gpuowl: runtime error

[Xyzzy]

4096MB seems to be fine. You don't get that much faster with more memory. Especially if you stall the program!

[petrw1]

Its running and it's a lot faster than a CPU.

A 6.7566 GhzDay P-1 in under 30 minutes with a P100 GPU

This would take 1 core of my 7820X 18 hours.
All 8 cores of my CPU could do just over 10 per day; this GPU could do almost 50 per day.

Observation: I specified B1=1,500,000 B2=30,000,000
I got Stage 1=2,164,271; Stage 2=1,460,867 primes and 2,880 (classes???)

PS I added the commas for readability

Observation#2: If I leave off the ,2 at the end of this worktodo.txt line
B1=1500000,B2=30000000;PFactor=0,1,2,40371047,-1,74,2
I get the error

2020-09-19 04:21:44 colab 1 FFT: 128K 256:1:256 (0.00 bpw)
2020-09-19 04:21:44 colab FFT size too large for exponent (0.00 bits/word).
2020-09-19 04:21:44 colab Exiting because "FFT size too large"

[petrw1]

2020-10-01 23:37:32 colab worktodo.txt line ignored: "B1=1500000,B2=30000000;PFactor=0,1,2,40382929,-1,74,2"


Thanks

Never mind, I got the new build which I understand no longer does P1 alone.

[moebius]

Originally Posted by petrw1

Never mind, I got the new build which I understand no longer does P1 alone.

https://mersenneforum.org/showpost.p...28&postcount=1

[kriesel]

Originally Posted by petrw1

2020-10-01 23:37:32 colab worktodo.txt line ignored: "B1=1500000,B2=30000000;PFactor=0,1,2,40382929,-1,74,2"

Thanks

Never mind, I got the new build which I understand no longer does P1 alone.

Gpuowl 7.0 is not ready yet, don't use it, per Mihai https://www.mersenneforum.org/showpo...0&postcount=44

[kriesel]

Originally Posted by moebius

Here is the executable for colab with the libstdc files

Tried that on a new colab account, got the following as entire output of the run attempt:
./gpuowl.exe: /usr/lib/x86_64-linux-gnu/libstdc++.so.6: version `GLIBCXX_3.4.26' not found (required by ./gpuowl.exe)


(edit: and I see others have hit that error too. Will try some of the posted resolution approaches at the next opportunities. If moebius or whoever were to post an updated archive file with the needed file included, that might help.)

[chalsall]

Originally Posted by kriesel

Tried that on a new colab account, got the following as entire output of the run attempt:
./gpuowl.exe: /usr/lib/x86_64-linux-gnu/libstdc++.so.6: version `GLIBCXX_3.4.26' not found (required by ./gpuowl.exe)

Quoting a thread almost exactly a year ago. Why?

I was drilling down to ensure I hadn't said anything stupid in my recent posts. I came across this example of someone well trained being able to reproduce excellence (once they understood the theory of operation).

We're all actually on the same side.

Correct?

[Dr Sardonicus]

Originally Posted by chalsall

Quoting a thread almost exactly a year ago. Why?

From my reading, I would guess it's because (my emphasis)

Originally Posted by kriesel

Tried that [the executable for colab from the quoted post] on a new colab account, got the following as entire output of the run attempt:
./gpuowl.exe: /usr/lib/x86_64-linux-gnu/libstdc++.so.6: version `GLIBCXX_3.4.26' not found (required by ./gpuowl.exe)

[moebius]

Originally Posted by kriesel

If moebius or whoever were to post an updated archive file with the needed file included, that might help.)

read this post, did You copy the libstdc files to your working directory? I can make a new binary, if you tell me the path to the old repository
https://mersenneforum.org/showpost.p...7&postcount=33

[kriesel]

Originally Posted by moebius

did You copy the libstdc files to your working directory?

yes. libstd... missing is not the issue, libcxx version is in that case.

I tried 3 different Linux for Colab versions and hit different issues in each, finally got Fan Ming's going, although it blocks mprime on the CPU due to excessive CPU usage in gpuowl. I think that is ~V6.11-329 based on dates. It does not give a version when it runs. Purpose of the exercise was to not only get that account going, but document what's needed to clone to other accounts' Google drives without building anew on each, or at every new launch that gets a different Colab VM.

From my "building with msys2" notes, to get an old gpuowl commit, find the hash corresponding to the version desired, by matching the leading digits of the hash to the trailing digits of the gpuowl version.
For V6.11-380-g79ea0cc that's 79ea0cc29184237b24018e9396df271ec2754e97. https://github.com/preda/gpuowl/comm...df271ec2754e97

Then do similar to

git clone --branch v6 https://github.com/preda/gpuowl
cd gpuowl  
git checkout 79ea0cc29184237b24018e9396df271ec2754e97
or perhaps git reset --hard 79ea0cc29184237b24018e9396df271ec2754e97

There's more on my 3 attempts with downloaded builds in the second half of https://www.mersenneforum.org/showpo...0&postcount=7; use browser search for "build anew" to find the start of the relevant section.

[moebius]

GLIBCXX_3.4.26 comes with GCC 9.1.0. The gpuOwl 6.11.380 binary at https://mersenneforum.org/showpost.p...1&postcount=40 does not work for you, because the gcc version is not the same. I compiled it with gcc 9.1.0 https://gcc.gnu.org/onlinedocs/libst...anual/abi.html
It should work if you install gcc 9.1.0 and then copy the needed files with a notebook https://mersenneforum.org/showpost.p...7&postcount=33
Sure it isn't the same issue like in this thread? https://mersenneforum.org/showthread...511#post555511