Advanced search

Message boards : Graphics cards (GPUs) : how to choose which gpu (swan device 0, 1, 2, ) to use?

Author Message
erik
Send message
Joined: 30 Apr 19
Posts: 54
Credit: 168,971,875
RAC: 0
Level
Ile
Scientific publications
wat
Message 51780 - Posted: 11 May 2019 | 20:38:07 UTC

i am building a multi-gpu system (3 cards gtx 1070, 8gb) on threadripper-platform. but i want that gpugrid can use 2 of the cards for this project. 3rd card should left alone, so no gpugrid on 3rd card at all.

how to configure this in boinc? plz, when you are refering to app_info.xml to adjust it, plz post a full text of it so that i only need to copy-paste it. because i am totall noob about coding.

i am running windows 10 (64bit with 16gb ram, nvme ssd), swan_sync is set to 1 (enabled)

Zalster
Avatar
Send message
Joined: 26 Feb 14
Posts: 211
Credit: 4,496,324,562
RAC: 0
Level
Arg
Scientific publications
watwatwatwatwatwatwatwat
Message 51781 - Posted: 11 May 2019 | 21:40:39 UTC - in response to Message 51780.

I'm pretty sure you can't do that in the app_info. More than likely you will need to make a cc_config using the options command and exclude in the options section. Normally you would also need a app section but think you might be able skip that and exclude all work. Not 100 on that part. I know in the past I could exclude specific work units I didn't want on a GPU from different project this way.

You will need to experiment with this and see if it works for you. To make a xml file open notepad, paste the following in it. When you go to save (AND THIS IS IMPORTANT) make sure to select SAVE AS. Not just SAVE. Once the new window opens up, change save type to *.* not the .txt that it wants to default to. Then type in cc_config.xml It may give you an argument about using the .xml rather than .txt Tell it yes, you do want to use .xml and not .txt If it saves it as cc_config.xml.txt BOINC will not read it and it will be ignored. Move it to the BOINC folder not the GPUGrid project folder. It has to be higher up in order for BOINC to read it. You will need to restart BOINC as well as just running re-read config files will not do the trick. Good luck



<cc_config>
<options>
<exclude_gpu>
<url>http://www.gpugrid.net/</url>
<device_num>2</device_num>
</exclude_gpu>
</options>
</cc_config>



Z


The original cc_config.xml looks like the following. I only included it in case you need the app section. You would need one for each application if it doesn't like the above.

<cc_config>
<options>
<exclude_gpu>
<url>http://www.gpugrid.net/</url>
<device_num>2</device_num>
<app></app>
</exclude_gpu>
<exclude_gpu>
<url>http://www.gpugrid.net/</url>
<device_num>2</device_num>
<app></app>
</exclude_gpu>
</options>
</cc_config>

____________

erik
Send message
Joined: 30 Apr 19
Posts: 54
Credit: 168,971,875
RAC: 0
Level
Ile
Scientific publications
wat
Message 51783 - Posted: 11 May 2019 | 22:42:41 UTC - in response to Message 51781.


The original cc_config.xml looks like the following. I only included it in case you need the app section. You would need one for each application if it doesn't like the above.

<cc_config>
<options>
<exclude_gpu>
<url>http://www.gpugrid.net/</url>
<device_num>2</device_num>
<app></app>
</exclude_gpu>
<exclude_gpu>
<url>http://www.gpugrid.net/</url>
<device_num>2</device_num>
<app></app>
</exclude_gpu>
</options>
</cc_config>
my cc_config.xml file i have created a while ago to use 2 gpu's in my 2nd computer, with <use all gpu> 1. should i remove this part at all from cc_config? or should i adjust 1 => 0.66 (percentile)? having 3 cards, using 2 for boinc (2/3= 0.66)?

next question....i would like to understand why i need <app> section in your 2nd coding? my 3rd system (this topic is about it) is dedicated to run 2 projects:
world comunnity grid (only cpu usage)
and gpugrid (only gpu usage)

if i do need <app> section, can i put there acemdlong as the name of app?

