mersenneforum.org  

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

Reply
 
Thread Tools
Old 2020-10-24, 11:24   #133
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

111368 Posts
Default

Quote:
Originally Posted by preda View Post
Were you using -log 100000 by any chance in your config?
Yes I think so. config.txt last modifed 18 Oct 2020 was
Code:
-user kriesel -cpu asr2/radeonvii1 -d 1 -maxAlloc 14G -proof 8 -log 100000 -use NO_ASM
kriesel is online now   Reply With Quote
Old 2020-10-24, 11:29   #134
preda
 
preda's Avatar
 
"Mihai Preda"
Apr 2015

24×83 Posts
Default

Quote:
Originally Posted by kriesel View Post
Yes I think so. config.txt last modifed 18 Oct 2020 was
Code:
-user kriesel -cpu asr2/radeonvii1 -d 1 -maxAlloc 14G -proof 8 -log 100000 -use NO_ASM
NO_ASM on R7 -- some like it slow?
preda is offline   Reply With Quote
Old 2020-10-24, 11:32   #135
preda
 
preda's Avatar
 
"Mihai Preda"
Apr 2015

24·83 Posts
Default

Quote:
Originally Posted by kriesel View Post
I see by scanning the commit, you've changed it from loop on repeated error, to stop the program on repeated error. Either of those results in up to a lost gpu day if checked daily. (Saves a little electricity, loses a lot of time.)
OK I understand, I'll consider implementing this. I still consider "stop the GPU" a safe bail-out, while "keep running 100% doing nothing" a waste.
preda is offline   Reply With Quote
Old 2020-10-24, 11:35   #136
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

2·2,351 Posts
Default

Quote:
Originally Posted by preda View Post
OK I understand, I'll consider implementing this. I still consider "stop the GPU" a safe bail-out, while "keep running 100% doing nothing" a waste.
It's not keep doing nothing, it's try something else (next assignment) and see if that works.
kriesel is online now   Reply With Quote
Old 2020-10-24, 11:37   #137
preda
 
preda's Avatar
 
"Mihai Preda"
Apr 2015

24×83 Posts
Default

Quote:
Originally Posted by kriesel View Post
It's not keep doing nothing, it's try something else (next assignment) and see if that works.
Yes I agree. I wanted to say that the fix to the loop you reported was high-priority because otherwise it represented such a waste.
preda is offline   Reply With Quote
Old 2020-10-24, 11:46   #138
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

2·2,351 Posts
Default

Quote:
Originally Posted by preda View Post
NO_ASM on R7 -- some like it slow?
No, avoiding the delay of a compile that fails is why I skip the ASM attempt that generates an error and NO_ASM retry. Same worktodo file, same gpuowl version, same Win 10 system:

