GpuOwl 7.x

[preda]

Originally Posted by frmky

Using ./gpuowl -prp 6972649 -b1 1500000 -b2 10000000 -maxAlloc 4.5G

Thanks! I pushed a fix, should be fine now. The problem was affecting P2 where the number of buffers was > 1024, which was the case for this small exponent.

[frmky]

Yes, that fixed it.

2020-10-18 12:36:19 Tesla K20c-0 6972649 P2(1.5M,10M) Setup 1024 P2 buffers in 3.7s
...
2020-10-18 12:38:35 Tesla K20c-0 6972649 GCD: 11352967935560261207067499316137

[Viliam Furik]

Now with the new version, removing the standalone P-1, could you fulfil last wish from version 6.x, that being, implement the functionality, that when you write known factor(s) in the worktodo line, it will not stop P-1 when finding the already known factor(s) after the first stage? I don't know as much about C and OpenCL as you, but I think you will agree, it can be solved with a simple condition and some extension of parsing script. (If the factor found in P1 consists of only the known factor(s), ignore it, and let the P2 run - or something similar)

And to the new version, do you plan to make PRP-CF available?

[preda]

Originally Posted by Viliam Furik

Now with the new version, removing the standalone P-1, could you fulfil last wish from version 6.x, that being, implement the functionality, that when you write known factor(s) in the worktodo line, it will not stop P-1 when finding the already known factor(s) after the first stage? I don't know as much about C and OpenCL as you, but I think you will agree, it can be solved with a simple condition and some extension of parsing script. (If the factor found in P1 consists of only the known factor(s), ignore it, and let the P2 run - or something similar)

And to the new version, do you plan to make PRP-CF available?

No plans for PRP-CF. I don't have a good understanding of that need myself, and I think that need is adequately covered by mprime.

Could you explain a bit more about what you're trying to do by not stopping P-1 when it finds a factor?

About accepting a factor in the worktodo line, that was the case before (the factor was completly ignored), but it turned out that was a problem because some users assumed that PRP-CF must be supported because the worktodo line with a factor is not rejected. Now it's rejected for hopefully less confusion.

[Viliam Furik]

Originally Posted by preda

No plans for PRP-CF. I don't have a good understanding of that need myself, and I think that need is adequately covered by mprime.

Well, but the mprime or Prime95 can't run on Radeon VII or any other GPU whatsoever.

Originally Posted by preda

Could you explain a bit more about what you're trying to do by not stopping P-1 when it finds a factor?

Simply the same mechanics as in Prime95/mprime. When you create a line in worktodo.txt for P-1, and you include the known factors at the end like this -> Pminus1=AID,1,2,prime,-1,99,0,"factor,factor" , it will ignore the factors if they are found. gpuOwl 6.x doesn't accept the factors, so it will always stop after first stage if a factor is found.

[Viliam Furik]

Originally Posted by preda

About accepting a factor in the worktodo line, that was the case before (the factor was completly ignored), but it turned out that was a problem because some users assumed that PRP-CF must be supported because the worktodo line with a factor is not rejected. Now it's rejected for hopefully less confusion.

In what version was it available? I might just use it if I find the right .exe file.

[preda]

Originally Posted by Viliam Furik

In what version was it available? I might just use it if I find the right .exe file.

I think any v6.x accepts factors at the end of the assignment, and completely *ignores* them. Does not read them, does not handle them in any way, as if they weren't there. Does not continue P-1 past them, P-1 works in the same way as if the factors were not there. Not very useful IMO.

[Viliam Furik]

Originally Posted by preda

I think any v6.x accepts factors at the end of the assignment, and completely *ignores* them. Does not read them, does not handle them in any way, as if they weren't there. Does not continue P-1 past them, P-1 works in the same way as if the factors were not there. Not very useful IMO.

What is the specific syntax needed? Prime95 style is not working.

[preda]

Originally Posted by Viliam Furik

What is the specific syntax needed? Prime95 style is not working.

I don't know, maybe only works for PRP. Anyway I consider that (not rejecting factors at the end) a bug.

[kriesel]

Originally Posted by Viliam Furik

What is the specific syntax needed? Prime95 style is not working.

None. Gpuowl has never supported PRP cofactor tests, or known factors in PRP or P-1. V6.11 ignored the additional input, including nonsensical extra input, in a test on PRP.
https://www.mersenneforum.org/showth...tor#post558573
https://www.mersenneforum.org/showth...tor#post556524
Gpuowl does a lot, but not everything that mprime/prime95 do, or that we users wish it would do.

[preda]

In a recent commit (tagged v7.1), GpuOwl switched to "Proof v2".

This is a change in the way the proof is generated -- slightly different residues are used. Not much changes client-side, except that the transition between proof versions *can't* be done half-way through the PRP test (because: different residues are used for the two versions).

So finish the PRP test with the old version, and start a new one with the new version.

The "Proof v2" is already accepted (and preferred) by the server, and is the same format Prime95 uses.