I have an Alienware R7 machine that has two GPUs: one Intel 630 and one GeForce GTX 1080.

The BOINC app is running one thread for each cpu core, one for the Intel GPU and one for the GeForce GPU. According to the Windows Task Manager the CPU and the Intel GPU are each getting slammed while the NVIDIA GPU is lazily chilling between 1%-2% utilization. I do see that there is one NVIDIA thread pushing out tasks according to BOINC. I expect that BOINC isn't fully taking advantage of all of the cores on my NVIDIA GPU.

How do I know?

NVIDIA GeForce GTX 1080 driver version 398.36.

Log Snippet:
7/21/2018 8:49:23 PM | | Unrecognized tag in cc_config.xml: <dont_use_wsl>
7/21/2018 8:49:23 PM | | Starting BOINC client version 7.12.1 for windows_x86_64
7/21/2018 8:49:23 PM | | log flags: file_xfer, sched_ops, task
7/21/2018 8:49:23 PM | | Libraries: libcurl/7.47.1 OpenSSL/1.0.2g zlib/1.2.8
7/21/2018 8:49:23 PM | | Data directory: C:\ProgramData\BOINC
7/21/2018 8:49:23 PM | | Running under account jsart
7/21/2018 8:49:23 PM | | CUDA: NVIDIA GPU 0: GeForce GTX 1080 (driver version 398.36, CUDA version 9.2, compute capability 6.1, 4096MB, 3555MB available, 8876 GFLOPS peak)
7/21/2018 8:49:23 PM | | OpenCL: NVIDIA GPU 0: GeForce GTX 1080 (driver version 398.36, device version OpenCL 1.2 CUDA, 8192MB, 3555MB available, 8876 GFLOPS peak)
7/21/2018 8:49:23 PM | | OpenCL: Intel GPU 0: Intel(R) UHD Graphics 630 (driver version 22.20.16.4758, device version OpenCL 2.1, 6469MB, 6469MB available, 230 GFLOPS peak)
7/21/2018 8:49:23 PM | | OpenCL CPU: Intel(R) Core(TM) i7-8700 CPU @ 3.20GHz (OpenCL driver vendor: Intel(R) Corporation, driver version 7.5.0.550, device version OpenCL 2.1 (Build 550))
7/21/2018 8:49:23 PM | | Host name: DESKTOP-ERKQBTM
7/21/2018 8:49:23 PM | | Processor: 12 GenuineIntel Intel(R) Core(TM) i7-8700 CPU @ 3.20GHz [Family 6 Model 158 Stepping 10]
7/21/2018 8:49:23 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 fma cx16 sse4_1 sse4_2 movebe popcnt aes f16c rdrandsyscall nx lm avx avx2 vmx smx tm2 pbe fsgsbase bmi1 hle smep bmi2
7/21/2018 8:49:23 PM | | OS: Microsoft Windows 10: Core x64 Edition, (10.00.17134.00)
7/21/2018 8:49:23 PM | | Memory: 15.81 GB physical, 18.69 GB virtual
7/21/2018 8:49:23 PM | | Disk: 1.80 TB total, 1.66 TB free
7/21/2018 8:49:23 PM | | Local time is UTC -5 hours
7/21/2018 8:49:23 PM | | No WSL found.
7/21/2018 8:49:23 PM | | Config: don't compute while Wow.exe is running
7/21/2018 8:49:23 PM | | Config: don't use GPUs while Wow.exe is running
7/21/2018 8:49:23 PM | | Config: report completed tasks immediately
7/21/2018 8:49:23 PM | | Config: use all coprocessors
7/21/2018 8:49:23 PM | SETI@home | URL http://setiathome.berkeley.edu/; Computer ID 8548586; resource share 100
7/21/2018 8:49:23 PM | SETI@home | General prefs: from SETI@home (last modified 21-Jul-2018 14:58:50)
7/21/2018 8:49:23 PM | SETI@home | Computer location: home
7/21/2018 8:49:23 PM | SETI@home | General prefs: no separate prefs for home; using your defaults
7/21/2018 8:49:23 PM | | Reading preferences override file
7/21/2018 8:49:23 PM | | Preferences:
7/21/2018 8:49:23 PM | | max memory usage when active: 4047.81 MB
7/21/2018 8:49:23 PM | | max memory usage when idle: 14572.13 MB
7/21/2018 8:49:23 PM | | max disk usage: 10.00 GB
7/21/2018 8:49:23 PM | | max CPUs used: 11
7/21/2018 8:49:23 PM | | suspend work if non-BOINC CPU load exceeds 25%
7/21/2018 8:49:23 PM | | (to change preferences, visit a project web site or select Preferences in the Manager)
7/21/2018 8:49:23 PM | | Setting up project and slot directories
7/21/2018 8:49:23 PM | | Checking active tasks
7/21/2018 8:49:23 PM | | Setting up GUI RPC socket

I don't know how to delete my question, but it looks like my NVIDIA GPU is popping out tasks much faster than the other GPU or CPUs. I'm going to assume that the PhysX/CUDA engine just doesn't show up on the Windows Task manager. Sorry guys.