Zalster
Avatar
Send message
Joined: 26 Feb 14
Posts: 211
Credit: 4,496,324,562
RAC: 0
Level
Arg
Scientific publications
watwatwatwatwatwatwatwat
Message 51784 - Posted: 12 May 2019 | 0:19:16 UTC - in response to Message 51783.
Last modified: 12 May 2019 | 0:20:06 UTC

my cc_config.xml file i have created a while ago to use 2 gpu's in my 2nd computer, with <use all gpu> 1. should i remove this part at all from cc_config? or should i adjust 1 => 0.66 (percentile)? having 3 cards, using 2 for boinc (2/3= 0.66)?


The only reason to have a <use all gpus> is if you have different GPUs in the same system. Example 1080, 1070, 1060. You would set it to 1 to mean on, 0 means off. There are no other values (0.33, 0.66.) If all the GPUs in your system are same then you can delete that. If you are using a mixed system, you will need to keep that and incorporate in into the new cc_config.


next question....i would like to understand why i need <app> section in your 2nd coding? my 3rd system (this topic is about it) is dedicated to run 2 projects:
world comunnity grid (only cpu usage)
and gpugrid (only gpu usage)

if i do need <app> section, can i put there acemdlong as the name of app?


You would need the <app> section if you decided you want to process a specific type of work unit but not another from a particular project. Here work is divided into 2 types, long and short. Other projects don't make that distinction. So for a project where all work is available, the only to prevent the project from sending a particular work unit would be to name it. You might want to do that if certain work units took an excessively long time to complete or ran your GPU too hot. To avoid that, you would basically say "I don't want these particular work units"

Try it without the <app> section first and see if it avoids sending any work to the "3rd GPU" GPU numbering is 0, 1,2 Unfortunately how BOINC numbers your GPUs is different from how M$ numbers your GPUs. If all the GPUs are the same and you don't care which of the 3 isn't being used then you can just proceed. If however there is a particular GPU you don't want used, then you will have to systematically go thru the numbering in your cc_config, ie change it from 0, 1, 2 until you figure out which card goes with that value. Beware, turning off your computer will change that value, so you will have to repeat this process each time.

If the shorten cc_config with the exclude doesn't work, then the longer one with the <app> will be needed. Yes you can put acemdlong in there but you will also need one for the short work units as well.
____________

Keith Myers
Send message
Joined: 13 Dec 17
Posts: 1284
Credit: 4,925,581,959
RAC: 6,507,650
Level
Arg
Scientific publications
watwatwatwatwat
Message 51812 - Posted: 14 May 2019 | 17:17:36 UTC
Last modified: 14 May 2019 | 17:18:24 UTC

You use how BOINC numbers the card in the Event Log startup to determine which card is 0,1,2. That numbering scheme is how you define the gpu_exclude statement in cc_config.xml. You can skip the app section. Just the project url that needs a specific gpu # excluded and the <device_num> definition.

<exclude_gpu>
<url>http://www.gpugrid.net/</url>
<device_num>0</device_num>
<type>NVIDIA</type>
</exclude_gpu>

erik
Send message
Joined: 30 Apr 19
Posts: 54
Credit: 168,971,875
RAC: 0
Level
Ile
Scientific publications
wat
Message 51832 - Posted: 16 May 2019 | 12:01:30 UTC - in response to Message 51812.

You use how BOINC numbers the card in the Event Log startup to determine which card is 0,1,2. That numbering scheme is how you define the gpu_exclude statement in cc_config.xml. You can skip the app section. Just the project url that needs a specific gpu # excluded and the <device_num> definition.
<exclude_gpu>
<url>http://www.gpugrid.net/</url>
<device_num>0</device_num>
<type>NVIDIA</type>
</exclude_gpu>

event log of boinc? or event log of windows itself?

