2 GPUs One Case? - I'm not seeing something

[litehouse43]

Hey Guys,
Re-configuring some hardware combinations and decided to put my two 660 Ti's in the same case. Having some problems, been troubleshooting all morning and I'm still missing something. Checked all the seating of the cards, power supply is more than sufficient, updated drivers with both cards in, cc_config file is posted below (think it's fine), and posted my event log. I have 4 GPUGRID tasks waiting and the one running does say NIVIDIA (Device 0) so it's recognizing the second. GPU-Z is also showing the second. Any suggestions? I'm clearly missing something.

2/1/2014 3:30:39 PM | | Starting BOINC client version 7.2.33 for windows_x86_64
2/1/2014 3:30:39 PM | | log flags: file_xfer, sched_ops, task
2/1/2014 3:30:39 PM | | Libraries: libcurl/7.25.0 OpenSSL/1.0.1 zlib/1.2.6
2/1/2014 3:30:39 PM | | Data directory: C:\ProgramData\BOINC
2/1/2014 3:30:39 PM | | Running under account Jack Main
2/1/2014 3:30:39 PM | | CUDA: NVIDIA GPU 0: GeForce GTX 660 Ti (driver version 332.21, CUDA version 6.0, compute capability 3.0, 2048MB, 1903MB available, 2985 GFLOPS peak)
2/1/2014 3:30:39 PM | | CUDA: NVIDIA GPU 1: GeForce GTX 660 Ti (driver version 332.21, CUDA version 6.0, compute capability 3.0, 2048MB, 1958MB available, 2985 GFLOPS peak)
2/1/2014 3:30:39 PM | | OpenCL: NVIDIA GPU 0: GeForce GTX 660 Ti (driver version 332.21, device version OpenCL 1.1 CUDA, 2048MB, 1903MB available, 2985 GFLOPS peak)
2/1/2014 3:30:39 PM | | OpenCL: NVIDIA GPU 1: GeForce GTX 660 Ti (driver version 332.21, device version OpenCL 1.1 CUDA, 2048MB, 1958MB available, 2985 GFLOPS peak)
2/1/2014 3:30:39 PM | | Host name: JackMain-PC
2/1/2014 3:30:39 PM | | Processor: 12 GenuineIntel Intel(R) Core(TM) i7-3930K CPU @ 3.20GHz [Family 6 Model 45 Stepping 7]
2/1/2014 3:30:39 PM | | Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss htt tm pni ssse3 cx16 sse4_1 sse4_2 popcnt aes syscall nx lm vmx tm2 dca pbe
2/1/2014 3:30:39 PM | | OS: Microsoft Windows 7: Ultimate x64 Edition, Service Pack 1, (06.01.7601.00)
2/1/2014 3:30:39 PM | | Memory: 15.95 GB physical, 31.90 GB virtual
2/1/2014 3:30:39 PM | | Disk: 931.29 GB total, 877.89 GB free
2/1/2014 3:30:39 PM | | Local time is UTC -7 hours
2/1/2014 3:30:39 PM | | VirtualBox version: 4.3.6
2/1/2014 3:30:39 PM | | Config: use all coprocessors
2/1/2014 3:30:39 PM | fightmalaria@home | URL http://boinc.ucd.ie/fmah/; Computer ID 30675; resource share 100
2/1/2014 3:30:39 PM | WUProp@Home | URL http://wuprop.boinc-af.org/; Computer ID 65361; resource share 100
2/1/2014 3:30:39 PM | GPUGRID | URL http://www.gpugrid.net/; Computer ID 166965; resource share 100
2/1/2014 3:30:39 PM | World Community Grid | URL http://www.worldcommunitygrid.org/; Computer ID 2783951; resource share 5000
2/1/2014 3:30:39 PM | World Community Grid | General prefs: from World Community Grid (last modified 17-Sep-2013 23:22:34)
2/1/2014 3:30:39 PM | World Community Grid | Host location: none
2/1/2014 3:30:39 PM | World Community Grid | General prefs: using your defaults
2/1/2014 3:30:39 PM | | Preferences:
2/1/2014 3:30:39 PM | | max memory usage when active: 16331.61MB
2/1/2014 3:30:39 PM | | max memory usage when idle: 16331.61MB
2/1/2014 3:30:39 PM | | max disk usage: 100.00GB
2/1/2014 3:30:39 PM | | (to change preferences, visit a project web site or select Preferences in the Manager)
2/1/2014 3:30:39 PM | | Not using a proxy
2/1/2014 3:30:55 PM | World Community Grid | Restarting task MCM1_0001836_5656_1 using mcm1 version 728 in slot 1
2/1/2014 3:30:55 PM | World Community Grid | Restarting task MCM1_0001836_6160_1 using mcm1 version 728 in slot 10
2/1/2014 3:30:55 PM | World Community Grid | Restarting task MCM1_0001836_6606_1 using mcm1 version 728 in slot 8
2/1/2014 3:30:55 PM | WUProp@Home | Restarting task wu_v4_1389300887_520746_0 using data_collect_v4 version 413 (nci) in slot 0
2/1/2014 3:30:55 PM | World Community Grid | Restarting task MCM1_0001836_7037_1 using mcm1 version 728 in slot 4
2/1/2014 3:30:55 PM | World Community Grid | Restarting task MCM1_0001836_6272_0 using mcm1 version 728 in slot 2
2/1/2014 3:30:55 PM | World Community Grid | Restarting task MCM1_0001836_6613_1 using mcm1 version 728 in slot 13
2/1/2014 3:30:55 PM | World Community Grid | Restarting task MCM1_0001836_6432_0 using mcm1 version 728 in slot 14
2/1/2014 3:30:55 PM | World Community Grid | Restarting task MCM1_0001836_6831_1 using mcm1 version 728 in slot 15
2/1/2014 3:30:55 PM | World Community Grid | Restarting task MCM1_0001836_6789_0 using mcm1 version 728 in slot 16
2/1/2014 3:30:55 PM | World Community Grid | Restarting task MCM1_0001836_5925_1 using mcm1 version 728 in slot 17
2/1/2014 3:30:55 PM | World Community Grid | Restarting task MCM1_0001836_5545_1 using mcm1 version 728 in slot 18
2/1/2014 3:30:55 PM | World Community Grid | Restarting task MCM1_0001836_5659_0 using mcm1 version 728 in slot 5
2/1/2014 3:30:55 PM | GPUGRID | Restarting task gluilex8x70-NOELIA_DIPEPT1-1-2-RND7398_0 using acemdlong version 815 (cuda42) in slot 19



