2nd GPU not seen (revisited)


log in

Advanced search

Questions and Answers : GPU applications : 2nd GPU not seen (revisited)

Author Message
jay_e
Send message
Joined: 6 Apr 03
Posts: 42
Credit: 198,361
RAC: 14
United States
Message 1459451 - Posted: 1 Jan 2014, 3:47:06 UTC
Last modified: 1 Jan 2014, 3:49:16 UTC

Greetings!

"It seemed like a good idea at the time." And then..

I have been using an ATI 7750 GPU and successfully crunching.
Since I used the 7750 as my video device, I thought I would get an older
graphics card (Nvidia 8400GS) for video and free up all of the 7750 for crunching.

I did read the helpful posts here about editing the cc_config.xml.
I put in the "use all" flag. (it is in the BOINC data directory. (all RW permissions)

Here it is:


<!--
This is a minimal configuration file cc_config.xml of the BOINC core client.
For a complete list of all available options and logging flags and their
meaning see: http://boinc.berkeley.edu/trac/wiki/ClientMessages
-->
<cc_config>
<log_flags>
<task>1</task>
<file_xfer>1</file_xfer>
<sched_ops>1</sched_ops>
</log_flags>
<options>
<use_all_gpus>1</use_all_gpus>
</options>
</cc_config>


Environment: Linux - Ubuntu Studio

Tue 31 Dec 2013 09:38:05 PM EST | | Starting BOINC client version 7.0.65 for x86_64-pc-linux-gnu
Tue 31 Dec 2013 09:38:05 PM EST | | log flags: file_xfer, sched_ops, task
Tue 31 Dec 2013 09:38:05 PM EST | | Libraries: libcurl/7.29.0 OpenSSL/1.0.1c zlib/1.2.7 libidn/1.25 librtmp/2.3
Tue 31 Dec 2013 09:38:05 PM EST | | Data directory: /var/lib/boinc-client
Tue 31 Dec 2013 09:38:05 PM EST | | Processor: 8 AuthenticAMD AMD FX(tm)-8150 Eight-Core Processor [Family 21 Model 1 Stepping 2]
Tue 31 Dec 2013 09:38:05 PM EST | | Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good nopl nonstop_tsc extd_apicid aperfmperf pni pclmulqdq monitor ssse3 cx16 sse4_1 sse4_2 popcnt aes xsave avx lahf_lm cmp_legacy svm extapic cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw ibs xop skinit wdt lwp fma4 nodeid_msr topoext perfctr_core arat cpb hw_pstate npt lbrv svm_lock nrip_save tsc_scale vmcb_clean flushbyasid decodeassists pausefilter pfthreshold
Tue 31 Dec 2013 09:38:05 PM EST | | OS: Linux: 3.8.0-34-lowlatency
Tue 31 Dec 2013 09:38:05 PM EST | | Memory: 7.70 GB physical, 9.31 GB virtual
Tue 31 Dec 2013 09:38:05 PM EST | | Disk: 9.61 GB total, 7.12 GB free
Tue 31 Dec 2013 09:38:05 PM EST | | Local time is UTC -5 hours
Tue 31 Dec 2013 09:38:05 PM EST | | No usable GPUs found
Tue 31 Dec 2013 09:38:05 PM EST | | app version refers to missing GPU type ATI
Tue 31 Dec 2013 09:38:05 PM EST | Einstein@Home | Application uses missing ATI GPU
Tue 31 Dec 2013 09:38:05 PM EST | | app version refers to missing GPU type ATI
Tue 31 Dec 2013 09:38:05 PM EST | Einstein@Home | Application uses missing ATI GPU
Tue 31 Dec 2013 09:38:05 PM EST | Einstein@Home | Found app_config.xml
Tue 31 Dec 2013 09:38:05 PM EST | | Config: use all coprocessors
Tue 31 Dec 2013 09:38:05 PM EST | | Config: GUI RPCs allowed from:
Tue 31 Dec 2013 09:38:05 PM EST | World Community Grid | URL http://www.worldcommunitygrid.org/; Computer ID 2635735; resource share 20
Tue 31 Dec 2013 09:38:05 PM EST | Einstein@Home | URL http://einstein.phys.uwm.edu/; Computer ID 9846906; resource share 20
Tue 31 Dec 2013 09:38:05 PM EST | malariacontrol.net | URL http://www.malariacontrol.net/; Computer ID 731405; resource share 20
Tue 31 Dec 2013 09:38:05 PM EST | LHC@home 1.0 | URL http://lhcathomeclassic.cern.ch/sixtrack/; Computer ID 10310841; resource share 20
Tue 31 Dec 2013 09:38:05 PM EST | SETI@home Beta Test | URL http://setiweb.ssl.berkeley.edu/beta/; Computer ID 69470; resource share 20


Is it important that the 'use all' statement is listed after the log about missing GPU ???