16/05/2019 13:44:22 | | Starting BOINC client version 7.14.2 for windows_x86_64
16/05/2019 13:44:22 | | log flags: file_xfer, sched_ops, task
16/05/2019 13:44:22 | | Libraries: libcurl/7.47.1 OpenSSL/1.0.2g zlib/1.2.8
16/05/2019 13:44:22 | | Data directory: C:\ProgramData\BOINC
16/05/2019 13:44:22 | | Running under account erik
16/05/2019 13:44:22 | | CUDA: NVIDIA GPU 0: GeForce GTX 1070 (driver version 419.67, CUDA version 10.1, compute capability 6.1, 4096MB, 3556MB available, 6609 GFLOPS peak)
16/05/2019 13:44:22 | | CUDA: NVIDIA GPU 1: GeForce GTX 1070 (driver version 419.67, CUDA version 10.1, compute capability 6.1, 4096MB, 3556MB available, 6609 GFLOPS peak)
16/05/2019 13:44:22 | | OpenCL: NVIDIA GPU 0: GeForce GTX 1070 (driver version 419.67, device version OpenCL 1.2 CUDA, 8192MB, 3556MB available, 6609 GFLOPS peak)
16/05/2019 13:44:22 | | OpenCL: NVIDIA GPU 1: GeForce GTX 1070 (driver version 419.67, device version OpenCL 1.2 CUDA, 8192MB, 3556MB available, 6609 GFLOPS peak)
16/05/2019 13:44:22 | | Host name: x399-taichi
16/05/2019 13:44:22 | | Processor: 32 AuthenticAMD AMD Ryzen Threadripper 1950X 16-Core Processor [Family 23 Model 1 Stepping 1]
16/05/2019 13:44:22 | | Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 htt pni ssse3 fma cx16 sse4_1 sse4_2 movebe popcnt aes f16c rdrandsyscall nx lm avx avx2 svm sse4a osvw skinit wdt tce topx page1gb rdtscp fsgsbase bmi1 smep
16/05/2019 13:44:22 | | OS: Microsoft Windows 10: Enterprise x64 Edition, (10.00.18362.00)
16/05/2019 13:44:22 | | Memory: 15.88 GB physical, 18.76 GB virtual
16/05/2019 13:44:22 | | Disk: 237.91 GB total, 190.63 GB free
16/05/2019 13:44:22 | | Local time is UTC +2 hours
16/05/2019 13:44:22 | | No WSL found.
16/05/2019 13:44:22 | GPUGRID | Found app_config.xml
16/05/2019 13:44:22 | | Config: use all coprocessors
16/05/2019 13:44:22 | GPUGRID | URL http://www.gpugrid.net/; Computer ID 505426; resource share 100
16/05/2019 13:44:22 | GPUGRID | General prefs: from GPUGRID (last modified 30-Apr-2019 12:55:37)
16/05/2019 13:44:22 | GPUGRID | Host location: none
16/05/2019 13:44:22 | GPUGRID | General prefs: using your defaults
16/05/2019 13:44:22 | | Reading preferences override file
16/05/2019 13:44:22 | | Preferences:
16/05/2019 13:44:22 | | max memory usage when active: 8132.65 MB
16/05/2019 13:44:22 | | max memory usage when idle: 14638.77 MB
16/05/2019 13:44:22 | | max disk usage: 190.88 GB
16/05/2019 13:44:22 | | suspend work if non-BOINC CPU load exceeds 90%
16/05/2019 13:44:22 | | (to change preferences, visit a project web site or select Preferences in the Manager)
16/05/2019 13:44:22 | | Setting up project and slot directories
16/05/2019 13:44:22 | | Checking active tasks
16/05/2019 13:44:22 | | Setting up GUI RPC socket
16/05/2019 13:44:22 | | Checking presence of 34 project files
16/05/2019 13:46:30 | GPUGRID | project resumed by user
16/05/2019 13:46:32 | GPUGRID | Sending scheduler request: Requested by project.
16/05/2019 13:46:32 | GPUGRID | Not requesting tasks: "no new tasks" requested via Manager
16/05/2019 13:46:34 | GPUGRID | Scheduler request completed


I got 2 of the same card, exactly the same, msi gtx1070 8gb. the first is connected to x16-slot, the second is connected to x8-slot.
Boinc is right now running on the 2nd card via x8-slot, I can see this with gpu-z because this card is loaded for 90%.

