Trial Factoring tests
log in

Advanced search

Message boards : Number crunching : Trial Factoring tests

1 · 2 · 3 · 4 . . . 7 · Next
Author Message
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 5926 - Posted: 29 Mar 2020, 7:07:32 UTC
Last modified: 29 Mar 2020, 9:51:47 UTC

This thread is for testing the ERROR: unknown option '--device' and Multi-GPU using the app_config.xml from the FAQ

Pls post your feedback or suggestions to change something in the app_config.xml file

Multi-GPU is only supported by the mfaktc app (CUDA)

If you can run the opencl app in linux standalone post the result of command
ldd mfakto-x64 here. If it runs in standlone it will also run with the wrapper.

DeleteNull
Volunteer developer
Volunteer tester
Send message
Joined: 29 Nov 14
Posts: 79
Credit: 353,525,022
RAC: 650,683
Message 5930 - Posted: 29 Mar 2020, 11:24:36 UTC - in response to Message 5926.

Firt of all "ldd":
wolfgang@ryzen3700x:~/Downloads/mfact/mfakto> ldd mfakto-x64
linux-vdso.so.1 (0x00007ffc00020000)
libOpenCL.so.1 => /usr/lib64/libOpenCL.so.1 (0x00007efbff4a2000)
libstdc++.so.6 => /usr/lib64/libstdc++.so.6 (0x00007efbff2bf000)
libm.so.6 => /lib64/libm.so.6 (0x00007efbff179000)
libgcc_s.so.1 => /lib64/libgcc_s.so.1 (0x00007efbff15f000)
libc.so.6 => /lib64/libc.so.6 (0x00007efbfef99000)
libdl.so.2 => /lib64/libdl.so.2 (0x00007efbfef94000)
libpthread.so.0 => /lib64/libpthread.so.0 (0x00007efbfef70000)
/lib64/ld-linux-x86-64.so.2 (0x00007efbff6d4000)

Second: you have to specify platform and device, ... for me:
-d 11 (only one platform: NVIDIA, and only one GPU RTX2070)
It seems that numbering starts with 1 (not 0).

But:
mfakto-x64 makes an inline compilation and this fails with an error:
Don't understand command line argument "-O3"!

The complete output is this:
wolfgang@ryzen3700x:~/Downloads/mfact/mfakto> ./mfakto-x64 -tf 3321932839 32 95 -d 11 -v 2
mfakto 0.14 (64bit build)


Runtime options
Inifile mfakto.ini
Verbosity 2
SieveOnGPU yes
MoreClasses yes
GPUSievePrimes 82486
GPUSieveProcessSize 24Ki bits
GPUSieveSize 96Mi bits
FlushInterval 8
WorkFile worktodo.txt
ResultsFile results.txt
Checkpoints enabled
CheckpointDelay 300s
Stages enabled
StopAfterFactor class
PrintMode compact
V5UserID none
ComputerID none
ProgressHeader "Date Time | class Pct | time ETA | GHz-d/day Sieve Wait"
ProgressFormat "%d %T | %C %p%% | %t %e | %g %s %W%%"
TimeStampInResults no
VectorSize 2
GPUType AUTO
SmallExp no
UseBinfile mfakto_Kernels.elf
Compiletime options
Select device - Get device info - Device 1/1: GeForce RTX 2070 (NVIDIA Corporation),
device version: OpenCL 1.2 CUDA, driver version: 440.64
Extensions: cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_fp64 cl_khr_byte_addressable_store cl_khr_icd cl_khr_gl_sharing cl_nv_compiler_options cl_nv_device_attribute_query cl_nv_pragma_unroll cl_nv_copy_opts cl_nv_create_buffer cl_khr_int64_base_atomics cl_khr_int64_extended_atomics
Global memory:8368685056, Global memory cache: 1179648, local memory: 49152, workgroup size: 1024, Work dimensions: 3[1024, 1024, 64, 0, 0] , Max clock speed:1830, compute units:36
Compiling kernels (build options: "-I. -DVECTOR_SIZE=2 -O3 -DMORE_CLASSES -DCL_GPU_SIEVE").
BUILD OUTPUT
Error in processing command line: Don't understand command line argument "-O3"!
END OF BUILD OUTPUT
Error -11 (Build program failure): clBuildProgram
ERROR: init_CL(3, 11) failed

Stef42
Send message
Joined: 22 Dec 14
Posts: 15
Credit: 66,697,537
RAC: 87,293
Message 5932 - Posted: 29 Mar 2020, 11:54:33 UTC - in response to Message 5930.