I also think that it is weird that the log does not have an GPUapp-confi-relatedentry from Seti Beta similar to that of Einstein. ( I also had a recent problem of no finding set-beta work for ATI. (Perhaps none is available.)
I was careful to not wipe out ATi drivers when I tried to add Nvidia video drivers.
The Ubuntu fglrx (for OpenCL) is still there.
'i' means installed

-- ATI drivers
i A fglrx-amdcccle-updates - Catalyst Control Center for the AMD graphics accelerators
i fglrx-updates - Video driver for the AMD graphics accelerators

-- nvidia drivers (just used for video)
i nvidia-common - transitional package for ubuntu-drivers-common


I used the 'lspci' unix command that shows what devices the OS sees.
it gave:
01:00.0 VGA compatible controller: NVIDIA Corporation G98 [GeForce 8400 GS Rev. 2] (rev a1)
and
06:00.0 VGA compatible controller: Advanced Micro Devices [AMD] nee ATI Cape Verde PRO [Radeon HD 7750]


I am stumped. Please help.
Should I just take out the Nvidia 8400GS?
Need any logging?

THANKS in advance,
Jay

Happy New Year!
[edit: fix typo.]

OzzFan
Volunteer tester
Avatar
Send message
Joined: 9 Apr 02
Posts: 13542
Credit: 29,405,219
RAC: 15,935
United States
Message 1459469 - Posted: 1 Jan 2014, 5:08:01 UTC - in response to Message 1459451.

A couple of tips:

1) The nVidia 8400 is not a CUDA compatible device, so it won't crunch. If I remember correctly, an 8800 or later is required.

2) Because BOINC doesn't detect a usable first GPU, it is unable to work with any GPU devices (might be a bug, don't know that it's worth fixing given the age of non-CUDA complaint GPUs).
____________

jay_e
Send message
Joined: 6 Apr 03
Posts: 42
Credit: 198,361
RAC: 14
United States
Message 1459488 - Posted: 1 Jan 2014, 7:07:46 UTC - in response to Message 1459469.
Last modified: 1 Jan 2014, 7:15:12 UTC

Greetings!

Thanks for the tips - but they didn't apply.
I did not want to use the 8400 as a computing device for BOINC.
I want to use it only as a display device.
The other card, an ATI 7750 ( with 2GB ram) is the Computing device for BOINC.

Before I added the 8400, it crunched fine.
After I added the 8400, Boinc can no longer see the 7750,
although it is there.

The idea is/was to not use the 7750 as a display device, (use the 8400 for display)
thus
*possibly* freeing up the 7750 to work faster, or to be able to to do 2 WU at once.

if you would like, look at
http://einstein.phys.uwm.edu/workunit.php?wuid=181252573
This is an example where the 7750 ran slower than a GTX 460.

Now, BOINC does not see the 7750 at all.

Jay

[Edit fix typo. Add statement below]
I should say - that in the web-pages where a person sets up the project preferences, I checked 'use ATI GPU' and not 'use NVIDIA GPU'.

OzzFan
Volunteer tester
Avatar
Send message
Joined: 9 Apr 02
Posts: 13542
Credit: 29,405,219
RAC: 15,935
United States
Message 1459494 - Posted: 1 Jan 2014, 7:24:41 UTC - in response to Message 1459488.
Last modified: 1 Jan 2014, 7:30:30 UTC

If that's the case:

1) You do not need to have <use_all_gpus>1</use_all_gpus> in your cc_config.xml. That option tells BOINC to crunch on all devices.

2) The crunching device must be the primary display device. If you want to have a lesser-powerful graphics card for basic display, it should be the secondary.

3) Having a separate display device will not necessarily free up the crunching device by giving it more resources. You simply need to create an app_config.xml, and preferably download the most recent version of BOINC to make sure app_config.xml works correctly.

