mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > Hardware > GPU Computing > GpuOwl

Reply
 
Thread Tools
Old 2018-03-01, 19:00   #1
SELROC
 

22·1,259 Posts
Default gpuowl: runtime error

This on debian buster,
Here's an extract of the program output:


Note: using short, fused carry and fused tail kernels
OpenCL compilation in 616 ms, with " -DEXP=84674341u -I. -cl-fast-relaxed-math -cl-kernel-arg-info "
PRP-3: FFT 5000K (625 * 4096 * 2) of 84674341 (16.54 bits/word) [2018-03-01 19:57:46 CET]
Starting at iteration 0
error -55 (fft4K)
gpuowl: clwrap.h:267: void run(cl_queue, cl_kernel, size_t, size_t, const string&): Assertion `check(clEnqueueNDRangeKernel(queue, kernel, 1, __null, &workSize, &groupSize, 0, __null, __null), name.c_str())' failed.
Aborted
  Reply With Quote
Old 2018-03-01, 20:17   #2
preda
 
preda's Avatar
 
"Mihai Preda"
Apr 2015

1,297 Posts
Default

-55 is "invalid work item size". Apparently the OpenCL system you're using does not support a 512 workgroup size.

Quote:
Originally Posted by SELROC View Post
This on debian buster,
Here's an extract of the program output:


Note: using short, fused carry and fused tail kernels
OpenCL compilation in 616 ms, with " -DEXP=84674341u -I. -cl-fast-relaxed-math -cl-kernel-arg-info "
PRP-3: FFT 5000K (625 * 4096 * 2) of 84674341 (16.54 bits/word) [2018-03-01 19:57:46 CET]
Starting at iteration 0
error -55 (fft4K)
gpuowl: clwrap.h:267: void run(cl_queue, cl_kernel, size_t, size_t, const string&): Assertion `check(clEnqueueNDRangeKernel(queue, kernel, 1, __null, &workSize, &groupSize, 0, __null, __null), name.c_str())' failed.
Aborted
preda is online now   Reply With Quote
Old 2018-03-01, 20:20   #3
SELROC
 

3×31×79 Posts
Default

Quote:
Originally Posted by preda View Post
-55 is "invalid work item size". Apparently the OpenCL system you're using does not support a 512 workgroup size.
Thanks, what should I do to make the program work ?

perhaps install a different opencl package ?
  Reply With Quote
Old 2018-03-01, 20:47   #4
SELROC
 

10111100002 Posts
Default

Quote:
Originally Posted by SELROC View Post
Thanks, what should I do to make the program work ?

perhaps install a different opencl package ?
I can modify the program if this is necessary, but I would need guidance from the author
  Reply With Quote
Old 2018-03-02, 08:25   #5
SELROC
 

1010100111012 Posts
Default

Quote:
Originally Posted by preda View Post
-55 is "invalid work item size". Apparently the OpenCL system you're using does not support a 512 workgroup size.
Effectively the max work group size is 256.

How should I modify the program to make it work with this hardware ?

Thank you
  Reply With Quote
Old 2018-03-02, 08:32   #6
preda
 
preda's Avatar
 
"Mihai Preda"
Apr 2015

129710 Posts
Default

Quote:
Originally Posted by SELROC View Post
Effectively the max work group size is 256.

How should I modify the program to make it work with this hardware ?

Thank you
There's no easy way (to use workgroup 256 in this situation) I can think of.

If you install amdgpu-pro or ROCm, you should be able to use WG up to 1024.

In the next update I'll try to move back to 256.
preda is online now   Reply With Quote
Old 2018-03-02, 08:50   #7
SELROC
 

2×5×677 Posts
Default

Quote:
Originally Posted by preda View Post
There's no easy way (to use workgroup 256 in this situation) I can think of.

If you install amdgpu-pro or ROCm, you should be able to use WG up to 1024.

In the next update I'll try to move back to 256.

Thanks very much, I look forward for the mods.

selroc
  Reply With Quote
Old 2018-03-02, 17:04   #8
SELROC
 

22·3·193 Posts
Default

Quote:
Originally Posted by preda View Post
There's no easy way (to use workgroup 256 in this situation) I can think of.

If you install amdgpu-pro or ROCm, you should be able to use WG up to 1024.

