Google Diet Colab Notebook

[Fan Ming]

Complied new version of gpuowl (commit E5A8F2C:https://github.com/preda/gpuowl/comm...c5dcd953ae33cf) on Google colab.
Not tested, but should work.

[moebius]

Originally Posted by Fan Ming

Complied new version of gpuowl.Not tested, but should work.

Unfortunately, it still doesn't work for me.

2020-06-29 14:12:09 gpuowl
2020-06-29 14:12:09 Note: not found 'config.txt'
2020-06-29 14:12:09 config: -prp 333999549
2020-06-29 14:12:09 device 0, unique id ''
2020-06-29 14:12:09 Exception gpu_error: clGetPlatformIDs(16, platforms, (unsigned *) &nPlatforms) at clwrap.cpp:71 getDeviceIDs
2020-06-29 14:12:09 Bye

[kriesel]

Originally Posted by moebius

Unfortunately, it still doesn't work for me.

2020-06-29 14:12:09 gpuowl
2020-06-29 14:12:09 Note: not found 'config.txt'
2020-06-29 14:12:09 config: -prp 333999549
2020-06-29 14:12:09 device 0, unique id ''
2020-06-29 14:12:09 Exception gpu_error: clGetPlatformIDs(16, platforms, (unsigned *) &nPlatforms) at clwrap.cpp:71 getDeviceIDs
2020-06-29 14:12:09 Bye

Consider using worktodo.txt for what changes (work assignments), and config.txt for what doesn't normally change. Create a config.txt that tells gpuowl what device to use, and retry.

[Fan Ming]

Originally Posted by moebius

Unfortunately, it still doesn't work for me.

2020-06-29 14:12:09 gpuowl
2020-06-29 14:12:09 Note: not found 'config.txt'
2020-06-29 14:12:09 config: -prp 333999549
2020-06-29 14:12:09 device 0, unique id ''
2020-06-29 14:12:09 Exception gpu_error: clGetPlatformIDs(16, platforms, (unsigned *) &nPlatforms) at clwrap.cpp:71 getDeviceIDs
2020-06-29 14:12:09 Bye

Is this a GPU session(Runtime type->Accelerator should be GPU)? It seems that no GPU was detected.

[moebius]

Originally Posted by Fan Ming

Is this a GPU session(Runtime type->Accelerator should be GPU)? It seems that no GPU was detected.

No, but I've changed this. Thanks for the support.

Unfortunately, the Tesla K80 does not seem to run flawlessly.


2020-06-29 17:02:24 gpuowl
2020-06-29 17:02:24 Note: not found 'config.txt'
2020-06-29 17:02:24 config: -prp 333999549
2020-06-29 17:02:24 device 0, unique id ''
2020-06-29 17:02:24 Tesla K80-0 333999549 FFT: 18M 1K:9:1K (17.70 bpw)
2020-06-29 17:02:24 Tesla K80-0 Expected maximum carry32: 6E890000
2020-06-29 17:02:26 Tesla K80-0 OpenCL args "-DEXP=333999549u -DWIDTH=1024u -DSMALL_HEIGHT=1024u -DMIDDLE=9u -DPM1=0 -DMM2_CHAIN=1u -DMAX_ACCURACY=1 -DWEIGHT_STEP_MINUS_1=0x1.e0807b194ba13p-3 -DIWEIGHT_STEP_MINUS_1=-0x1.8530a90b718e5p-3 -cl-unsafe-math-optimizations -cl-std=CL2.0 -cl-finite-math-only "
2020-06-29 17:02:29 Tesla K80-0

2020-06-29 17:02:29 Tesla K80-0 OpenCL compilation in 2.77 s
2020-06-29 17:02:36 Tesla K80-0 333999549 OK 0 loaded: blockSize 400, 0000000000000003
2020-06-29 17:02:53 Tesla K80-0 333999549 EE 800 0.00%; 13847 us/it; ETA 53d 12:43; e5e2d3b8fb255c88 (check 5.99s)
2020-06-29 17:02:59 Tesla K80-0 333999549 OK 0 loaded: blockSize 400, 0000000000000003
2020-06-29 17:03:16 Tesla K80-0 333999549 EE 800 0.00%; 13862 us/it; ETA 53d 14:06; e5e2d3b8fb255c88 (check 6.00s) 1 errors
2020-06-29 17:03:22 Tesla K80-0 333999549 OK 0 loaded: blockSize 400, 0000000000000003
2020-06-29 17:03:39 Tesla K80-0 333999549 EE 800 0.00%; 13889 us/it; ETA 53d 16:36; e5e2d3b8fb255c88 (check 5.99s) 2 errors
2020-06-29 17:03:39 Tesla K80-0 3 sequential errors, will stop.
2020-06-29 17:03:39 Tesla K80-0 Exiting because "too many errors"
2020-06-29 17:03:39 Tesla K80-0 Bye

[Fan Ming]

Originally Posted by moebius

No, but I've changed this. Thanks for the support.

Unfortunately, the Tesla K80 does not seem to run flawlessly.

