s), optionally current status(es) e.g. queued or running, and optionally what ETA(s) when applicable.
Color code below: purple reference; underlined blue links to mersenne.org exponent status, black undetermined, green matched, red mismatch (if any).

[1792] (0)

34643591 - kriesel gpuowl v6.11-380 matched
34696567 - kriesel gpuowl v6.11-380 matched
35184673 - kriesel gpuowl v6.11-380 matched
35381377 - sdbardwick prime95 v30.8b17 matched ("Torchwood Institute")
35478853 - sdbardwick prime95 v30.8b17 matched
[1890] (5)

36142801 - sdbardwick prime95 v30.8b17 matched
36211067 - sdbardwick prime95 v30.8b17 matched
36313813 - sdbardwick prime95 v30.8b17 matched
36497473 - sdbardwick prime95 v30.8b17 matched
36532159 - ATH prime95 v30.8b17 matched
[1944] (5)

36717713 - LaurV gpuOwl v6.11-380 matched
36841111 - LaurV gpuOwl v6.11-380 matched
37018711 - LaurV gpuOwl v6.11-380 matched
37047167 - LaurV gpuOwl v6.11-380 matched
38093491 - LaurV gpuOwl v6.11-380 matched
38208713 - LaurV gpuOwl v6.11-380 matched
38276081 - LaurV gpuOwl v6.11-380 matched
38363993 - LaurV gpuOwl v6.11-380 matched
[2048] (8)

38931791 - LaurV gpuOwl v6.11-380 matched
38976211 - LaurV gpuOwl v6.11-380 matched
38976221 - LaurV gpuOwl v6.11-380 matched
39052267 - LaurV gpuOwl v6.11-380 matched
39258293 - LaurV gpuOwl v6.11-380 matched
39839603 - LaurV gpuOwl v6.11-380 matched
40123351 - kriesel gpuowl v6.11-382 matched
40404289 - kriesel gpuowl v6.11-382 matched
40413371 - kriesel gpuowl v6.11-382 matched
40473841 - kriesel gpuowl v6.11-382 matched
40501819 - kriesel gpuowl v6.11-382 matched
[2160] (11)

40641659 - kriesel gpuowl v6.11-382 matched
[2187] (1)

41508253 - kriesel prime95 v30.8b15 matched
41518229 - kriesel gpuowl v6.11-380 matched
41856721 - kriesel gpuowl v6.11-380 matched
42791519 - kriesel gpuowl v6.11-380 matched
[2304] (4)

43883923 - kriesel mlucas v20.1.1 matched
44932729 - kriesel gpuowl v6.11-380 matched
45243557 - kriesel prime95 v30.8b15 matched
45285043 - kriesel prime95 v30.8b15 matched
[2520] (4)

48073099 - kriesel gpuowl v6.11-380 matched
48075583 - kriesel gpuowl v6.11-382 matched
48122471 - kriesel prime95 v30.8b14 matched
48429497 - kriesel prime95 v30.7b9 matched
[2592] (4)

48555343 - kriesel prime95 v30.8b17 matched
48677777 - kriesel prime95 v30.8b17 matched
49404263 - kriesel gpuowl v6.11-382 matched
49457687 - kriesel prime95 v30.7b9 matched
[2646] (4)

53998811 - kriesel prime95 v30.8b15 matched
54009271 - kriesel prime95 v30.8b17 matched
54010013 - kriesel prime95 v30.8b14 matched
55831921 - kruoli gpuowl v6.11-380 matched
56294479 - kruoli gpuowl v6.11-380 matched
56309111 - kruoli gpuowl v6.11-380 matched
[3024] (6)

57766307 - kruoli gpuowl v6.11-380 matched
57954781 - kriesel prime95 v30.8b15 matched
58370549 - kriesel gpuowl v6.11-382 matched
58370563 - kriesel gpuowl v6.11-382 matched
[3136] (4)

72366587 - kriesel gpuowl v6.11-382 matched
73604719 - kriesel mlucas v20.1.1 matched
73612841 - LaurV gpuOwl v6.11-380 matched
73614041 - LaurV gpuOwl v6.11-380 matched
73642033 - LaurV gpuOwl v6.11-380 matched
73684073 - LaurV gpuOwl v6.11-380 matched
73684703 - LaurV gpuOwl v6.11-380 matched
73685609 - LaurV gpuOwl v6.11-380 matched
73798027 - LaurV gpuOwl v6.11-380 matched
73802059 - LaurV gpuOwl v6.11-380 matched
73812071 - LaurV gpuOwl v6.11-380 matched
73901071 - LaurV gpuOwl v6.11-380 matched
[4032] (12)

77075387 - kriesel gpuowl v6.11-382 matched
77143147 - kriesel prime95 v30.8b14 matched
[4320] (2)

88680457 - kriesel gpuowl v6.11-380 matched
[5040] (1)

132000191 - kriesel gpuowl v6.11-380 matched
[7168] (1)

137362691 - LaurV gpuOwl v6.11-380 matched
[7776] (1)

666666667 -
[37800] (1)

sdbardwick · 2022-11-15, 00:34

I'll do these with P95 30.8b17:
35381377
35478853
36142801
36211067
36313813
36497473
ETA 3 days or so for all.
Finally, something my i9-11900 has something to do!