To run more than one GPU on a Seti at Home or other projects you need the following in a

cc_config.xml

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

The file is located in the hidden \ProgramData\BONIC sub-directory.

---

A proud member of the OFA (Old Farts Association).

To run more than one GPU on a Seti at Home or other projects you need ...

That's only needed when you have two or more GPUs of the same brand, but different models. It isn't necessary when all GPUs are the same brand and same model, or when they're all different brands.

Not needed when:
- 2x Nvidia GTX 1080
or
- 3x AMD RX Vega 64
or
- 1x Nvidia GTX 1080 and 1 Intel GPU

Needed when:
- 1 Nvidia GTX 1080 and 1 Nvidia GTX 770
or
- 2 AMD RX Vega 64 and 1 AMD HD 7870

---

To run more than one GPU on a Seti at Home or other projects you need ...

That's only needed when you have two or more GPUs of the same brand, but different models. It isn't necessary when all GPUs are the same brand and same model, or when they're all different brands.

I vaguely remember needing what I described when I was running two different GPU's (one GTX, one Quarto) so I am guessing that what I said doesn't hurt and might help.

Tom

---

A proud member of the OFA (Old Farts Association).

The other things I failed to note are:
1) Generally you don't want to use the Intel gpu. It slows most systems down.
You have a lot of computers on your list that would benefit from NOT processing Intel gpu tasks. Their RAC's will climb. True they have no gpus processing but that is still faster than having a "Intel gpu" trying to help.

2) You can run up to three tasks in parallel on most gtx 1080's.

You do that in side the "app_config.xml" by setting the gpu usage to .5 for 2 tasks or .33 for three tasks. See discussion below.

====app_config.xml======
<app_config>
<app>
<name>setiathome_v8</name>
<gpu_versions>
<gpu_usage>0.33</gpu_usage>
<cpu_usage>2</cpu_usage>
</gpu_versions>
</app>
<app>
<name>astropulse_v7</name>
<gpu_versions>
<gpu_usage>0.5</gpu_usage>
<cpu_usage>2</cpu_usage>
</gpu_versions>
</app>
</app_config>

====end of code============

Discussion.
This file will cause 2 cpus to be dedicated to each gpu task. It will cause up to 3 GPU setiathome (SOG/CUDA32 or42 or 52 in stock Seti) to be run at once.

It will cause 2 cpus to be dedicated to each astropulse. It will cause up to 2 GPU astropulses to be run at once.

Yes, there could be 2 astropulses and 1 SOG be run at once. Or 2 SOG's and 1 astropulses to be run at once. If there were no SOG's and 2 or more astropulses, two astropulses would be run at a time.

Tom

---

A proud member of the OFA (Old Farts Association).

Since the current batch of SETI applications are only capable of using one CPU setting <cpu_usage>x</cpu_usage> to anything greater than 1 is a waste of time, and may indeed be counter productive.

Also, depending on the GPU in use, the exact applications being run, and the rest of the supporting system, three concurrent tasks per GPU may not be the optimum in terms of performance. You need to see what the computation load on the GPU is, and this may not be directly related to the power draw. One tool that may be suitable to see the exact loading on the GPU is GPU-Z, others will recommend "SIV", but the data from that can be a little difficult to interpret.

---

Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?

Since the current batch of SETI applications are only capable of using one CPU setting <cpu_usage>x</cpu_usage> to anything greater than 1 is a waste of time, and may indeed be counter productive.

In another message thread it was suggested that for "optimum" load distribution between the cpus and the gpus that using a trick like 2 cpus for a gpu task would actually make the whole system more productive.

I can show that it doesn't appear to harm the gpu production and may even improve it because the gpu will never be competing for more cpu cycles.

Apparently by running the cpus at 95% (or whatever) rather than 100%, the spare cpu power can be used to maintain "full speed" on individual cpu tasks.

That said, I have had good luck running 1 cpu / gpu task. And in smaller core count computers with slower gpu's I am not sure I am getting much gain. But it appears to help in 4 or higher core count cpus with higher end cards (Gtx 1060+ maybe).

Tom

---

A proud member of the OFA (Old Farts Association).