Try the previous version I provided last year. Maybe optimizations in newer gpuowl caused unstable run, or the compiled new binary isn't well itself.

[Fan Ming]

Originally Posted by moebius

No, but I've changed this. Thanks for the support.

Unfortunately, the Tesla K80 does not seem to run flawlessly.


....
2020-06-29 17:02:24 Tesla K80-0 333999549 FFT: 18M 1K:9:1K (17.70 bpw)
2020-06-29 17:02:24 Tesla K80-0 Expected maximum carry32: 6E890000
2020-06-29 17:02:26 Tesla K80-0 OpenCL args "-DEXP=333999549u -DWIDTH=1024u -DSMALL_HEIGHT=1024u -DMIDDLE=9u -DPM1=0 -DMM2_CHAIN=1u -DMAX_ACCURACY=1 -DWEIGHT_STEP_MINUS_1=0x1.e0807b194ba13p-3 -DIWEIGHT_STEP_MINUS_1=-0x1.8530a90b718e5p-3 -cl-unsafe-math-optimizations -cl-std=CL2.0 -cl-finite-math-only "
2020-06-29 17:02:29 Tesla K80-0

Or try to cancel these optimizations.

[chalsall]

Originally Posted by moebius

Unfortunately, the Tesla K80 does not seem to run flawlessly.

When you get a K80 (the slowest GPU provided by Colab) choose "Factory Reset", and then reattach. Within a few attempts, you should get a P4, a P100, or a coveted T4.

[moebius]

1) LL in the double check range seems to work with the new compiled version.

2020-06-29 17:58:56 gpuowl
2020-06-29 17:58:56 Note: not found 'config.txt'
2020-06-29 17:58:56 config: -ll 53537243
2020-06-29 17:58:56 device 0, unique id ''
2020-06-29 17:58:56 Tesla T4-0 53537243 FFT: 2.75M 256:11:512 (18.57 bpw)
2020-06-29 17:58:56 Tesla T4-0 Expected maximum carry32: 46D30000
2020-06-29 17:58:56 Tesla T4-0 OpenCL args "-DEXP=53537243u -DWIDTH=256u -DSMALL_HEIGHT=512u -DMIDDLE=11u -DPM1=0 -DMM_CHAIN=1u -DMM2_CHAIN=1u -DMAX_ACCURACY=1 -DWEIGHT_STEP_MINUS_1=0x1.6730c3da2ac15p-2 -DIWEIGHT_STEP_MINUS_1=-0x1.09ea3b9d0d90fp-2 -cl-unsafe-math-optimizations -cl-std=CL2.0 -cl-finite-math-only "
2020-06-29 17:58:56 Tesla T4-0

2020-06-29 17:58:56 Tesla T4-0 OpenCL compilation in 0.01 s
2020-06-29 17:58:56 Tesla T4-0 53537243 LL 119000 loaded: b7fe47edf1df9c7f
2020-06-29 18:02:54 Tesla T4-0 53537243 LL 200000 0.37%; 2937 us/it; ETA 1d 19:31; d1177beaa336d0f0
2020-06-29 18:07:47 Tesla T4-0 53537243 LL 300000 0.56%; 2932 us/it; ETA 1d 19:21; 3cf6b471d02c92b7


2) canceling -cl-unsafe-math-optimizations

Unfortunately I am not so familiar with gpuowl to be able to do that quickly.


3) with -prp 333999549 the same error occurred again with the Tesla T4


I will run double checks for now.

[kriesel]

Originally Posted by moebius

2) canceling -cl-unsafe-math-optimizations

-h is your friend. Help output includes the option

-safeMath : do not use -cl-unsafe-math-optimizations (OpenCL)

[moebius]

Originally Posted by kriesel

-h is your friend. Help output includes the option

-safeMath : do not use -cl-unsafe-math-optimizations (OpenCL)

Thanks I missed that, I will try that later. I let the DC go through before colab throws me out.


2020-06-29 19:35:51 Tesla T4-0 53537243 OK 2000000 (jacobi == -1)
2020-06-29 19:40:44 Tesla T4-0 53537243 LL 2200000 4.11%; 2934 us/it; ETA 1d 17:51; 0ff568d9b44b522e
2020-06-29 19:45:37 Tesla T4-0 53537243 LL 2300000 4.30%; 2933 us/it; ETA 1d 17:44; 251446abea375f31
2020-06-29 19:50:31 Tesla T4-0 53537243 LL 2400000 4.48%; 2933 us/it; ETA 1d 17:40; ccfb451c0f8d3213
2020-06-29 19:55:24 Tesla T4-0 53537243 LL 2500000 4.67%; 2932 us/it; ETA 1d 17:34; 29045b8768c16eff
2020-06-29 20:00:17 Tesla T4-0 53537243 LL 2600000 4.86%; 2937 us/it; ETA 1d 17:33; bf1b2da355335b21
2020-06-29 20:00:17 Tesla T4-0 53537243 OK 2500000 (jacobi == -1)
2020-06-29 20:05:12 Tesla T4-0 53537243 LL 2700000 5.04%; 2941 us/it; ETA 1d 17:32; 4131529510cd3e19
/