I blocked Boinc for new task request until I can figure out which gpu (1st on x16-slot or 2nd on x8-slot) to rule out.

so, help me plz with <exclude> en <device_num> codes

Zalster
Avatar
Send message
Joined: 26 Feb 14
Posts: 211
Credit: 4,496,324,562
RAC: 0
Level
Arg
Scientific publications
watwatwatwatwatwatwatwat
Message 51837 - Posted: 16 May 2019 | 17:16:13 UTC - in response to Message 51832.
Last modified: 16 May 2019 | 17:16:27 UTC

.
<exclude_gpu>
<url>http://www.gpugrid.net/</url>
<device_num>0</device_num>
<type>NVIDIA</type>
</exclude_gpu>


I'm assuming this is the current cc_config in your system??

Boinc is right now running on the 2nd card via x8-slot, I can see this with gpu-z because this card is loaded for 90%.


I'll also assume you want GPUGrid to run on the 1st card since you mentioned this. If so, then it's an easy fix. Change the 0 to 1 and restart BOINC. It should now be crunching on the x16 PCIe slot. Something I forgot to add. You will need to add a section to the cc_config to block the new project from sending work to the other card being use exclusively for GPUGrid. Otherwise the other project will send work for it as well.

<exclude_gpu>
<url>http://www.gpugrid.net/</url>
<device_num>1</device_num>
<type>NVIDIA</type>
</exclude_gpu>
<exclude_gpu>
<url>url of other project</url>
<device_num>0</device_num>
<type>NVIDIA</type>
</exclude_gpu>


Good luck

Z
____________

erik
Send message
Joined: 30 Apr 19
Posts: 54
Credit: 168,971,875
RAC: 0
Level
Ile
Scientific publications
wat
Message 51842 - Posted: 16 May 2019 | 22:28:16 UTC - in response to Message 51837.

thanks for your reply.
i was wondering if there is some logical way of thinking or numbering about pci-devices regarding boinc. boinc is running on 2nd gpu in x8-slot (pcie-bus 65 according windows 10 device manager). my 1st gpu is in first pcie-slot x16 (pcie-bus 9). one should expect that firstly pcie-bus 9 should be loaded in boinc. this is why i was wondering about the logical thinking of boinc

Zalster
Avatar
Send message
Joined: 26 Feb 14
Posts: 211
Credit: 4,496,324,562
RAC: 0
Level
Arg
Scientific publications
watwatwatwatwatwatwatwat
Message 51843 - Posted: 16 May 2019 | 22:33:37 UTC - in response to Message 51842.

That I can not answer. Might ask RIchard or Keith as they seem familiar as to how BOINC is compiled to run.

Keith Myers
Send message
Joined: 13 Dec 17
Posts: 1284
Credit: 4,925,581,959
RAC: 6,507,650
Level
Arg
Scientific publications
watwatwatwatwat
Message 51848 - Posted: 17 May 2019 | 6:22:20 UTC - in response to Message 51832.

event log of boinc? or event log of windows itself?

The Event Log of BOINC is how the cards are numbered for the gpu_exclude device number.

erik
Send message
Joined: 30 Apr 19
Posts: 54
Credit: 168,971,875
RAC: 0
Level
Ile
Scientific publications
wat
Message 51849 - Posted: 17 May 2019 | 8:38:38 UTC - in response to Message 51848.

event log of boinc? or event log of windows itself?

The Event Log of BOINC is how the cards are numbered for the gpu_exclude device number.

did you see my event log of boinc?
i have 2 of the same gpu, excatly the same names. same factory, same memory.

now i can mamage which card to exclude because i now have 2 gpu, this is now easy.

how to manage this when you have 4 of the same cards?

Zalster
Avatar
Send message
Joined: 26 Feb 14
Posts: 211
Credit: 4,496,324,562
RAC: 0
Level
Arg
Scientific publications
watwatwatwatwatwatwatwat
Message 51852 - Posted: 17 May 2019 | 14:18:00 UTC - in response to Message 51849.
Last modified: 17 May 2019 | 14:18:14 UTC

how to manage this when you have 4 of the same cards?


Now that I can answer. I decide beforehand how many cards I want to run on this project and just assign them a number. Since they are all the same and the PCIe slots are all the same, it's not a hard choice. But I'm betting most of your motherboard's slots are not PCIex16 or x8. In which case, you will have to experiment with the numbering and see what number corresponds to which GPU.
____________

mmonnin
Send message
Joined: 2 Jul 16
Posts: 332
Credit: 3,772,896,065
RAC: 4,765,302
Level
Arg
Scientific publications
watwatwatwatwat
Message 51857 - Posted: 17 May 2019 | 21:57:47 UTC - in response to Message 51852.

how to manage this when you have 4 of the same cards?


