Advanced search

Message boards : Graphics cards (GPUs) : GPU Load at 0%?

Author Message
theo
Send message
Joined: 15 Jan 19
Posts: 4
Credit: 22,607,027
RAC: 0
Level
Pro
Scientific publications
wat
Message 51347 - Posted: 17 Jan 2019 | 16:54:01 UTC

Hi,

I just started with this project using BOINC. My GPU is a NVidia GTX 1060 6GB.
On average, a task seems to take ~9 hours for my system.

MY question is why does my GPU load vary over the course of task running? It seems that over periods of 5 or more minutes, the GPU loads goes from 0% to 95% and back again while running the entire time. Is this normal? Is it because the rest of my system is bottle-necking my card?

Thanks

Aurum
Avatar
Send message
Joined: 12 Jul 17
Posts: 399
Credit: 13,024,100,382
RAC: 766,238
Level
Trp
Scientific publications
watwatwat
Message 51349 - Posted: 17 Jan 2019 | 20:56:53 UTC - in response to Message 51347.

Hi Theo & welcome, That sounds normal. Some calculations are best done by the CPU and those that can be done in parallel are best for the GPU. The work flow continuously goes back and forth between the CPU and the GPU. That's why implementing SWAN_SYNC speeds it up by dedicating a CPU logical thread for each WU running.

If you're using Windows you can watch it using the free utility GPU-Z and with Linux Mint I watch the System Monitor.

Is this your desktop PC or is it a dedicated DC rig??? What CPU do you have?

You can also run two WUs in parallel on the same GPU using an app_config.xml file placed in your Project folder.

Profile Beyond
Avatar
Send message
Joined: 23 Nov 08
Posts: 1112
Credit: 6,162,416,256
RAC: 0
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 51350 - Posted: 17 Jan 2019 | 21:14:25 UTC - in response to Message 51349.

Is this your desktop PC or is it a dedicated DC rig??? What CPU do you have?

You can also run two WUs in parallel on the same GPU using an app_config.xml file placed in your Project folder.

AuthenticAMD
AMD FX(tm)-6300 Six-Core Processor [Family 21 Model 2 Stepping 0] (6 processors)

Running 2 up on a 1060 will get you dangerously close to missing the 24hr bonus unless you set the project priority to 0. Then you'll probably waste more time in the UL/DL process than you would gain. Not the best solution.

Profile Retvari Zoltan
Avatar
Send message
Joined: 20 Jan 09
Posts: 2343
Credit: 16,201,255,749
RAC: 7,520
Level
Trp
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 51351 - Posted: 17 Jan 2019 | 23:39:40 UTC - in response to Message 51347.

Perhaps you have the default setting of "suspending GPU apps while the computer is in use" in BOINC manager.

theo
Send message
Joined: 15 Jan 19
Posts: 4
Credit: 22,607,027
RAC: 0
Level
Pro
Scientific publications
wat
Message 51352 - Posted: 18 Jan 2019 | 4:24:31 UTC - in response to Message 51349.
Last modified: 18 Jan 2019 | 4:30:15 UTC

Hi Theo & welcome, That sounds normal. Some calculations are best done by the CPU and those that can be done in parallel are best for the GPU. The work flow continuously goes back and forth between the CPU and the GPU. That's why implementing SWAN_SYNC speeds it up by dedicating a CPU logical thread for each WU running.

Thanks for the welcome. This seems like a worthy project for my time. What is SWAN_SYNC? Where can I read about it?

Is this your desktop PC or is it a dedicated DC rig??? What CPU do you have?

I am on windows 10. I have an older AMD CPU (FX 6300 6 CORE 3.5GHz) and 16GB of DDR3 along with the GTX 1060 6GB GPU.
I just thought I would help compute with my spare CPU/GPU hours.

You can also run two WUs in parallel on the same GPU using an app_config.xml file placed in your Project folder.

So if I run two WUs in parallel I wonder if I could manage the to get the WU to be 180 degrees out of synch so that each use the GPU in different times.

theo
Send message
Joined: 15 Jan 19
Posts: 4
Credit: 22,607,027
RAC: 0
Level
Pro
Scientific publications
wat
Message 51353 - Posted: 18 Jan 2019 | 4:28:09 UTC - in response to Message 51350.

AuthenticAMD
AMD FX(tm)-6300 Six-Core Processor [Family 21 Model 2 Stepping 0] (6 processors)


Wow, I didn't realize you could read my system specs on my profile. I was thinking of upgrading my mobo and CPU RAM to the next generation of hardware. Not sure if it would speed up my processing time with the GPU appreciably.

Running 2 up on a 1060 will get you dangerously close to missing the 24hr bonus unless you set the project priority to 0. Then you'll probably waste more time in the UL/DL process than you would gain. Not the best solution.