Why run mfakto on a NVIDIA GPU?! That won't work anyway. Mfakto is AMD-GPU only, mfaktc is NVIDIA GPU only.

DeleteNull
Volunteer developer
Volunteer tester
Send message
Joined: 29 Nov 14
Posts: 79
Credit: 353,525,022
RAC: 650,683
Message 5933 - Posted: 29 Mar 2020, 12:01:27 UTC - in response to Message 5932.

mfakto runs on OpenCl devices. That can be AMD/ATI, NVidia, Intel and also some CPU.

And RTX2070 is capable to run CUDA and OpenCL-Code

Stef42
Send message
Joined: 22 Dec 14
Posts: 15
Credit: 66,697,537
RAC: 87,293
Message 5934 - Posted: 29 Mar 2020, 12:20:55 UTC - in response to Message 5933.

Are you sure? The readme states only to support AMD GPU's, since the kernels used are made for AMD GPU's specifically.

zombie67 [MM]
Avatar
Send message
Joined: 4 Dec 14
Posts: 30
Credit: 1,080,232,404
RAC: 3,936,650
Message 5935 - Posted: 29 Mar 2020, 14:46:30 UTC
Last modified: 29 Mar 2020, 15:15:21 UTC

With regard to the app_config.xml file, to run one pask per GPU:

1) I am getting the following error in the event log, when BOINC reads the app_config.xml file:

Sun 29 Mar 2020 07:30:41 AM PDT | SRBase | Entry in app_config.xml for app 'GPU72', plan class '' doesn't match any app versions


2) Even with the error message, it seems to be working. I am running one task on each GPU, where it used to run both tasks on just one of the GPUs.

If you want host details, it is this one.

Edit: It was working for the first two tasks. But once they completed, and started new tasks, it went back to running both tasks on a single GPU.
____________
Reno, NV
Team: SETI.USA

Sergey Kovalchuk
Send message
Joined: 27 Oct 15
Posts: 12
Credit: 5,358,644
RAC: 0
Message 5936 - Posted: 29 Mar 2020, 15:53:49 UTC - in response to Message 5935.

<app_config>
<app_version>
<app_name>GPU72</app_name>
<plan_class>cuda100</plan_class> # 1
<cmdline>-d c</cmdline> # 2 - try multi-GPU, 2 tasks * 2 GPU
</app_version>
</app_config>

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 5942 - Posted: 29 Mar 2020, 17:59:41 UTC - in response to Message 5935.
Last modified: 29 Mar 2020, 18:08:24 UTC

With regard to the app_config.xml file, to run one pask per GPU:

1) I am getting the following error in the event log, when BOINC reads the app_config.xml file:

Sun 29 Mar 2020 07:30:41 AM PDT | SRBase | Entry in app_config.xml for app 'GPU72', plan class '' doesn't match any app versions


2) Even with the error message, it seems to be working. I am running one task on each GPU, where it used to run both tasks on just one of the GPUs.

If you want host details, it is this one.

Edit: It was working for the first two tasks. But once they completed, and started new tasks, it went back to running both tasks on a single GPU.


The error should be gone after you have reported a result and the issue give me some headaches.

edit:
from the readme, the mfaktc program can only run a single instance per GPU, so it cant work, try the first option to run all GPUs on one WU

zombie67 [MM]
Avatar
Send message
Joined: 4 Dec 14
Posts: 30
Credit: 1,080,232,404
RAC: 3,936,650
Message 5945 - Posted: 29 Mar 2020, 18:59:47 UTC - in response to Message 5936.

<app_config>
<app_version>
<app_name>GPU72</app_name>
<plan_class>cuda100</plan_class> # 1
<cmdline>-d c</cmdline> # 2 - try multi-GPU, 2 tasks * 2 GPU
</app_version>
</app_config>


I am having trouble understanding what you mean with #2. I am already running two tasks, as I have two GPUs. But both tasks are running on just one of the GPUs*. What are you suggesting "c" should be, to get one task per GPU? It is currently set to "1", as it says in the FAQ.