Code:
2020-10-24 06:38:22 gpuowl v7.0-35-gf06bc5b
2020-10-24 06:38:22 config: -user kriesel -cpu asr2/radeonvii1 -d 1 -maxAlloc 14G -proof 8 -log 100000
2020-10-24 06:38:22 device 1, unique id ''
2020-10-24 06:38:22 asr2/radeonvii1 worktodo.txt line ignored: "# B1=8310000,B2=249300000;PRP=0,1,2,957156667,-1,86,2 got stuck in EE loop in P1"
2020-10-24 06:38:22 asr2/radeonvii1 843112609 FFT: 48M 4K:12:512 (16.75 bpw)
2020-10-24 06:38:30 asr2/radeonvii1 843112609 OpenCL args "-DEXP=843112609u -DWIDTH=4096u -DSMALL_HEIGHT=512u -DMIDDLE=12u -DAMDGPU=1 -DCARRY64=1 -DCARRYM64=1 -DWEIGHT_STEP_MINUS_1=0xc.0c8ba145d86d8p-6 -DIWEIGHT_STEP_MINUS_1=-0xa.23d68ede2efd8p-6  -cl-unsafe-math-optimizations -cl-std=CL2.0 -cl-finite-math-only "
2020-10-24 06:38:30 asr2/radeonvii1 843112609 ASM compilation failed, retrying compilation using NO_ASM
2020-10-24 06:38:36 asr2/radeonvii1 843112609 OpenCL compilation in 6.46 s
2020-10-24 06:38:39 asr2/radeonvii1 843112609 maxAlloc: 14.0 GB
2020-10-24 06:38:40 asr2/radeonvii1 843112609 P1(7520000) 10848976 bits
2020-10-24 06:38:49 asr2/radeonvii1 843112609 OK  10894000 loaded: blockSize 500, 6cd3c74e38c2ef7f
2020-10-24 06:38:49 asr2/radeonvii1 843112609 validating proof residues for power 8
2020-10-24 06:38:51 asr2/radeonvii1 843112609 Proof using power 8
2020-10-24 06:39:11 asr2/radeonvii1 843112609 OK  10895000   1.29% 66174a251f389b1c 16720 us/it; ETA 161d 01:14
2020-10-24 06:39:14 asr2/radeonvii1 843112609 P2(7520000,225.6M) Warning: not enough memory for efficient P2. Increase -maxAlloc if possible
2020-10-24 06:39:18 asr2/radeonvii1 843112609 P2(7520000,225.6M) B1=7520000, B2=225600000, D=210: 10940144 primes in [23433795, 225600165], selected 9651616 (88.2%) (1288528 doubles + 8363088 singles)
2020-10-24 06:39:18 asr2/radeonvii1 843112609 P2(7520000,225.6M) D=210; from B2=23433795 : 962697 blocks starting at 111590
2020-10-24 06:39:18 asr2/radeonvii1 843112609 P2(7520000,225.6M) Acquired memory lock 'memlock-1'
2020-10-24 06:39:18 asr2/radeonvii1 843112609 P2(7520000,225.6M) Allocated 24 buffers
2020-10-24 06:39:19 asr2/radeonvii1 843112609 P2(7520000,225.6M) Starting P1 GCD
2020-10-24 06:39:23 asr2/radeonvii1 843112609 P2(7520000,225.6M) Setup 24 P2 buffers in 4984.0 ms
2020-10-24 06:39:23 asr2/radeonvii1 843112609 P2(7520000,225.6M)  23434005 (  0%); 11 muls, 14134 us/mul

Last fiddled with by kriesel on 2020-10-24 at 11:52
kriesel is online now   Reply With Quote
Old 2020-10-24, 11:54   #139
preda
 
preda's Avatar
 
"Mihai Preda"
Apr 2015

53016 Posts
Default

Quote:
Originally Posted by kriesel View Post
No, avoiding the delay of a compile that fails is why I skip the ASM attempt that generates an error and NO_ASM retry. Same worktodo file, same gpuowl version, same Win 10 system:
Yep. I forgot that the driver on Windows does not support ASM.
preda is offline   Reply With Quote
Old 2020-10-24, 11:57   #140
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

2×2,351 Posts
Default

Quote:
Originally Posted by preda View Post
Yep. I forgot that the driver on Windows does not support ASM.
How much difference do you see in iteration timings between ASM and NO_ASM on linux on Radeon VII? Only in ROCm?
kriesel is online now   Reply With Quote
Old 2020-10-24, 11:58   #141
Viliam Furik
 
Jul 2018
Martin, Slovakia

3768 Posts
Default

Quote:
Originally Posted by kriesel View Post
No, that's why I skip the ASM attempt that generates an error and NO_ASM retry.
Here I was, thinking that this is normal...

I can confirm the ASM error. I have Radeon VII on Windows 10, too.

-----
EDIT:
It seems like you have discussed the problem while I was writing my post.

Last fiddled with by Viliam Furik on 2020-10-24 at 12:00
Viliam Furik is offline   Reply With Quote
Old 2020-10-24, 17:59   #142
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

125E16 Posts
Default

Quote:
Originally Posted by Viliam Furik View Post
I can confirm the ASM error. I have Radeon VII on Windows 10, too.