Sorry, what is the 24hr bonus?

theo
Send message
Joined: 15 Jan 19
Posts: 4
Credit: 22,607,027
RAC: 0
Level
Pro
Scientific publications
wat
Message 51354 - Posted: 18 Jan 2019 | 4:29:40 UTC - in response to Message 51351.

Perhaps you have the default setting of "suspending GPU apps while the computer is in use" in BOINC manager.


That's a good thought, but no. I made sure that the task was running the whole time and not suspended.

Wow, you sure have a lot of computation credits.

rod4x4
Send message
Joined: 4 Aug 14
Posts: 266
Credit: 2,219,935,054
RAC: 0
Level
Phe
Scientific publications
watwatwatwatwatwatwatwatwatwat
Message 51355 - Posted: 18 Jan 2019 | 7:27:26 UTC - in response to Message 51354.
Last modified: 18 Jan 2019 | 7:42:43 UTC

I noticed your jobs stop a lot. Not just CPU processing. Note the frequent temperature resets on your "StdErr Output" log (Check your Tasks for this info)

Your last GPU job stopped 46 times! The other jobs stopped almost the same amount.

Does your PC or Boinc Client have any suspend settings enabled?

Are you running anything else on the CPUs or any High Utilization processes?

Check the Event Log in Windows for any errors.

Is the CPU overheating?

Profile Beyond
Avatar
Send message
Joined: 23 Nov 08
Posts: 1112
Credit: 6,162,416,256
RAC: 0
Level
Tyr
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 51357 - Posted: 19 Jan 2019 | 0:28:30 UTC - in response to Message 51353.

AuthenticAMD
AMD FX(tm)-6300 Six-Core Processor [Family 21 Model 2 Stepping 0] (6 processors)
Wow, I didn't realize you could read my system specs on my profile. I was thinking of upgrading my mobo and CPU RAM to the next generation of hardware. Not sure if it would speed up my processing time with the GPU appreciably.

http://www.gpugrid.net/show_host_detail.php?hostid=496793

Probably no difference. Your 1060 is doing well, and welcome to the project.

Running 2 up on a 1060 will get you dangerously close to missing the 24hr bonus unless you set the project priority to 0. Then you'll probably waste more time in the UL/DL process than you would gain. Not the best solution.
Sorry, what is the 24hr bonus?

You get a 25% point bonus for finishing a WU (from when downloaded) in 48 hours and a 50% bonus for finishing in 24 hours. It adds up quickly.

goldfinch
Send message
Joined: 5 May 19
Posts: 31
Credit: 395,274,685
RAC: 742,447
Level
Asp
Scientific publications
wat
Message 51770 - Posted: 6 May 2019 | 14:42:35 UTC
Last modified: 6 May 2019 | 15:22:14 UTC

Hi All,

Sorry if already answered somewhere else - couldn't find a reply. I joined GPU-Grid only a couple of days ago as I got a laptop with GeForce GTX 1060 (NVidia drivers v.430) and integrated Intel UHD Graphics 630. Laptop is running on Win10 Home. However, all I see in BOINC logs is

7/05/2019 12:37:10 AM | GPUGRID | Requesting new tasks for Intel GPU

How can I make my NVidia card work? Or, preferrably, both?

Also, when I check Windows Task Manager, I see that acemd-922-80.exe consumes 0% GPU, with GPU Engine column blank against this task. Does it make any difference? Should it show anything?

Thanks in advance.

Vlad.

UPD: After a while I could see 95% utilisation of Cuda engine of NVidia GPU in Task Manager. Probably, it's just the amount of computations that makes it so slow...

erik
Send message
Joined: 30 Apr 19
Posts: 54
Credit: 168,971,875
RAC: 0
Level
Ile
Scientific publications
wat
Message 51833 - Posted: 16 May 2019 | 14:45:30 UTC
Last modified: 16 May 2019 | 14:46:07 UTC

swan_sync put to 1 in windows under system variables. you have to make new variable called swan_sync (doesn't matter capital or small fonts). and then reboot windows.

i wouldn't putt 2 WU on 1 card.
you can also install msi afterburner (even when your gpu is not from msi, it will work) and create fan-profile (in the settings of msi afterburner) so that your gpu isn't throttling. because my suspecsion is that your gpu is throttling.
you dont need to overclock your gpu. but if you want to overclock, do it only core clock +100mhz, very safe value. dont overclock memory speed, you will get only errored calculation when you overclock memory speed.

my gtx1060 are running at around 57-58°C with fan speeds of 90%. maybe noisy, but my cpu-cooler makes more noise than gpu-fan -:)

Post to thread

Message boards : Graphics cards (GPUs) : GPU Load at 0%?

//