Note: BOINCmgr shows one task running on GPU0 and one task on GPU1, even though both tasks are actually running on just one of the GPUs.
____________
Reno, NV
Team: SETI.USA

[AF>Libristes] ElGuillermo
Send message
Joined: 12 Jan 15
Posts: 1
Credit: 6,534,513
RAC: 0
Message 5946 - Posted: 29 Mar 2020, 19:10:52 UTC

[copied from GPU72 thread]

Hi !

Some "old" hardware here :
01:00.0 VGA compatible controller: Advanced Micro Devices, Inc. [AMD/ATI] Cape Verde XT [Radeon HD 7770/8760 / R7 250X]

Seen in Linux Mint 19.3 as:
OpenCL: AMD/ATI GPU 0: AMD VERDE (DRM 2.50.0, 5.3.0-28-generic, LLVM 9.0.0) (driver version 19.2.8, device version OpenCL 1.1 Mesa 19.2.8, 2044MB, 2044MB available, 800 GFLOPS peak)


All the WUs go in error at start :

<core_client_version>7.9.3</core_client_version> <![CDATA[ <message> process exited with code 195 (0xc3, -61)</message> <stderr_txt> 20:15:06 (16011): wrapper (7.2.26012): starting 20:15:06 (16011): wrapper: running mfakto-x64 ( --device 0) ERROR: unknown option '--device' 20:15:07 (16011): mfakto-x64 exited; CPU time 0.002208 20:15:07 (16011): app exit status: 0x100 20:15:07 (16011): called boinc_finish </stderr_txt> ]]>


I've tried the fix suggested ...to no avail :/
<app_config> <app> <name>GPU72</name> <max_concurrent>1</max_concurrent> <gpu_versions> <gpu_usage>1</gpu_usage> <cpu_usage>0.01</cpu_usage> </gpu_versions> </app> <app_version> <app_name>GPU72</app_name> <plan_class>opencl_ati_101</plan_class> <cmdline>-d 0</cmdline> </app_version> </app_config>



Here's the output from command "./mfakto-x64 -d 0 -st" :
mfakto 0.14 (64bit build) Runtime options Inifile mfakto.ini Verbosity 1 SieveOnGPU yes MoreClasses yes GPUSievePrimes 82486 GPUSieveProcessSize 24Ki bits GPUSieveSize 96Mi bits FlushInterval 8 WorkFile worktodo.txt ResultsFile results.txt Checkpoints enabled CheckpointDelay 300s Stages enabled StopAfterFactor class PrintMode compact V5UserID none ComputerID none TimeStampInResults no VectorSize 2 GPUType AUTO SmallExp no UseBinfile mfakto_Kernels.elf Compiletime options Select device - Error: No platform found ERROR: init_CL(3, 0) failed


And the output from command "./mfakto-x64 -st"
mfakto 0.14 (64bit build) Runtime options Inifile mfakto.ini Verbosity 1 SieveOnGPU yes MoreClasses yes GPUSievePrimes 82486 GPUSieveProcessSize 24Ki bits GPUSieveSize 96Mi bits FlushInterval 8 WorkFile worktodo.txt ResultsFile results.txt Checkpoints enabled CheckpointDelay 300s Stages enabled StopAfterFactor class PrintMode compact V5UserID none ComputerID none TimeStampInResults no VectorSize 2 GPUType AUTO SmallExp no UseBinfile mfakto_Kernels.elf Compiletime options Select device - Error: No platform found ERROR: init_CL(3, 0) failed



Ask for any details if needed :)

Sergey Kovalchuk
Send message
Joined: 27 Oct 15
Posts: 12
Credit: 5,358,644
RAC: 0
Message 5948 - Posted: 29 Mar 2020, 19:12:27 UTC - in response to Message 5945.

mfakto has option "-d c" - force using all CPUs
mfaktc does not have such an option,
so my tip is past the target :(

bluestang
Send message
Joined: 6 Jun 19
Posts: 59
Credit: 1,268,237,570
RAC: 0
Message 5949 - Posted: 29 Mar 2020, 20:24:06 UTC - in response to Message 5948.

