Trial Factoring
log in

Advanced search

Message boards : Number crunching : Trial Factoring

Previous · 1 . . . 4 · 5 · 6 · 7 · 8 · 9 · 10 . . . 23 · Next
Author Message
Sergey Kovalchuk
Send message
Joined: 27 Oct 15
Posts: 12
Credit: 5,358,644
RAC: 0
Message 5917 - Posted: 28 Mar 2020, 19:06:56 UTC

there is no <cmdline> either in the app_version section or in workunit
so that it can be added to the app_config file without any problems

<app_version>
<app_name>GPU72</app_name>
<plan_class>opencl_ati_101</plan_class>
<cmdline> -d 1 </cmdline>
</app_version>

that's just different tasks need different parameters, as an option to launch the second instance of BOINC for the second GPU with the option "-d 1" in this project

PS in this way I was able to run some tasks on Intel GPU although the server did not recognize this GPU

Profile rebirther
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 2 Jan 13
Posts: 7232
Credit: 42,729,227
RAC: 31
Message 5918 - Posted: 28 Mar 2020, 19:34:45 UTC - in response to Message 5917.
Last modified: 28 Mar 2020, 19:50:06 UTC

there is no either in the app_version section or in workunit
so that it can be added to the app_config file without any problems


GPU72
opencl_ati_101
-d 1


that's just different tasks need different parameters, as an option to launch the second instance of BOINC for the second GPU with the option "-d 1" in this project

PS in this way I was able to run some tasks on Intel GPU although the server did not recognize this GPU


Intel GPU would run but wasnt setup, too slow.
If the -d 1 option can run a second instance (dual card) its nearly the same as the app_config in FAQ (both cards running one WU). If somebody can test it and confirm. If one of both configs are working this would be great to track down some issues. I will correct some things in FAQ about the app_config.xml

Update:
FAQ app_config.xml section updated, pls test so we can track down things

bluestang
Send message
Joined: 6 Jun 19
Posts: 59
Credit: 1,268,237,570
RAC: 0
Message 5919 - Posted: 28 Mar 2020, 20:07:07 UTC - in response to Message 5918.
Last modified: 28 Mar 2020, 20:22:07 UTC

Wrong again. -d 1 option does nothing on my 2x 1660ti setup. Only 1 GPU is being utilized by the app. And it is not even the 1st GPU in my system being used...it is the 2nd GPU. Makes no sense at all.

EDIT: The -d command only tells the app what GPU to use. It does nothing for running 1 WU on 2 GPUs like is being said in previous posts in this thread.

KEP
Volunteer tester
Send message
Joined: 28 Nov 14
Posts: 88
Credit: 1,102,770
RAC: 0
Message 5920 - Posted: 28 Mar 2020, 20:28:11 UTC - in response to Message 5919.

Dear bluestang

I have to ask you to cool down a bit. What you are requesting is not what mfakt(o)(c) was designed for. Having a multiGPU system, doesn't automatically means that the app you use, is able to utilize both GPUs on 1 task. It may be possible in the future, also there may be a tweak that can be done, but as far as I know, mfakt(o)(c) is not multiGPU useable. So in stead of the harsh tone (as I at least read your statements), then please show some support and constructive way to getting this issue solved. This is also brand new territory for Reb, so do not expect that everything will be easy solvable or easily adapted to your needs.

Take care and please show some patience and think about how your posts can be read :)

Stay safe :)

Profile rebirther
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 2 Jan 13
Posts: 7232
Credit: 42,729,227
RAC: 31
Message 5921 - Posted: 28 Mar 2020, 20:35:20 UTC - in response to Message 5920.

KEP, every info is welcome, we need more users which running into these issues and want to help to test. Lets see how we can fix this, linux opencl has a 100% failure rate and mac is using the same config. If we can fix linux then mac too.

Dirk Broer
Send message
Joined: 2 Jan 15
Posts: 71
Credit: 129,980,152
RAC: 80,689
Message 5922 - Posted: 29 Mar 2020, 0:28:47 UTC - in response to Message 5921.
Last modified: 29 Mar 2020, 0:29:28 UTC

KEP, every info is welcome, we need more users which running into these issues and want to help to test. Lets see how we can fix this, linux opencl has a 100% failure rate and mac is using the same config. If we can fix linux then mac too.


