Can someone produce a binary of gpuowl?

[drkirkby]

I can't seem to compile this on CentOs 7.8, despite using the latest version of gcc and GMP.



Can anyone produce a binary, and some sort of test code that I could test?



Dave

[drkirkby]

I should have added - I need it for Linux.

[M344587487]

Here's the latest git compiled on a machine with AVX2 (not sure if it matters, probably not), you can test roughly with something like this or any large enough integer:

./gpuowl -prp 17323111

[drkirkby]

Thank you. Unfortuately, it needs a glibc later than I have installed on my CentOS 7.8 system. An attempt to install a later version of glibc in /usr/local, then set LD_LIBRARY_PATH fails. In fact during the configure process I had to set an option to disable sanity checks, as ones advised against installing glibc in /usr/local. If I installed in /usr/, I'm likely to break the OS, so I will leave it.

[M344587487]

Here's a few attempts to statically link libc, they all work locally and produce different binaries so the options did "something". It's a stab in the dark as I'm not familiar with static linking and am not set up to properly test. Worth a shot at least.

[Xyzzy]

Originally Posted by drkirkby

…CentOs 7.8…

https://arstechnica.com/gadgets/2021...ction-servers/

[drkirkby]

I tried compiling gpuowl on my PC, which runs CentOS 7.9, but I concluded the OS needed too many changes to make it practical. In particular, it seems glibc would need updating, which is a risky thing to do.

That aside, are GPUs able to do PRP tests quickly? I've currently got a couple of 26-core CPUs. A 110 million exponent PRP test takes two days, using one worker, so I can get about 1 PRP test done per day on average. At the moment I'm not achieving that, as I'm doing a huge exponent,

https://www.mersenne.org/report_expo...33&full=1which would have 100 million decimal digits. But that's the first and last time I will attempt that. I have considered giving up before, but with more than 90% of the PRP test complete, it is too late to give up now.

[M344587487]

Modern GPU's are the fastest way to do a PRP test, using gpuowl. A Radeon VII is something like an order of magnitude faster than a consumer CPU with dual channel memory. Did you try the few attempts I made at a static build of gpuowl on your other thread?

[Xyzzy]

Originally Posted by drkirkby

I tried compiling gpuowl on my PC, which runs CentOS 7.9, but I concluded the OS needed too many changes to make it practical.

https://developers.redhat.com/blog/2...-subscription/

[retina]

Originally Posted by Xyzzy

https://developers.redhat.com/blog/2...-subscription/

Need to register / activate / login to run the OS? I think that turns the computer into yet another IoT device. I would be worried if that was my system.

[drkirkby]

Originally Posted by M344587487

Here's a few attempts to statically link libc, they all work locally and produce different binaries so the options did "something". It's a stab in the dark as I'm not familiar with static linking and am not set up to properly test. Worth a shot at least.

Thank you. I had not seen your updates to this thread, until you mentioned it in another thread. Unfortunately, none of them work - see error messages at the end of the thread. A later version of glibc is needed, and it's not practical to upgrade that. Even if one builds a local copy, LD_LIBRARY_PATH etc will not allow one to use another glibc. In fact, when I tried to compile glibc, it reported that it should not be installed in /usr/local, and I had to add an option to bypass sanity checks to even allow me to build it. But I can't get the binary to use it.

I will soon be ditching CentOS in favor of Ubuntu or similar. I used CentOS as it was required for a trial of some very expensive commercial software. The trial was an absolute disaster, with the program crashing every 5 minutes.

[dkirkby@jackdaw gpuowl-static]$ ./gpuowla
./gpuowla: /lib64/libquadmath.so.0: version `QUADMATH_1.2' not found (required by ./gpuowla)
./gpuowla: /lib64/libm.so.6: version `GLIBC_2.29' not found (required by ./gpuowla)
[dkirkby@jackdaw gpuowl-static]$ ./gpuowlb
./gpuowlb: /lib64/libquadmath.so.0: version `QUADMATH_1.2' not found (required by ./gpuowlb)
./gpuowlb: /lib64/libm.so.6: version `GLIBC_2.29' not found (required by ./gpuowlb)
./gpuowlb: /lib64/libstdc++.so.6: version `GLIBCXX_3.4.20' not found (required by ./gpuowlb)
./gpuowlb: /lib64/libstdc++.so.6: version `CXXABI_1.3.9' not found (required by ./gpuowlb)
./gpuowlb: /lib64/libstdc++.so.6: version `CXXABI_1.3.11' not found (required by ./gpuowlb)
./gpuowlb: /lib64/libstdc++.so.6: version `GLIBCXX_3.4.26' not found (required by ./gpuowlb)
./gpuowlb: /lib64/libstdc++.so.6: version `GLIBCXX_3.4.22' not found (required by ./gpuowlb)
./gpuowlb: /lib64/libstdc++.so.6: version `GLIBCXX_3.4.21' not found (required by ./gpuowlb)
[dkirkby@jackdaw gpuowl-static]$ ./gpuowlc
./gpuowlc: /lib64/libquadmath.so.0: version `QUADMATH_1.2' not found (required by ./gpuowlc)
./gpuowlc: /lib64/libm.so.6: version `GLIBC_2.29' not found (required by ./gpuowlc)
[dkirkby@jackdaw gpuowl-static]$ [dkirkby@jackdaw gpuowl-static]$ ./gpuowla
bash: [dkirkby@jackdaw: command not found...
[dkirkby@jackdaw gpuowl-static]$ ./gpuowla: /lib64/libquadmath.so.0: version `QUADMATH_1.2' not found (required by ./gpuowla)
> ./gpuowla: /lib64/libm.so.6: version `GLIBC_2.29' not found (required by ./gpuowla)
> [dkirkby@jackdaw gpuowl-static]$ ./gpuowlb
> ./gpuowlb: /lib64/libquadmath.so.0: version `QUADMATH_1.2' not found (required by ./gpuowlb)
-bash: syntax error near unexpected token `('
[dkirkby@jackdaw gpuowl-static]$ ./gpuowlb: /lib64/libm.so.6: version `GLIBC_2.29' not found (required by ./gpuowlb)
> ./gpuowlb: /lib64/libstdc++.so.6: version `GLIBCXX_3.4.20' not found (required by ./gpuowlb)
> ./gpuowlb: /lib64/libstdc++.so.6: version `CXXABI_1.3.9' not found (required by ./gpuowlb)
-bash: syntax error near unexpected token `('
[dkirkby@jackdaw gpuowl-static]$ ./gpuowlb: /lib64/libstdc++.so.6: version `CXXABI_1.3.11' not found (required by ./gpuowlb)
> ./gpuowlb: /lib64/libstdc++.so.6: version `GLIBCXX_3.4.26' not found (required by ./gpuowlb)
> ./gpuowlb: /lib64/libstdc++.so.6: version `GLIBCXX_3.4.22' not found (required by ./gpuowlb)
-bash: syntax error near unexpected token `('
[dkirkby@jackdaw gpuowl-static]$ ./gpuowlb: /lib64/libstdc++.so.6: version `GLIBCXX_3.4.21' not found (required by ./gpuowlb)
> [dkirkby@jackdaw gpuowl-static]$ ./gpuowlc
> ./gpuowlc: /lib64/libquadmath.so.0: version `QUADMATH_1.2' not found (required by ./gpuowlc)
> ./gpuowlc: /lib64/libm.so.6: version `GLIBC_2.29' not found (required by ./gpuowlc)
-bash: syntax error near unexpected token `('
[dkirkby@jackdaw gpuowl-static]$