Now that I can answer. I decide beforehand how many cards I want to run on this project and just assign them a number. Since they are all the same and the PCIe slots are all the same, it's not a hard choice. But I'm betting most of your motherboard's slots are not PCIex16 or x8. In which case, you will have to experiment with the numbering and see what number corresponds to which GPU.


GPUGrid tasks report the PCI ID

From your PC with 4x Tis
# PCI ID : 0000:09:00.0
# PCI ID : 0000:06:00.0
# PCI ID : 0000:05:00.0
# PCI ID : 0000:0A:00.0

BOINC gets it from somewhere.

Keith Myers
Send message
Joined: 13 Dec 17
Posts: 1284
Credit: 4,925,581,959
RAC: 6,507,650
Level
Arg
Scientific publications
watwatwatwatwat
Message 51858 - Posted: 17 May 2019 | 21:59:20 UTC - in response to Message 51849.

event log of boinc? or event log of windows itself?

The Event Log of BOINC is how the cards are numbered for the gpu_exclude device number.

did you see my event log of boinc?
i have 2 of the same gpu, excatly the same names. same factory, same memory.

now i can mamage which card to exclude because i now have 2 gpu, this is now easy.

how to manage this when you have 4 of the same cards?

That becomes more difficult. You then have to ID the card by the PCIe BusID number or the UUID which is an even better way as that number is unique to every card like a fingerprint.

You can use the nvidia-smi with the -L parameter to show the UUID of a card.
nvidia-smi is located at C:\Program Files\NVIDIA Corporation\NVSMI in Windows.
nvidia-smi -L
GPU 0: GeForce GTX 1070 (UUID: GPU-e25636cc-94d1-42ca-9b31-6e0018dfedd6)
GPU 1: GeForce RTX 2080 (UUID: GPU-089608fe-cba5-4711-bf68-085fd0711d8c)
GPU 2: GeForce GTX 1080 (UUID: GPU-2d04c292-2286-8b37-d8b8-c01f67e752fd)


To figure out which 1070 is which you could always resort to the "finger test" IOW stop a card's fan so that your monitoring software shows some card's fan speed drop to zero. Then you know which 1070 is which.

erik
Send message
Joined: 30 Apr 19
Posts: 54
Credit: 168,971,875
RAC: 0
Level
Ile
Scientific publications
wat
Message 51860 - Posted: 17 May 2019 | 22:50:32 UTC - in response to Message 51858.
Last modified: 17 May 2019 | 22:59:53 UTC

event log of boinc? or event log of windows itself?

The Event Log of BOINC is how the cards are numbered for the gpu_exclude device number.

did you see my event log of boinc?
i have 2 of the same gpu, excatly the same names. same factory, same memory.

now i can mamage which card to exclude because i now have 2 gpu, this is now easy.

how to manage this when you have 4 of the same cards?

That becomes more difficult. You then have to ID the card by the PCIe BusID number or the UUID which is an even better way as that number is unique to every card like a fingerprint.

You can use the nvidia-smi with the -L parameter to show the UUID of a card.
nvidia-smi is located at C:\Program Files\NVIDIA Corporation\NVSMI in Windows.
nvidia-smi -L
GPU 0: GeForce GTX 1070 (UUID: GPU-e25636cc-94d1-42ca-9b31-6e0018dfedd6)
GPU 1: GeForce RTX 2080 (UUID: GPU-089608fe-cba5-4711-bf68-085fd0711d8c)
GPU 2: GeForce GTX 1080 (UUID: GPU-2d04c292-2286-8b37-d8b8-c01f67e752fd)


To figure out which 1070 is which you could always resort to the "finger test" IOW stop a card's fan so that your monitoring software shows some card's fan speed drop to zero. Then you know which 1070 is which.