Something tells me that you tried to port a Windows CUDA app with a bare minimum of re-compiling. Try to make the developer start from OpenCL on a Linux system (preferably with a Radeon card) for the next OpenCL version.

bluestang
Send message
Joined: 6 Jun 19
Posts: 59
Credit: 1,268,237,570
RAC: 0
Message 5923 - Posted: 29 Mar 2020, 1:03:52 UTC - in response to Message 5920.

Dear bluestang

I have to ask you to cool down a bit. What you are requesting is not what mfakt(o)(c) was designed for. Having a multiGPU system, doesn't automatically means that the app you use, is able to utilize both GPUs on 1 task. It may be possible in the future, also there may be a tweak that can be done, but as far as I know, mfakt(o)(c) is not multiGPU useable. So in stead of the harsh tone (as I at least read your statements), then please show some support and constructive way to getting this issue solved. This is also brand new territory for Reb, so do not expect that everything will be easy solvable or easily adapted to your needs.

Take care and please show some patience and think about how your posts can be read :)

Stay safe :)


I am cool. You're reading it wrong. I'm trying to help get this figured out. It was said that the app will use both GPUs to run 1 WU and it wasn't and I'm trying to help figure out why. It's nothing against Rebirther, so don't make it so.

Either way, I figured out how to run on both cards by using my own method. After I run out of Seti WUs I'll be back to kick some tail :)

mmonnin
Send message
Joined: 1 Feb 17
Posts: 27
Credit: 311,202,073
RAC: 681,256
Message 5924 - Posted: 29 Mar 2020, 2:02:53 UTC

I haven't seen it yet but I'm guessing FP32?

Also what makes this run so much faster on Turing cards?

I see credit/runtime has doubled to 3k per task.

Profile rebirther
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 2 Jan 13
Posts: 7232
Credit: 42,729,227
RAC: 31
Message 5925 - Posted: 29 Mar 2020, 5:20:15 UTC - in response to Message 5924.

I haven't seen it yet but I'm guessing FP32?

Also what makes this run so much faster on Turing cards?

I see credit/runtime has doubled to 3k per task.


There is a higher bit lvl now and took twice in time.

KEP
Volunteer tester
Send message
Joined: 28 Nov 14
Posts: 88
Credit: 1,102,770
RAC: 0
Message 5927 - Posted: 29 Mar 2020, 8:46:11 UTC - in response to Message 5923.

I am cool. You're reading it wrong. I'm trying to help get this figured out. It was said that the app will use both GPUs to run 1 WU and it wasn't and I'm trying to help figure out why. It's nothing against Rebirther, so don't make it so.

Either way, I figured out how to run on both cards by using my own method. After I run out of Seti WUs I'll be back to kick some tail :)


Okay, then I appologize. Since you found a solution, would you mind posting it here or have you already given it to Reb? ... nice job, sure looking forward to see how much tail you can kick :)

DeleteNull
Volunteer developer
Volunteer tester
Send message
Joined: 29 Nov 14
Posts: 79
Credit: 353,382,922
RAC: 644,911
Message 5929 - Posted: 29 Mar 2020, 9:50:16 UTC - in response to Message 5868.

Nvidia on Linux

<message>
process exited with code 195 (0xc3, -61)</message>
<stderr_txt>
22:54:06 (28652): wrapper (7.2.26012): starting
22:54:06 (28652): wrapper: running ./mfaktc.exe ( --device 0)
./mfaktc.exe: error while loading shared libraries: libcudart.so.10.1: cannot open shared object file: No such file or directory
22:54:07 (28652): ./mfaktc.exe exited; CPU time 0.001267
22:54:07 (28652): app exit status: 0x7f00
22:54:07 (28652): called boinc_finish
</stderr_txt>

This code tries running an .exe on Linux! Can be done under WINE (emulator), but not native.


The file ending ".exe" or ".com" doesn't matter in Linux.
The file must have the executable flag x. And it has.
-rwxrwxrwx 1 wolfgang users 2715824 4. Sep 2019 mfaktc.exe

Just copy the file libcudart.so.10.1 into your /lib64 folder (or whatever your distribution needs)

mmonnin
Send message
Joined: 1 Feb 17
Posts: 27
Credit: 311,202,073
RAC: 681,256
Message 5931 - Posted: 29 Mar 2020, 11:27:59 UTC - in response to Message 5925.

I haven't seen it yet but I'm guessing FP32?

Also what makes this run so much faster on Turing cards?

I see credit/runtime has doubled to 3k per task.


There is a higher bit lvl now and took twice in time.


You commented on the part that wasn't even a question...