LaurV · 2022-11-15, 03:52

Everything under 40M, for a start.
gpuOwl 6 (same one compiled by kriesel) Radeon vii, win 7, 2 cards should go through them like cheese.
36717713
36841111
37018711
37047167
38093491
38208713
38276081
38363993
38931791
38976211
38976221
39052267
39258293
39839603

p.s. modified the initial post to save you the work

, I also added green for the matched to be easy to see

LaurV · 2022-11-15, 07:41

[Trivia]
As you were kind enough to provide proper links, I played along by clicking on each one

you know that saying that every link must be clicked...
What's puzzling here, almost all of them (with only 2-3 exceptions, which were recently tested) have a "bad residue" test sometimes in 2007-2008 (few in or after 2011).
I mean, all, not only those I reserved.
A lot of wasted computing power, at the time...
[/Trivia]

edit (kriesel): Yes, only one below 57M did not have a bad first LL. Many of those are from 2010.
Your link clicking was redundant in a sense. As a former webmaster I am in the habit of testing all links posted. The links were however created for the convenience of all recheck participants or the merely curious.

ATH · 2022-11-15, 17:38

I don't really think it is needed to check all these with the shift count enabled, and I do not have much resources now that I stopped my Colab accounts.
If there is a bug it is probably only in the 20000K FFT.

If you find something in those you already queued now, I can take a few.

kruoli · 2022-11-15, 19:37

If possible, I would like to take
55831921
56294479
56309111
57766307
for PRP. Or is LL preferred here?

kriesel · 2022-11-15, 20:00

Quote:

Originally Posted by kruoli

If possible, I would like to take
55831921
56294479
56309111
57766307
for PRP. Or is LL preferred here?

Definitely LL only, since it is the CUDALucas LL final res64 we are attempting to confirm with other software, as an attempt to identify possible fft-length-dependent CUDALucas bugs. CUDALucas is LL only. I've added you to the list in post one for those. What software will you be using?

edit: PRP would only provide confirmation the LL final residue should not be zero, but provide no information about what the LL res64 should be beyond that.

kruoli · 2022-11-15, 20:15

Good to now.

I will use GPUowl, and if those do not match, I will queue them on another machine with mprime/Prime95.

Thank you for the clarification!

kriesel · 2022-11-15, 20:49

Quote:

Originally Posted by kruoli

I will use GPUowl

which version, v6.11-380?

kruoli · 2022-11-15, 21:42

Right now I have not yet a fixed version. Do you recommend this version for this project?

kriesel · 2022-11-15, 21:50

Quote:

Originally Posted by kruoli

Right now I have not yet a fixed version. Do you recommend this version for this project?

Gpuowl V6.11-380 or some versions near it support LL with Jacobi symbol check and are about as fast as available. Gpuowl V7.x omits LL so is not usable for this.