For reference, I have an ATi Radeon HD 6870 that crunches two workunits at a time and has no problem with most basic graphics functions. I do have to disable BOINC while playing graphics intensive games on that machine though. I would imagine your GPU should be just as capable.

Profile Ageless
Avatar
Send message
Joined: 9 Jun 99
Posts: 12259
Credit: 2,553,709
RAC: 770
Netherlands
Message 1459685 - Posted: 1 Jan 2014, 18:24:26 UTC - in response to Message 1459469.

@Charlie,

1) The nVidia 8400 is not a CUDA compatible device, so it won't crunch. If I remember correctly, an 8800 or later is required.

You remember incorrectly. All Geforce 8xxx are CUDA capable. See https://developer.nvidia.com/cuda-gpus for the full list.

@jay_e,
- You made sure to reinstall the AMD drivers after you installed the Nvidia drivers? Especially with OpenCL, the drivers installed by the latest hardware capable of OpenCL will overwrite the drivers of any previous hardware, which may break detection capability of the previous hardware.
- You left a monitor attached to the AMD card? Or else a VGA dummy? What happens when you reattach the monitor to the AMD device?
____________
Jord

Fighting for the correct use of the apostrophe, together with Weird Al Yankovic

jay_e
Send message
Joined: 6 Apr 03
Posts: 42
Credit: 198,361
RAC: 14
United States
Message 1460081 - Posted: 2 Jan 2014, 20:27:55 UTC - in response to Message 1459685.

Thanks Jord,

Linux is pretty good at telling you what you are about to 'break' when installing.
Since I wanted the NVidia just for display, I got the simplest drives that did not break the ATI drivers.

But I definitely trust you and reinstalled the ATI drivers.
( and reset ati to fill in an Xorg stanza.)
I also reset & removed all projects, uninstalled all boinc, and deleted the Boinc data dir. Then reinstalled boinc. put in a cc-config with use all CPU, and
app_configs for different project with app names from the newly loaded client-state.

You found the problem.
Interesting stuff with 2 cards and one display.
I read that the dummy plug was needed for windows.
( did not expect that for linux since the lspci showed both cards when only one connection.)

Using one cable to display, and no dummy plug, BOINC only sees the video card that is connected.

So.
I connected a vga cable between the display and the Old Nvidia card, and
a DVI-I (single) cable between the ATI card and on a 2nd port on the same display.

At boot, my pc recognized both cards. But I only used the VGA connection.

I did a shutdown and switched ports on the display.
One-at-a-time, both ports and cards work.

BOINC now sees the ATI card.
That was the problem.
Thank you!!

The purpose of this is to test if I get *any* better performance when
offloading the display work from the ATI card.
( was getting less-than-expected performance from the ATI card.)

Will now try to get some set ATI WU and look up wingman's speeds.
Which is best comparison? Run time or CPU time?

Here is an example
http://setiweb.ssl.berkeley.edu/beta/workunit.php?wuid=5912795

Using Running times:


  • an Nvidia (Coprocessors [4] NVIDIA GeForce GTX 660 (2048MB) driver: 331.58 OpenCL: 1.01) -- 742 seconds
  • my 7750 -- 4,118 seconds
  • a cpu -- 13,117 seconds



The 7750 has 2GB of memory. I'm not sure how the double precision is implemented. (perhaps slowly :-) )


Any insights?


Thanks again for the tip about the dummy plug.

Jay

OzzFan
Volunteer tester
Avatar
Send message
Joined: 9 Apr 02
Posts: 13542
Credit: 29,405,219
RAC: 15,935
United States
Message 1460090 - Posted: 2 Jan 2014, 20:55:27 UTC - in response to Message 1460081.

The 7750 has 2GB of memory. I'm not sure how the double precision is implemented. (perhaps slowly :-)


It is my understanding that SETI uses single precision floating point only.

Profile arkayn
Volunteer tester
Avatar
Send message
Joined: 14 May 99
Posts: 3595
Credit: 47,370,649
RAC: 3,276
United States
Message 1460182 - Posted: 3 Jan 2014, 2:48:27 UTC - in response to Message 1460090.

The 7750 has 2GB of memory. I'm not sure how the double precision is implemented. (perhaps slowly :-)


It is my understanding that SETI uses single precision floating point only.


Yes it does, but the 7750 has 1/16th double precision to single precision.
____________

Questions and Answers : GPU applications : 2nd GPU not seen (revisited)

Copyright © 2014 University of California