UUID number is that the same what i can see in windows Register under CurrentContorlSet/Video? because msi afterburner shows me those regiater-names or -numbers.

i was wrong when i said that my first x16-slot has pcie-bus device 9. my first pci-slot (is x16) has pcie-bus device 65, where my first card is inserted and where hdmi screen is connected to.
my 2nd card which was as firstly loaded in boinc for calculations has pcie-device bus 9.
could it be that boinc firstly takes the gpu-card to use / to load with the lowest number of pcie-device-ID (meaning pci device 9, x8-slot)?

that finger testing of the fan is very good option to see which card is what

those nvidia commands i will try monday as my computers are at my work

EDIT...i have found also some interessting thing about tasks. when you locate in which slot in boinc is task running (by clicking on Properties of a task), for example slot/0, in the map of slot-0 i can find a file called init_data.xml.
inside this xml-file there is a option called device_num. here i can see that a specific task is running on <device_num>0 or <device_num>1.
this also gives me some clue which card is loaded with calculations

Keith Myers
Send message
Joined: 13 Dec 17
Posts: 1284
Credit: 4,925,581,959
RAC: 6,507,650
Level
Arg
Scientific publications
watwatwatwatwat
Message 51862 - Posted: 18 May 2019 | 3:58:29 UTC - in response to Message 51860.

From what I can gather, BOINC numbers the cards based on highest performance. So a RTX 2080 is numbered as Device 0 and then a GTX 1080 is numbered as Device 1 and then a GTX 1070 is numbered as Device 2.

When you have identical cards, then I think BOINC numbers the cards according to lowest PCIe BusId. So PCIe BusID 5 is Device 0, BusID 6 is Device 1 and BusID 9 is Device 2. That is how my 3 identical GTX 1070 Ti get numbered.

Jozef J
Send message
Joined: 7 Jun 12
Posts: 112
Credit: 1,118,845,172
RAC: 0
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwat
Message 51865 - Posted: 18 May 2019 | 9:20:00 UTC

Simply: Bionc manager is overmatured undeveloped software where we can not:
ADD project per cpu core ---
ADD project per gpu ---in simple GUI

In todays 21.C is using appinfo and cc_config like using https://en.wikipedia.org/wiki/Punched_tape )) just in electronics form we dont use paper but we must use that app and cc.. witch is not user friendly and dificult for normal users and they think this manipulation in OS folders and adding this app/cc is "hacking" ..)) ..noobs are hapy but this will not get more people to boinc.
Why anderson dont use other people to develop boinc manager?? like in mobile OS where all mobile OS make big steps in last years, we use 15 years old boinc manager.
I think it will not better in future as i can see here:
[url]https://boinc.berkeley.edu/trac/wiki/ProjectMain [/url]

erik
Send message
Joined: 30 Apr 19
Posts: 54
Credit: 168,971,875
RAC: 0
Level
Ile
Scientific publications
wat
Message 51867 - Posted: 18 May 2019 | 11:11:13 UTC - in response to Message 51865.

Simply: Bionc manager is overmatured undeveloped software where we can not:
ADD project per cpu core ---
ADD project per gpu ---in simple GUI

i totally agree with that this would be awesome feature of boinc

mmonnin
Send message
Joined: 2 Jul 16
Posts: 332
Credit: 3,772,896,065
RAC: 4,765,302
Level
Arg
Scientific publications
watwatwatwatwat
Message 51868 - Posted: 18 May 2019 | 11:44:11 UTC

Thats probably easier to do for a Program that runs 1 single project, like FAH, but more complicated for the general user with multiple projects.

Jozef J
Send message
Joined: 7 Jun 12
Posts: 112
Credit: 1,118,845,172
RAC: 0
Level
Met
Scientific publications
watwatwatwatwatwatwatwatwatwatwat
Message 51869 - Posted: 18 May 2019 | 13:35:28 UTC

OMG people ask for this "feature" from first days of seti project....you can find in waybackmachine
Simply "you not abble for 20 years "delivering" something.. becouse there is problem with , berkeley infrastructure read your computer like "host" , whole host, and this segmenting" on cpu core,gpu core is "unreadable" for them.. but in modern era it can be do it over java for exemple,not very best but will work
Also some boinc project managers dont like becouse they will more benefit from your whole host PC , than like only few cores...)
sorry for off topic , but on boinc page is this problem ignored yeaars

erik
Send message
Joined: 30 Apr 19
Posts: 54
Credit: 168,971,875
RAC: 0
Level
Ile
Scientific publications
wat
Message 51964 - Posted: 2 Jun 2019 | 20:07:45 UTC

a little update how device numbering in boinc is.
boinc sees multi-gpu setup the same way as gpu-z sees.

