Is this possible at all?
 

From the Recent Cleared Report:

ramgeis Manual testing [B]999999937[/B] C 2018-09-10 09:52 [B]1843.9[/B] 50800.332 4307529C0B5EB6__

A few thoughts:

1. I am under the impression that testing above ~500M is not yet supported by the client(s) used. Even if it is currently possible, this test was started more than 5 years ago, and at that time I´m nearly 100% sure it was not the case.

2. Assuming it is / was possible, shall we have any confidence that a test that ran for 1843.9 days is to be trusted? And no, I will not volunteer to run the DC... LOL.

I am curious about the story of this test, but oh well, that´s probably only me... :rolleyes:

The longest assignment (as of the end of last year) was the LL test of [M]M332197309[/M] by Lelikoid, which took 3197.2 days (that's 8 ¾ years).

And yeah, it's marked "Suspect"...

[QUOTE=lycorn;495843]From the Recent Cleared Report:

ramgeis Manual testing [B]999999937[/B] C 2018-09-10 09:52 [B]1843.9[/B] 50800.332 4307529C0B5EB6__

A few thoughts:

1. I am under the impression that testing above ~500M is not yet supported by the client(s) used. Even if it is currently possible, this test was started more than 5 years ago, and at that time I´m nearly 100% sure it was not the case.

2. Assuming it is / was possible, shall we have any confidence that a test that ran for 1843.9 days is to be trusted? And no, I will not volunteer to run the DC... LOL.

I am curious about the story of this test, but oh well, that´s probably only me... :rolleyes:[/QUOTE]

All I can add is that Prime95 is limited to a little under 600M.
However, I believe there is other GPU or CUDA based software that does not have this limit.

[QUOTE=petrw1;495845]All I can add is that Prime95 is limited to a little under 600M.[/QUOTE]

True. I did mean to write "600M", it was a typo.

[QUOTE=petrw1;495845]However, I believe there is other GPU or CUDA based software that does not have this limit.[/QUOTE]

That´s possible, but were those clients available in summer 2013?

[QUOTE=lycorn;495846]That´s possible, but were those clients available in summer 2013?[/QUOTE]

I don't know about CUDA but searching for chalsall in the GPUto72 forum brings up threads for that GPU project that start on 2011 12-01

Since version 29.2 there have been 50M FFT for FMA only:
[url]https://mersenneforum.org/showpost.php?p=460914&postcount=1[/url]

but that should only be enough for 930M-940M at the most, and that version only came out 1 year and 3 months ago.

CUDALucas has up to 64M FFT and Mlucas at least 64M, but then it should have been a manual turn in.


Edit: Oh it was a manuel test, but with CUDALucas it would not have the information about number of days it took.

[QUOTE=science_man_88;495850]I don't know about CUDA but searching for chalsall in the GPUto72 forum brings up threads for that GPU project that start on 2011 12-01[/QUOTE]

Sigh... Your point being?

[QUOTE=lycorn;495843]From the Recent Cleared Report:

ramgeis Manual testing [B]999999937[/B] C 2018-09-10 09:52 [B]1843.9[/B] 50800.332 4307529C0B5EB6__

A few thoughts:

1. I am under the impression that testing above ~500M is not yet supported by the client(s) used. Even if it is currently possible, this test was started more than 5 years ago, and at that time I´m nearly 100% sure it was not the case.

2. Assuming it is / was possible, shall we have any confidence that a test that ran for 1843.9 days is to be trusted? And no, I will not volunteer to run the DC... LOL.

I am curious about the story of this test, but oh well, that´s probably only me... :rolleyes:[/QUOTE]

Hi, I'm happy to answer your implicit questions. ;-)

- the software used for the test was CUDALucas v2.06beta
- although I reserved the exponent (mainly for for doing idle TF work) 5 years ago I started with the LL test about half a year ago when the hardware to do it in a reasonable time was available (Nvidia GV100)
- the FFT size used was 57344K, roughly 16 ms/iteration
- I kept some intermediate save files in case someone is interested ;-)

[QUOTE=ramgeis;495854]Hi, I'm happy to answer your implicit questions. ;-)[/QUOTE]

Cool. Why did you run a LL test that high?

Some additional information:


- beforehand I did a couple of DCs to check the hardware/software combination

- the ROE reported was all the time beween 0.19 and 0.23, mostly in the 0.21 area
- at the moment I'm running a couple of DCs to check the current hardware situation

[QUOTE=chalsall;495857]Cool. Why did you run a LL test that high?[/QUOTE]

Because it seemed to be possible, so I had to try it out. :-)