"-d c" on AMD GPUs will error out in a couple seconds. It does not work. Also, "-d 0" or "-d 1" do nothing on AMD GPUs in my system (2x RX 580) as to which GPU to use. It doesn't even use the 1st GPU in the system. The app is using the 2nd GPU.

There are serious issues/limitations with this app. Especially for AMD GPUs.

Also, I have yet to have any of the above command line switches do anything for NVIDIA GPUs either as far as letting the app run 1 WU on both GPUs. It doesn't work.

This app needs to be set up to run 1 WU per GPU plain and simple and its not for some reason.

I'm not bitching really, I'm just trying to give info to get this sorted.

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 5950 - Posted: 29 Mar 2020, 20:40:07 UTC - in response to Message 5949.
Last modified: 29 Mar 2020, 20:58:50 UTC

If nothing is work correctly on multi GPU then you can only set 1 GPU in cc_config file but you want of course use all for other projects too.

Update:
there is a exclude_gpu option for cc_config which can be defined for which app the second GPU is not running but can run on another project. I will post this later.

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 5952 - Posted: 29 Mar 2020, 21:06:24 UTC

I have changed one line in app_config to run only one WU max, maybe this could help or not.

Gigacruncher [TSBTs Pirate]
Send message
Joined: 28 Mar 20
Posts: 48
Credit: 8,419,360
RAC: 0
Message 5953 - Posted: 29 Mar 2020, 21:10:57 UTC

Reb,

You need launching multiple mfakto instances to get advantage of multi GPU.
Read:

https://www.mersenneforum.org/showpost.php?p=526674&postcount=1538

https://www.mersenneforum.org/showpost.php?p=533736&postcount=1573

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 5954 - Posted: 29 Mar 2020, 21:15:56 UTC

If the app_config is not working try this here:

<exclude_gpu> <url>http://srbase.my-firewall.org/sr5/</url> <device_num>1</device_num> <type>NVIDIA|ATI</type> <app>GPU72</app> </exclude_gpu>


change the type what you have in your PC and put this code in the option section of cc_config

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 5955 - Posted: 29 Mar 2020, 21:18:39 UTC - in response to Message 5953.

Reb,

You need launching multiple mfakto instances to get advantage of multi GPU.
Read:

https://www.mersenneforum.org/showpost.php?p=526674&postcount=1538

https://www.mersenneforum.org/showpost.php?p=533736&postcount=1573


Thx but not possible in this server config.

bluestang
Send message
Joined: 6 Jun 19
Posts: 59
Credit: 1,268,237,570
RAC: 0
Message 5956 - Posted: 29 Mar 2020, 21:19:38 UTC - in response to Message 5955.

Reb,

You need launching multiple mfakto instances to get advantage of multi GPU.
Read:

https://www.mersenneforum.org/showpost.php?p=526674&postcount=1538

https://www.mersenneforum.org/showpost.php?p=533736&postcount=1573


Thx but not possible in this server config.


What about on the client side?

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 5957 - Posted: 29 Mar 2020, 21:27:56 UTC - in response to Message 5956.
Last modified: 29 Mar 2020, 21:29:27 UTC

Reb,

You need launching multiple mfakto instances to get advantage of multi GPU.
Read:

https://www.mersenneforum.org/showpost.php?p=526674&postcount=1538

https://www.mersenneforum.org/showpost.php?p=533736&postcount=1573


Thx but not possible in this server config.


What about on the client side?


no, doesnt work. All the work which sending out is server related. The code must be changed in the app itself.

bluestang
Send message
Joined: 6 Jun 19
Posts: 59
Credit: 1,268,237,570
RAC: 0
Message 5958 - Posted: 29 Mar 2020, 21:34:05 UTC - in response to Message 5957.

Ok, thanks. Also, appreciate the quick replies from you.

Now, take a few days off to relax and then get back to work fixing the app lol :)

1 · 2 · 3 · 4 . . . 7 · Next
Post to thread

Message boards : Number crunching : Trial Factoring tests


Main page · Your account · Message boards


Copyright © 2014-2024 BOINC Confederation / rebirther