under ADVANCED tab you have option called CUDA, under CUDA you can see MULTI BOARD and then No (0) or No (1) etc etc. those device numbers in gpu-z are the same device numbers in boinc. and THIS numbering is done according to PCIE BUS ID. lower PCIE ID gets lower device number. you should say: of course, nothing strange or news with this.

now comes strange part....my motherboard asrock x399 taichi (for threadripper cpu's) has the following pcie-slots, counting from the pcie-slot most nearby cpu and memory banks:

- x16 electrial (pcie bus id 66)
- x16 slot (eletrical x8, pcie bus id 66)
- x1 slot electrical (pcie bus id 7)
- x16 electrical (pcie bus id 10)
- x16 slot (electrical x8, pcie bus id 9)

according to this pcie bus id's my device 0 in boinc and in gpu-z is x1-slot.

all this information is just for sharing.
i am not asking why 1st pcie-slot (most nearby cpu and memory banks) has the highest pcie bus id. but just sharing my experience.

oh yeah, forgotten to say.....swan_sync enabled under windows 10, tried with different gtx-cards (1060 and 1070), alsways multi-gpu in alls the same gtx-cards or all mixed gtx-cards.

so, i am saying: the card type has nothing to do with pcie bus id numbering

have fun !!!

erik
Send message
Joined: 30 Apr 19
Posts: 54
Credit: 168,971,875
RAC: 0
Level
Ile
Scientific publications
wat
Message 52205 - Posted: 7 Jul 2019 | 11:00:58 UTC

another question....rigth now i am testing gtx 1080 Ti card. its load is about 50%. this card is connected to pcie-x8 slot. is this normal for 1080 Ti? 50% load of gpu?

if it is not, is it possible to sign specific app (acemdlong or acemdshort) to this specific card? and then using gpu_usage 0.5 comment to load 2 WU on this card so it finally get almost 100% gpu load?

Erich56
Send message
Joined: 1 Jan 15
Posts: 1090
Credit: 6,603,906,926
RAC: 18,783,925
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwat
Message 52206 - Posted: 7 Jul 2019 | 14:37:49 UTC - in response to Message 52205.

another question....rigth now i am testing gtx 1080 Ti card. its load is about 50%. this card is connected to pcie-x8 slot. is this normal for 1080 Ti? 50% load of gpu?

Did you enable SwanSync ?

erik
Send message
Joined: 30 Apr 19
Posts: 54
Credit: 168,971,875
RAC: 0
Level
Ile
Scientific publications
wat
Message 52207 - Posted: 8 Jul 2019 | 9:00:38 UTC - in response to Message 52206.

another question....rigth now i am testing gtx 1080 Ti card. its load is about 50%. this card is connected to pcie-x8 slot. is this normal for 1080 Ti? 50% load of gpu?

Did you enable SwanSync ?

yes, swan_syn is enabled. i am running windows 10, 64bit with latest windows updates, nvidia driver version 430. gpu load for gtx 1080 Ti is round 55-59%, connected to pcie-x8 slot (and also electrically x8-slot)

so once again my question....is it possible to asign a specific gpu-card to a specific application? meaning...i would like to asign gtx1080 Ti to only acemdlong-app. is this possible?

Profile Retvari Zoltan
Avatar
Send message
Joined: 20 Jan 09
Posts: 2343
Credit: 16,201,255,749
RAC: 6,169
Level
Trp
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 52209 - Posted: 8 Jul 2019 | 19:20:12 UTC - in response to Message 52207.

meaning...i would like to asign gtx1080 Ti to only acemdlong-app. is this possible?