2022-11-14, 23:39	#1
kriesel "TF79LL86GIMPS96gpu17" Mar 2017 US midwest 2²×3×643 Posts	Rechecking CUDALucas-only LL results These are the 74 exponents returned by George's database query for exponents verified using CUDALucas only. They are candidates for independent verification of seed 4 final LL res64 only, using different software; gpuowl, prime95, or mlucas. Perhaps even cllucas, although I would avoid it due to speed and reliability and mainly because it was derived directly from CUDALucas. Note, these can not be registered with the PrimeNet server. I interpret ATH's posts as he intended to rerun one, has done so, and the rest are fair game for others, no toes mashed. So far, 73 matched, no mismatches yet, 1 to go as of mid 2022-12-04 (US CST), of which 1 is not yet queued. Good progress and encouraging results! I've added [CUDALucas original fft-length estimate] markers based on a pseudorandomly selected fft file from a Quadro 2000 CUDALucas benchmarking run from long ago. [length] divides those preceding, that could have used it, from those following, for which it is expected to be too small. Units are K words. This is only an estimate of how many and which lengths the preceding CUDALucas runs might have involved, since different GPU models benchmarking often select different fft lengths to record in the fft file, or fastest fft lengths may be nonreproducible from one benchmark run to another on the same GPU, or users may force specific lengths. We don't have data on which GPU models were used on specific previous CUDALucas LL runs. Following the length is (estimated count). The point is to estimate coverage of CUDALucas fft lengths. I count 17 populated estimated lengths (out of 88 in the applicable range in the selected fft benchmark file), of which 16 have at least one completed matched result below, and 16 have all completed. Note 20000K and all other fft lengths suitable for 100Mdigit Mersennes are absent from the list below. When claiming or updating one or more for this additional checking with anything except CUDALucas, please state which exponent(s), and for each, what application(s) will be used, including what version(s), optionally current status(es) e.g. queued or running, and optionally what ETA(s) when applicable. Color code below: purple reference; underlined blue links to mersenne.org exponent status, black undetermined, green matched, red mismatch (if any). [1792] (0) 34643591 - kriesel gpuowl v6.11-380 matched 34696567 - kriesel gpuowl v6.11-380 matched 35184673 - kriesel gpuowl v6.11-380 matched 35381377 - sdbardwick prime95 v30.8b17 matched ("Torchwood Institute") 35478853 - sdbardwick prime95 v30.8b17 matched [1890] (5) 36142801 - sdbardwick prime95 v30.8b17 matched 36211067 - sdbardwick prime95 v30.8b17 matched 36313813 - sdbardwick prime95 v30.8b17 matched 36497473 - sdbardwick prime95 v30.8b17 matched 36532159 - ATH prime95 v30.8b17 matched [1944] (5) 36717713 - LaurV gpuOwl v6.11-380 matched 36841111 - LaurV gpuOwl v6.11-380 matched 37018711 - LaurV gpuOwl v6.11-380 matched 37047167 - LaurV gpuOwl v6.11-380 matched 38093491 - LaurV gpuOwl v6.11-380 matched 38208713 - LaurV gpuOwl v6.11-380 matched 38276081 - LaurV gpuOwl v6.11-380 matched 38363993 - LaurV gpuOwl v6.11-380 matched [2048] (8) 38931791 - LaurV gpuOwl v6.11-380 matched 38976211 - LaurV gpuOwl v6.11-380 matched 38976221 - LaurV gpuOwl v6.11-380 matched 39052267 - LaurV gpuOwl v6.11-380 matched 39258293 - LaurV gpuOwl v6.11-380 matched 39839603 - LaurV gpuOwl v6.11-380 matched 40123351 - kriesel gpuowl v6.11-382 matched 40404289 - kriesel gpuowl v6.11-382 matched 40413371 - kriesel gpuowl v6.11-382 matched 40473841 - kriesel gpuowl v6.11-382 matched 40501819 - kriesel gpuowl v6.11-382 matched [2160] (11) 40641659 - kriesel gpuowl v6.11-382 matched [2187] (1) 41508253 - kriesel prime95 v30.8b15 matched 41518229 - kriesel gpuowl v6.11-380 matched 41856721 - kriesel gpuowl v6.11-380 matched 42791519 - kriesel gpuowl v6.11-380 matched [2304] (4) 43883923 - kriesel mlucas v20.1.1 matched 44932729 - kriesel gpuowl v6.11-380 matched 45243557 - kriesel prime95 v30.8b15 matched 45285043 - kriesel prime95 v30.8b15 matched [2520] (4) 48073099 - kriesel gpuowl v6.11-380 matched 48075583 - kriesel gpuowl v6.11-382 matched 48122471 - kriesel prime95 v30.8b14 matched 48429497 - kriesel prime95 v30.7b9 matched [2592] (4) 48555343 - kriesel prime95 v30.8b17 matched 48677777 - kriesel prime95 v30.8b17 matched 49404263 - kriesel gpuowl v6.11-382 matched 49457687 - kriesel prime95 v30.7b9 matched [2646] (4) 53998811 - kriesel prime95 v30.8b15 matched 54009271 - kriesel prime95 v30.8b17 matched 54010013 - kriesel prime95 v30.8b14 matched 55831921 - kruoli gpuowl v6.11-380 matched 56294479 - kruoli gpuowl v6.11-380 matched 56309111 - kruoli gpuowl v6.11-380 matched [3024] (6) 57766307 - kruoli gpuowl v6.11-380 matched 57954781 - kriesel prime95 v30.8b15 matched 58370549 - kriesel gpuowl v6.11-382 matched 58370563 - kriesel gpuowl v6.11-382 matched [3136] (4) 72366587 - kriesel gpuowl v6.11-382 matched 73604719 - kriesel mlucas v20.1.1 matched 73612841 - LaurV gpuOwl v6.11-380 matched 73614041 - LaurV gpuOwl v6.11-380 matched 73642033 - LaurV gpuOwl v6.11-380 matched 73684073 - LaurV gpuOwl v6.11-380 matched 73684703 - LaurV gpuOwl v6.11-380 matched 73685609 - LaurV gpuOwl v6.11-380 matched 73798027 - LaurV gpuOwl v6.11-380 matched 73802059 - LaurV gpuOwl v6.11-380 matched 73812071 - LaurV gpuOwl v6.11-380 matched 73901071 - LaurV gpuOwl v6.11-380 matched [4032] (12) 77075387 - kriesel gpuowl v6.11-382 matched 77143147 - kriesel prime95 v30.8b14 matched [4320] (2) 88680457 - kriesel gpuowl v6.11-380 matched [5040] (1) 132000191 - kriesel gpuowl v6.11-380 matched [7168] (1) 137362691 - LaurV gpuOwl v6.11-380 matched [7776] (1) 666666667 - [37800] (1) Last fiddled with by kriesel on 2022-12-04 at 16:52 Reason: status updates