-----
EDIT:
It seems like you have discussed the problem while I was writing my post.
Race conditions are not just for computers, or speed sports. 'salright.

Last fiddled with by kriesel on 2020-10-24 at 17:59
kriesel is online now   Reply With Quote
Old 2020-10-24, 18:05   #143
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

470210 Posts
Default -use STATS fail

It's repeatable in V7.0-35, on both error looping and normally running worktodo lines.
Seen on a system with 16GB ram, Celeron G1840 cpu, Win 10 Pro x64, Radeon VII, gpuowl-win v7.0-35-gf06bc5b.
Code:
2020-10-24 10:31:33 gpuowl v7.0-35-gf06bc5b
2020-10-24 10:31:33 config: -user kriesel -cpu asr2/radeonvii1 -d 1 -maxAlloc 14G -proof 8 -log 100000 -use STATS,NO_ASM
2020-10-24 10:31:33 device 1, unique id ''
2020-10-24 10:31:33 asr2/radeonvii1 worktodo.txt line ignored: "# B1=8310000,B2=249300000;PRP=0,1,2,957156667,-1,86,2 got stuck in EE loop in P1"
2020-10-24 10:31:33 asr2/radeonvii1 957156667 FFT: 52M 4K:13:512 (17.55 bpw)
2020-10-24 10:31:42 asr2/radeonvii1 957156667 OpenCL args "-DEXP=957156667u -DWIDTH=4096u -DSMALL_HEIGHT=512u -DMIDDLE=13u -DAMDGPU=1 -DCARRY64=1 -DCARRYM64=1 -DMM_CHAIN=2u -DMM2_CHAIN=2u -DMAX_ACCURACY=1 -DWEIGHT_STEP_MINUS_1=0xb.9677ce2235ecp-5 -DIWEIGHT_STEP_MINUS_1=-0x8.81d5d7e6b906p-5 -DNO_ASM=1 -DSTATS=1  -cl-unsafe-math-optimizations -cl-std=CL2.0 -cl-finite-math-only "
2020-10-24 10:31:49 asr2/radeonvii1 957156667 OpenCL compilation in 6.90 s
2020-10-24 10:31:52 asr2/radeonvii1 957156667 maxAlloc: 14.0 GB
2020-10-24 10:31:52 asr2/radeonvii1 957156667 P1(8310000) 11988434 bits
2020-10-24 10:31:55 asr2/radeonvii1 957156667 Acquired memory lock 'memlock-1'
2020-10-24 10:31:55 asr2/radeonvii1 957156667 P1(8310000) using 49 buffers
2020-10-24 10:31:59 asr2/radeonvii1 957156667 P1(8310000) releasing 49 buffers
2020-10-24 10:32:00 asr2/radeonvii1 957156667 Released memory lock 'memlock-1'
2020-10-24 10:32:00 asr2/radeonvii1 Exception gpu_error: OUT_OF_RESOURCES carryFused at clwrap.cpp:325 run
2020-10-24 10:32:00 asr2/radeonvii1 Bye

