gpuOwL: an OpenCL program for Mersenne primality testing

[M344587487]

Damn the DC did use the wrong type, the worktodo was not wrong I guess gpuowl can only do type 4. Can anyone point me in the direction of a DC with type 4 exponent to grab? Don;t think the DB can be searched by type and the recent results are all type 5 or 1.

[kriesel]

Originally Posted by M344587487

Damn the DC did use the wrong type, the worktodo was not wrong I guess gpuowl can only do type 4. Can anyone point me in the direction of a DC with type 4 exponent to grab? Don;t think the DB can be searched by type and the recent results are all type 5 or 1.

The PRP residue type depends on the version of gpuowl you run. To verify quickly you have the right version of gpuowl for the desired residue type, run a small exponent such as 1257787. For example, V3.8 gpuowl does type 1 PRP residue. Exponent details at mersenne.org will show what residue type the first run was done with; https://www.mersenne.org/report_expo...4100000&full=1.
Additionally verify the shift was not zero. Gpuowl does zero shift only, so is not a good DC for a zero shift first test, and a triple check would follow. See also http://www.mersenne.ca/exponent/84004321 which does not indicate residue type there.
Zero shift first PRP tests should be DC by prime95 not gpuowl, to ensure a nonzero DC. See 84006781.

[M344587487]

Thanks, I was lazy researching but would never have figured to use v3.8.

[kriesel]

Originally Posted by M344587487

Thanks, I was lazy researching but would never have figured to use v3.8.

If you want to qualify your modification or gpu, you could try
https://www.mersenne.org/report_expo...4102691&full=1. Since I ran it, I have a list of interim residues a short run could be checked against.

It was with the wrong set of system software, not enough up to date, gpuowl started but after a while the residue was zeroed out, and a page fault occurred.


After installing the latest kernel, modules and headers, v. 5.0, then gpuowl works without problems and the zero-residue error is gone.

Originally Posted by SELROC

It was with the wrong set of system software, not enough up to date, gpuowl started but after a while the residue was zeroed out, and a page fault occurred.


After installing the latest kernel, modules and headers, v. 5.0, then gpuowl works without problems and the zero-residue error is gone.

The same error has occurred again, zero-residue, but this time gpuowl reloaded the last checkpoint and continued.


I found that I was running a version of openowl compiled against old libs. Now I have recompiled and waiting to see if the error occurs again.

[kriesel]

Originally Posted by kriesel

If you want to qualify your modification or gpu, you could try
https://www.mersenne.org/report_expo...4102691&full=1. Since I ran it, I have a list of interim residues a short run could be checked against.

There's also a table of interim residues for powers of ten iterations for known Mersenne primes, in the attachment of post 5 at https://www.mersenneforum.org/showthread.php?t=24003

[kriesel]

I've extended the third attachment at https://www.mersenneforum.org/showpo...35&postcount=2 to include a plot of the per iteration times on an RX480 gpu for whichever version of gpuowl tested fastest for each fft length. An approximation of the asymptotic scaling of gpuowl is p^1.078 for ms/iter, p^2.078 for exponent run time, from the trend line fit for fft lengths 6M to 144M (exponents 100M<p<2520M). This is close to values obtained for CUDALucas and prime95, ~2.094, and finally provides for gpuowl a scaling above p², which is more consistent with expectations of scaling n² log n log log n from such reliable sources as Knuth.

Originally Posted by SELROC

The same error has occurred again, zero-residue, but this time gpuowl reloaded the last checkpoint and continued.


I found that I was running a version of openowl compiled against old libs. Now I have recompiled and waiting to see if the error occurs again.

After some hour no error. Will see today the continuation.

[kracker]

Originally Posted by M344587487

Damn the DC did use the wrong type, the worktodo was not wrong I guess gpuowl can only do type 4. Can anyone point me in the direction of a DC with type 4 exponent to grab? Don;t think the DB can be searched by type and the recent results are all type 5 or 1.

^ TIL... Is there any way at all to lookup exponents that have been done with a specific type?

[GP2]

Originally Posted by M344587487

Damn the DC did use the wrong type, the worktodo was not wrong I guess gpuowl can only do type 4. Can anyone point me in the direction of a DC with type 4 exponent to grab? Don;t think the DB can be searched by type and the recent results are all type 5 or 1.

You can find some PRPs that need DC from the following users:

Warning: not all are type 4. There's no way to filter by residue type, although you can click on the "Residue Type" column header to sort by it.

Also, I think gpuOwL only produces residues with shift count zero, so the double check will also have shift count zero, and some might insist that it's not a proper double check unless it's with a different shift count.

• Mihai Preda
• Kriesel
• kwe5ykdf
• tServo
• Franklin Webber
• Xebecer
• xx005fs
• SEL-ROC
• kracker