cc_config.xml

<cc_config>
<options>
<use_all_gpus>1</use_all_gpus>
</options>
</cc_config>

Thanks!
~Jack

[zombie67]

If you have CPU tasks in panic mode, it can cause your second (and beyond) GPU to stand idle. BOINC wants as many CPU threads for the CPU tasks as possible in that case.

Also, just so I am clear, is that your *whole* cc_config.xml? Or just the bit about GPUs? I want to make sure I am seeing it all.

[Fire$torm]

Some things to look out for...
*SLI/Crossfire: Somewhat dependent on project (and dependent on MB) but usually won't work or will be flaky, although I have seen reports of success.
*Dummy Plug: May be necessary on certain nVidia cards and like above is dependent on MB.
*Video Drivers (?): I seem to recall an issue some time ago about a GeForce driver that caused all manner of headache. Best to update to latest release.
*app_config.xml file: An old app_config still lingering in a project folder can cause problems.

An alternative for dual GPU systems is to have a Radeon/GeForce (mixed GPU) setup. The only caveat is that the Radeon card must be installed in the PCIe slot closest to the CPU. This is usually Slot-1. Slot-0 is almost always the "useless" PCIe X1 slot that is buried next to NB cooler or some other MB component.

[circeseye]

one thing if i remember right. it wont work one the newer ones thou. if ur running 2 video cards and u dont want them running together (crossfire/sli) do not add the bridge. then it is easier to run them with separate tasks. with the bridge on it will for some reason constantly try to use them together.

[Fire$torm]

one thing if i remember right. it wont work one the newer ones thou. if ur running 2 video cards and u dont want them running together (crossfire/sli) do not add the bridge. then it is easier to run them with separate tasks. with the bridge on it will for some reason constantly try to use them together.

yeah, what he said...

[litehouse43]

If you have CPU tasks in panic mode, it can cause your second (and beyond) GPU to stand idle. BOINC wants as many CPU threads for the CPU tasks as possible in that case.

Also, just so I am clear, is that your *whole* cc_config.xml? Or just the bit about GPUs? I want to make sure I am seeing it all.

Ya, that's all I have in that one.

Yup, no SLI/CF bridge on and no other moficiation xml files present. It's actually a completely new build.

[litehouse43]

Zombie - your the winner! WCG was in panic mode with too many CPU tasks that weren't scheduled to finish in time. Aborted half (since I was down for a few days) and now everything is running smooth. Thanks guys.

[EmSti]

You can also use app_config.xml to force a maximum number of wus on a project, then you don't have to worry about aborting jobs to regain control. Here is an example that limits leiden classical wus to 4 -

<app_config>
<app>
<name>classical</name>
<max_concurrent>4</max_concurrent>
</app>
</app_config>

App_config.xml is used in the newer boinc clients and replaces the need for app_info.xml for many items like running multiple wus per gpu. Simply add the app_config.xml to project directory and restart the boinc client. You can get the correct value for the <name> section by looking in the client_state.xml. For the leiden classical example, I looked for the project section in client_state.xml -
<project>
<master_url>http://boinc.gorlaeus.net/</master_url>
<project_name>Leiden Classical</project_name>
...

then further down the app section -

<app>
<name>classical</name>
<user_friendly_name>Classical</user_friendly_name>
<non_cpu_intensive>0</non_cpu_intensive>
</app>
<file>
<name>classical_5.56_windows_intelx86.exe</name>
<nbytes>1286144.000000</nbytes>
<max_nbytes>0.000000</max_nbytes>
<status>1</status>
<executable/>

You use the same case and spelling from the app section <name> in app_config.xml. If you get the name wrong, the bonic event log (Messages tab in BoincTasks) will warn you and show you the available application names.