2020-10-24 10:32:34 gpuowl v7.0-35-gf06bc5b
2020-10-24 10:32:34 config: -user kriesel -cpu asr2/radeonvii1 -d 1 -maxAlloc 15G -proof 8 -log 100000 -use STATS,NO_ASM
2020-10-24 10:32:34 device 1, unique id ''
2020-10-24 10:32:34 asr2/radeonvii1 worktodo.txt line ignored: "# B1=8310000,B2=249300000;PRP=0,1,2,957156667,-1,86,2 got stuck in EE loop in P1"
2020-10-24 10:32:34 asr2/radeonvii1 957156667 FFT: 52M 4K:13:512 (17.55 bpw)
2020-10-24 10:32:42 asr2/radeonvii1 957156667 OpenCL args "-DEXP=957156667u -DWIDTH=4096u -DSMALL_HEIGHT=512u -DMIDDLE=13u -DAMDGPU=1 -DCARRY64=1 -DCARRYM64=1 -DMM_CHAIN=2u -DMM2_CHAIN=2u -DMAX_ACCURACY=1 -DWEIGHT_STEP_MINUS_1=0xb.9677ce2235ecp-5 -DIWEIGHT_STEP_MINUS_1=-0x8.81d5d7e6b906p-5 -DNO_ASM=1 -DSTATS=1  -cl-unsafe-math-optimizations -cl-std=CL2.0 -cl-finite-math-only "
2020-10-24 10:32:49 asr2/radeonvii1 957156667 OpenCL compilation in 6.98 s
2020-10-24 10:32:53 asr2/radeonvii1 957156667 maxAlloc: 15.0 GB
2020-10-24 10:32:53 asr2/radeonvii1 957156667 P1(8310000) 11988434 bits
2020-10-24 10:32:55 asr2/radeonvii1 957156667 Acquired memory lock 'memlock-1'
2020-10-24 10:32:55 asr2/radeonvii1 957156667 P1(8310000) using 54 buffers
2020-10-24 10:33:00 asr2/radeonvii1 957156667 P1(8310000) releasing 54 buffers
2020-10-24 10:33:01 asr2/radeonvii1 957156667 Released memory lock 'memlock-1'
2020-10-24 10:33:01 asr2/radeonvii1 Exception gpu_error: OUT_OF_RESOURCES carryFused at clwrap.cpp:325 run
2020-10-24 10:33:01 asr2/radeonvii1 Bye

2020-10-24 10:34:13 gpuowl v7.0-35-gf06bc5b
2020-10-24 10:34:13 config: -user kriesel -cpu asr2/radeonvii1 -d 1 -maxAlloc 15G -proof 8 -log 100000 -use STATS,NO_ASM
2020-10-24 10:34:13 device 1, unique id ''
2020-10-24 10:34:13 asr2/radeonvii1 worktodo.txt line ignored: "# B1=8310000,B2=249300000;PRP=0,1,2,957156667,-1,86,2 got stuck in EE loop in P1"
2020-10-24 10:34:13 asr2/radeonvii1 843112609 FFT: 48M 4K:12:512 (16.75 bpw)
2020-10-24 10:34:21 asr2/radeonvii1 843112609 OpenCL args "-DEXP=843112609u -DWIDTH=4096u -DSMALL_HEIGHT=512u -DMIDDLE=12u -DAMDGPU=1 -DCARRY64=1 -DCARRYM64=1 -DWEIGHT_STEP_MINUS_1=0xc.0c8ba145d86d8p-6 -DIWEIGHT_STEP_MINUS_1=-0xa.23d68ede2efd8p-6 -DNO_ASM=1 -DSTATS=1  -cl-unsafe-math-optimizations -cl-std=CL2.0 -cl-finite-math-only "
2020-10-24 10:34:28 asr2/radeonvii1 843112609 OpenCL compilation in 6.67 s
2020-10-24 10:34:31 asr2/radeonvii1 843112609 maxAlloc: 15.0 GB
2020-10-24 10:34:31 asr2/radeonvii1 843112609 P1(7520000) 10848976 bits
2020-10-24 10:34:34 asr2/radeonvii1 Exception gpu_error: OUT_OF_RESOURCES carryFused at clwrap.cpp:325 run
2020-10-24 10:34:34 asr2/radeonvii1 Bye
kriesel is online now   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
GpuOwl PRP-Proof changes preda GpuOwl 20 2020-10-17 06:51
gpuowl: runtime error SELROC GpuOwl 59 2020-10-02 03:56
gpuOWL for Wagstaff GP2 GpuOwl 22 2020-06-13 16:57
gpuowl tuning M344587487 GpuOwl 14 2018-12-29 08:11
How to interface gpuOwl with PrimeNet preda PrimeNet 2 2017-10-07 21:32

All times are UTC. The time now is 19:22.

Wed Nov 25 19:22:05 UTC 2020 up 76 days, 16:33, 3 users, load averages: 1.58, 1.60, 1.52

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.