2022-11-15, 00:34	#2
sdbardwick Aug 2002 North San Diego Coun 2⁴×3×17 Posts	I'll do these with P95 30.8b17: 35381377 35478853 36142801 36211067 36313813 36497473 ETA 3 days or so for all. Finally, something my i9-11900 has something to do! Last fiddled with by sdbardwick on 2022-11-15 at 00:36 Reason: Software info

2022-11-15, 03:52	#3
LaurV Romulan Interpreter "name field" Jun 2011 Thailand 2⁴×643 Posts	Everything under 40M, for a start. gpuOwl 6 (same one compiled by kriesel) Radeon vii, win 7, 2 cards should go through them like cheese. 36717713 36841111 37018711 37047167 38093491 38208713 38276081 38363993 38931791 38976211 38976221 39052267 39258293 39839603 p.s. modified the initial post to save you the work , I also added green for the matched to be easy to see Last fiddled with by LaurV on 2022-11-15 at 03:55

2022-11-15, 07:41	#4
LaurV Romulan Interpreter "name field" Jun 2011 Thailand 2830₁₆ Posts	[Trivia] As you were kind enough to provide proper links, I played along by clicking on each one you know that saying that every link must be clicked... What's puzzling here, almost all of them (with only 2-3 exceptions, which were recently tested) have a "bad residue" test sometimes in 2007-2008 (few in or after 2011). I mean, all, not only those I reserved. A lot of wasted computing power, at the time... [/Trivia] edit (kriesel): Yes, only one below 57M did not have a bad first LL. Many of those are from 2010. Your link clicking was redundant in a sense. As a former webmaster I am in the habit of testing all links posted. The links were however created for the convenience of all recheck participants or the merely curious. Last fiddled with by kriesel on 2022-11-24 at 11:23

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
inconsistencies between results.txt and results.json.txt	ixfd64	Software	3	2021-07-17 08:32
Statistical properties of categories of GIMPS results and interim results	kriesel	Probability & Probabilistic Number Theory	1	2019-05-22 22:59
Don't DC/LL them with CudaLucas	LaurV	Data	131	2017-05-02 18:41
settings for cudaLucas	fairsky	GPU Computing	11	2013-11-03 02:08
How to: send results back after using CUDALucas	Koyaanisqatsi	Information & Answers	5	2013-02-06 10:08

2022-11-15, 17:38	#5
ATH Einyen Dec 2003 Denmark 110101111010₂ Posts	I don't really think it is needed to check all these with the shift count enabled, and I do not have much resources now that I stopped my Colab accounts. If there is a bug it is probably only in the 20000K FFT. If you find something in those you already queued now, I can take a few.

2022-11-15, 19:37	#6
kruoli "Oliver" Sep 2017 Porta Westfalica, DE 1557₁₀ Posts	If possible, I would like to take 55831921 56294479 56309111 57766307 for PRP. Or is LL preferred here?

2022-11-15, 20:15	#8
kruoli "Oliver" Sep 2017 Porta Westfalica, DE 3²×173 Posts	Good to now. I will use GPUowl, and if those do not match, I will queue them on another machine with mprime/Prime95. Thank you for the clarification!

2022-11-15, 21:42	#10
kruoli "Oliver" Sep 2017 Porta Westfalica, DE 3²×173 Posts	Right now I have not yet a fixed version. Do you recommend this version for this project?