No, but you can exclude a GPU from a project.
See the <exclude_gpu> section here.

erik
Send message
Joined: 30 Apr 19
Posts: 54
Credit: 168,971,875
RAC: 0
Level
Ile
Scientific publications
wat
Message 52213 - Posted: 9 Jul 2019 | 10:42:27 UTC - in response to Message 52209.

meaning...i would like to asign gtx1080 Ti to only acemdlong-app. is this possible?

No, but you can exclude a GPU from a project.
See the <exclude_gpu> section here.

thanks for your reply. pitty i cann't asign specific device to specific application with specific setting for that device or appliction....would be great feature of BOINC

mmonnin
Send message
Joined: 2 Jul 16
Posts: 332
Credit: 3,772,896,065
RAC: 4,765,302
Level
Arg
Scientific publications
watwatwatwatwat
Message 52219 - Posted: 9 Jul 2019 | 23:23:44 UTC - in response to Message 52213.

meaning...i would like to asign gtx1080 Ti to only acemdlong-app. is this possible?

No, but you can exclude a GPU from a project.
See the <exclude_gpu> section here.

thanks for your reply. pitty i cann't asign specific device to specific application with specific setting for that device or appliction....would be great feature of BOINC


That is low GPU util even w/o swan sync.

The best way to use 1 GPU on one project and a 2nd GPU on another project is with separate BOINC Clients with the exclude function mentioned above. I have a guide for Windows and Linux at Overclock.net in the BOINC Guide sub forum if you're interested. My PCs with GPUs all have 1 client for CPU work and another for GPUs. It allows for finer control of work buffers, esp during competitions.

erik
Send message
Joined: 30 Apr 19
Posts: 54
Credit: 168,971,875
RAC: 0
Level
Ile
Scientific publications
wat
Message 52222 - Posted: 10 Jul 2019 | 5:29:35 UTC - in response to Message 52219.

meaning...i would like to asign gtx1080 Ti to only acemdlong-app. is this possible?

No, but you can exclude a GPU from a project.
See the <exclude_gpu> section here.

thanks for your reply. pitty i cann't asign specific device to specific application with specific setting for that device or appliction....would be great feature of BOINC


That is low GPU util even w/o swan sync.

The best way to use 1 GPU on one project and a 2nd GPU on another project is with separate BOINC Clients with the exclude function mentioned above. I have a guide for Windows and Linux at Overclock.net in the BOINC Guide sub forum if you're interested. My PCs with GPUs all have 1 client for CPU work and another for GPUs. It allows for finer control of work buffers, esp during competitions.
so, i have read your guide for 2nd client setup.

what you actually suggest is in 1 client to exclude gtx1070 and include gtx1080 with setting like i wish.

in another client (2nd client) to do the opposite...do i understand you correct?

because you make 2 folders for boinc:
C:\ProgramData\BOINC

C:\ProgramData\BOINC2

mmonnin
Send message
Joined: 2 Jul 16
Posts: 332
Credit: 3,772,896,065
RAC: 4,765,302
Level
Arg
Scientific publications
watwatwatwatwat
Message 52223 - Posted: 10 Jul 2019 | 9:55:20 UTC - in response to Message 52222.

meaning...i would like to asign gtx1080 Ti to only acemdlong-app. is this possible?

No, but you can exclude a GPU from a project.
See the <exclude_gpu> section here.

thanks for your reply. pitty i cann't asign specific device to specific application with specific setting for that device or appliction....would be great feature of BOINC


That is low GPU util even w/o swan sync.

The best way to use 1 GPU on one project and a 2nd GPU on another project is with separate BOINC Clients with the exclude function mentioned above. I have a guide for Windows and Linux at Overclock.net in the BOINC Guide sub forum if you're interested. My PCs with GPUs all have 1 client for CPU work and another for GPUs. It allows for finer control of work buffers, esp during competitions.
so, i have read your guide for 2nd client setup.

what you actually suggest is in 1 client to exclude gtx1070 and include gtx1080 with setting like i wish.

in another client (2nd client) to do the opposite...do i understand you correct?

because you make 2 folders for boinc:
C:\ProgramData\BOINC

C:\ProgramData\BOINC2


Yes, thats correct. One install, several client instances are opened with separate data directories (with their own configs).

Post to thread

Message boards : Graphics cards (GPUs) : how to choose which gpu (swan device 0, 1, 2, ) to use?

//