In the next update I'll try to move back to 256.
Somehow I got the program to work by reinstalling a fresh debian testing and amdgpu-pro

It is running right now
  Reply With Quote
Old 2018-03-02, 17:22   #9
SELROC
 

231218 Posts
Default

Quote:
Originally Posted by SELROC View Post
Somehow I got the program to work by reinstalling a fresh debian testing and amdgpu-pro

It is running right now


gpuOwL v2.0--mod GPU Mersenne primality checker
Ellesmere-36x1360-@4:0.0 Radeon RX 580 Series
Note: using short, fused carry and fused tail kernels
OpenCL compilation in 628 ms, with " -DEXP=84701459u -I. -cl-fast-relaxed-math -cl-kernel-arg-info "
PRP-3: FFT 5000K (625 * 4096 * 2) of 84701459 (16.54 bits/word) [2018-03-02 18:05:01 CET]
Starting at iteration 142500
OK 142500 / 84701459 [ 0.17%], 0.00 ms/it [0.00, 0.00], check 3.73s; ETA 0d 00:00; 821d3202550d3c23 [18:05:06]
OK 143000 / 84701459 [ 0.17%], 5.29 ms/it [5.29, 5.29], check 3.33s; ETA 5d 04:21; 71d0fd7863001c6d [18:05:11]
OK 144000 / 84701459 [ 0.17%], 5.00 ms/it [4.70, 5.29], check 3.32s; ETA 4d 21:24; 5f059bf89d226260 [18:05:20]
OK 145000 / 84701459 [ 0.17%], 5.01 ms/it [4.71, 5.30], check 3.32s; ETA 4d 21:36; 5055eab68c2355f2 [18:05:28]
OK 150000 / 84701459 [ 0.18%], 4.77 ms/it [4.71, 5.31], check 3.33s; ETA 4d 16:01; 6dc264064ce6830b [18:05:55]
OK 160000 / 84701459 [ 0.19%], 4.84 ms/it [4.72, 6.13], check 3.37s; ETA 4d 17:45; 7bd9b48b95f55663 [18:06:47]
OK 170000 / 84701459 [ 0.20%], 4.79 ms/it [4.72, 5.47], check 3.37s; ETA 4d 16:34; 4504231bf5b0b0af [18:07:38]
OK 180000 / 84701459 [ 0.21%], 4.80 ms/it [4.73, 5.49], check 3.37s; ETA 4d 16:46; 9500155d9891ec05 [18:08:30]
OK 200000 / 84701459 [ 0.24%], 4.79 ms/it [4.73, 5.38], check 3.40s; ETA 4d 16:20; 1cfb4f382ad7729b [18:10:09]
OK 220000 / 84701459 [ 0.26%], 4.79 ms/it [4.73, 5.40], check 3.38s; ETA 4d 16:21; 0d2bb5f1d3f7f8f0 [18:11:48]
OK 240000 / 84701459 [ 0.28%], 4.79 ms/it [4.73, 5.40], check 3.37s; ETA 4d 16:20; 009378cb131480c5 [18:13:27]
OK 260000 / 84701459 [ 0.31%], 4.82 ms/it [4.74, 5.41], check 3.38s; ETA 4d 17:04; 3c8ca29ba50ac4f6 [18:15:07]
OK 300000 / 84701459 [ 0.35%], 4.79 ms/it [4.73, 5.52], check 3.38s; ETA 4d 16:23; 2cef54e6f19b57f3 [18:18:22]
  Reply With Quote
Old 2018-03-03, 09:22   #10
SELROC
 

8410 Posts
Default

One thing I notice with two instances of gpuowl running: one instance gets stuck and the only way to stop it is reboot
  Reply With Quote
Old 2018-03-10, 15:02   #11
SELROC
 

6,961 Posts
Default

Quote:
Originally Posted by SELROC View Post
One thing I notice with two instances of gpuowl running: one instance gets stuck and the only way to stop it is reboot
Hello Mihai, have you attempted yet to reproduce the error ?

I have reinstalled debian-testing with amdgpu-pro and still getting the same error: if two instances of gpuowl are launched, the first remains in a blocked state and we can only reboot to stop it.

However, the normal reboot will not work (with a message: "watchdog did not stop") and we can only switch off the power to reboot.

My GPU hardware is Radeon RX 580
  Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
gpuowl tuning M344587487 GpuOwl 14 2018-12-29 08:11
How to interface gpuOwl with PrimeNet preda PrimeNet 2 2017-10-07 21:32
runtime question yoyo YAFU 1 2015-01-08 07:07
runtime error when using redc ltd GMP-ECM 5 2009-10-30 13:09
ECM Runtime and F20 D. B. Staple Factoring 11 2007-12-12 16:52

All times are UTC. The time now is 20:37.

Wed Oct 28 20:37:40 UTC 2020 up 48 days, 17:48, 1 user, load averages: 1.92, 1.91, 1.91

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2020, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.