Profile rebirther
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 2 Jan 13
Posts: 7232
Credit: 42,729,227
RAC: 31
Message 5938 - Posted: 29 Mar 2020, 16:56:30 UTC - in response to Message 5931.

I haven't seen it yet but I'm guessing FP32?

Also what makes this run so much faster on Turing cards?

I see credit/runtime has doubled to 3k per task.


There is a higher bit lvl now and took twice in time.


You commented on the part that wasn't even a question...


For the other ones I dont have an answer :/

Profile KPX
Send message
Joined: 7 Dec 14
Posts: 2
Credit: 25,917,824
RAC: 0
Message 5939 - Posted: 29 Mar 2020, 17:04:03 UTC

How long is the unit supposed to run? It's been running for 2 days on my notebook with Radeon Vega 10... Should I kill the unit?

Profile rebirther
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 2 Jan 13
Posts: 7232
Credit: 42,729,227
RAC: 31
Message 5940 - Posted: 29 Mar 2020, 17:08:56 UTC - in response to Message 5939.

How long is the unit supposed to run? It's been running for 2 days on my notebook with Radeon Vega 10... Should I kill the unit?


Take a look on the first line in your stderr.txt so you know how long it will be run but the deadline is 3d

mmonnin
Send message
Joined: 1 Feb 17
Posts: 27
Credit: 311,202,073
RAC: 681,256
Message 5941 - Posted: 29 Mar 2020, 17:53:56 UTC - in response to Message 5938.

I haven't seen it yet but I'm guessing FP32?

Also what makes this run so much faster on Turing cards?

I see credit/runtime has doubled to 3k per task.


There is a higher bit lvl now and took twice in time.


You commented on the part that wasn't even a question...


For the other ones I dont have an answer :/


I doubt its FP64 or my Radeon VII would be much better than the Turing cards. Maybe its the GDDR6 or if its just FP16 it could make the Turing cards much better than Pascal.

Dirk Broer
Send message
Joined: 2 Jan 15
Posts: 71
Credit: 129,980,152
RAC: 80,689
Message 5975 - Posted: 31 Mar 2020, 0:20:28 UTC - in response to Message 5929.
Last modified: 31 Mar 2020, 0:25:21 UTC

Just copy the file libcudart.so.10.1 into your /lib64 folder (or whatever your distribution needs)


Last time 'just copy this to that' worked for me was under DOS. This is not DOS, this is Linux. libcudart.so.10.1 is totally unknown yet in the repo that supplies my software.

bayanne
Send message
Joined: 16 Mar 20
Posts: 2
Credit: 0
RAC: 0
Message 5977 - Posted: 31 Mar 2020, 10:42:54 UTC
Last modified: 31 Mar 2020, 10:43:59 UTC

I have been running BOIC for many years on my MAC, but even though I have CUDA set up and running fine with GPU72, I cannot get SRBase to even get an exponent to test.

Am I missing something blindingly obvious?

I have selected NVIDIA and AMD in the properties, as well as GPU72 as an application

Mac OSX 10.11.6
CUDA Driver Version 8.0.90
GPU Driver Version 10.10.14 310.42.25f02
NVIDIA GeForce GTX 775M 2048 MB

Profile rebirther
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 2 Jan 13
Posts: 7232
Credit: 42,729,227
RAC: 31
Message 5978 - Posted: 31 Mar 2020, 10:58:47 UTC - in response to Message 5977.

I have been running BOIC for many years on my MAC, but even though I have CUDA set up and running fine with GPU72, I cannot get SRBase to even get an exponent to test.

Am I missing something blindingly obvious?

I have selected NVIDIA and AMD in the properties, as well as GPU72 as an application

Mac OSX 10.11.6
CUDA Driver Version 8.0.90
GPU Driver Version 10.10.14 310.42.25f02
NVIDIA GeForce GTX 775M 2048 MB


There is no cuda app setup. If you have an app let me know. I found some sourcecode for mfaktc but must be compile first.

Profile rebirther
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 2 Jan 13
Posts: 7232
Credit: 42,729,227
RAC: 31
Message 5986 - Posted: 2 Apr 2020, 19:30:11 UTC

I have removed the opencl apps for linux and mac for the meanwhile until we got a fix (standalone tests are needed)

Previous · 1 . . . 4 · 5 · 6 · 7 · 8 · 9 · 10 . . . 23 · Next
Post to thread

Message boards : Number crunching : Trial Factoring


Main page · Your account · Message boards


Copyright © 2014-2024 BOINC Confederation / rebirther