Letting cuda WUs use a full thread/core

Shafa
Shafa
Joined: 31 May 05
Posts: 53
Credit: 627,005,014
RAC: 0

RE: Also, are there any

Quote:

Also, are there any nice programs for monitoring my GPU for linux as well as my CPU temps?

Well, depends what "nice" means to you.
I use GNOME Sensors Applet 2.2.3 and on a top bar I can see all temperatures (cpu, mb, hdd) as well as gpu temp.
Check it out - pure text or icons or mixed indication are possible. I prefer just pure text - then I can see only actual numbers and unit (°C, rpm).
No colored fiddleshit...

http://sensors-applet.sourceforge.net/index.php?content=screenshots
http://fedora4520.blogspot.com/2007/10/gnome-sensors-applet-in-acer-aspire.html
http://ossipedia.ipa.go.jp/appcatalog/image/lsc-492.png
http://lh5.ggpht.com/_AKAirBqiu8U/TEkB41lwefI/AAAAAAAAAPI/tg6Afol4Zsc/sensors-applet.jpg

FrankHagen
FrankHagen
Joined: 13 Feb 08
Posts: 102
Credit: 250,075
RAC: 0

RE: No colored

Quote:
No colored fiddleshit...

LOL - but that's exactly the thing most windozers call nice.. ;)

on the other hand - tools like MSI-afterburner where you get clock + voltage control and monitoring AIO are nice to have. of course this could be achieved without fragglecrap, but it's as it is..

Dirk
Dirk
Joined: 4 Jun 08
Posts: 35
Credit: 88,264,743
RAC: 0

RE: RE: No colored

Quote:
Quote:
No colored fiddleshit...

LOL - but that's exactly the thing most windozers call nice.. ;)

on the other hand - tools like MSI-afterburner where you get clock + voltage control and monitoring AIO are nice to have. of course this could be achieved without fragglecrap, but it's as it is..

Haha that's exactly what I call nice, I like to monitor what my GPU is doing.

GPU usage, temp, fan speed, core clock, shader clock, mem clock, vram usage.

Also allows for easy overclocking and fan speed control. I do need to confess that the green/black colour scheme of it pleases me :D

Dagorath
Dagorath
Joined: 22 Apr 06
Posts: 146
Credit: 226,423
RAC: 0

I also recommend the gnome

I also recommend the gnome sensors app Shafa refers to. For NVIDIA GPU the NVIDIA proprietary driver comes with an excellent command line tool named nvidia-settings for querying and setting all the driver attributes (settings). For example it shows you GPU temp and fan speed and allows you to set fan speed. You also get a nice GUI front end for the nvidia-settings tool but it crashes on start with my GTX 570 and driver 270.41.19. It works fine with driver 260.??. For that reason I strongly recommend starting with the 260.?? driver as it's sufficient for Einstein and GPUgrid CUDA apps and I suspect other projects' CUDA apps too.

TEMPERATURE WARNING

When I started crunching my very first CUDA task the GPU temp shot up to 95C but the fan speed stayed put at a constant 65% instead of rising to 100%. I fiddled around with it and came to the conclusion that I had it setup for fan speed to automatically rise and fall with temp and that there must be a bug in that mechanism. I ended up adjusting settings to take manual control of fan speed and now see GPU temp hovering between 59C and 65C. How I did that is detailed in this post at the BOINC dev forums. I'm running Fedora 14 and GTX 570. YMMV with Ubuntu and your GPUs. The bottom line is get familiar with your GPU temp and fan speed monitoring tools BEFORE you start that first CUDA task.

edit added:

Also see the nvclock and nvidia-smi CLI apps. Do "nvclock --help" and "nvidia-smi --help" for help summary. Do "man nvclock" and "man nvidia-smi" to see the fine manuals.

Dirk
Dirk
Joined: 4 Jun 08
Posts: 35
Credit: 88,264,743
RAC: 0

RE: I also recommend the

Quote:

I also recommend the gnome sensors app Shafa refers to. For NVIDIA GPU the NVIDIA proprietary driver comes with an excellent command line tool named nvidia-settings for querying and setting all the driver attributes (settings). For example it shows you GPU temp and fan speed and allows you to set fan speed. You also get a nice GUI front end for the nvidia-settings tool but it crashes on start with my GTX 570 and driver 270.41.19. It works fine with driver 260.??. For that reason I strongly recommend starting with the 260.?? driver as it's sufficient for Einstein and GPUgrid CUDA apps and I suspect other projects' CUDA apps too.

TEMPERATURE WARNING

When I started crunching my very first CUDA task the GPU temp shot up to 95C but the fan speed stayed put at a constant 65% instead of rising to 100%. I fiddled around with it and came to the conclusion that I had it setup for fan speed to automatically rise and fall with temp and that there must be a bug in that mechanism. I ended up adjusting settings to take manual control of fan speed and now see GPU temp hovering between 59C and 65C. How I did that is detailed in this post at the BOINC dev forums. I'm running Fedora 14 and GTX 570. YMMV with Ubuntu and your GPUs. The bottom line is get familiar with your GPU temp and fan speed monitoring tools BEFORE you start that first CUDA task.

edit added:

Also see the nvclock and nvidia-smi CLI apps. Do "nvclock --help" and "nvidia-smi --help" for help summary. Do "man nvclock" and "man nvidia-smi" to see the fine manuals.

The fermi cards do run hot yes, I believe the max temp for my GTX480 is 105c, of course it's a bad idea to run it constantly at high temps so as you said adjusting fan speed is important. The reason it runs so hot on stock settings seems obvious to me though, nvidia wanted to keep the fan noise down and the fan is very loud at high speed. Only when it passes 90c the fan on my card (at stock settings) will ramp up the speed. What I love about MSI afterburner is that I can easily adjust that fan speed/temp curve to what I want. I let the fan run a bit slower during the night because 80% fan speed is quite noisy when I try to sleep.

Just got a mail that my cpu cooler has been handed off to the delivery service so I'll hopefully get it tomorrow and install it asap. Sometime after that I'll start my linux adventure, already thanks for all the tips you guys gave me :)

Oh yea, I'll have to edit my app_info to run 4 tasks on my GPU under linux too, but I'll do that after I get it running smoothly first.

FrankHagen
FrankHagen
Joined: 13 Feb 08
Posts: 102
Credit: 250,075
RAC: 0

RE: Oh yea, I'll have to

Quote:
Oh yea, I'll have to edit my app_info to run 4 tasks on my GPU under linux too, but I'll do that after I get it running smoothly first.

it's really hard to tell, but i'd start with 2 and watch it. with that full-cpu app you'll get, 4 is most likely overkill.

Stranger7777
Stranger7777
Joined: 17 Mar 05
Posts: 436
Credit: 398,130,452
RAC: 124,739

For me, I'm using a Process

For me, I'm using a Process Tamer to adjust cpu usage of BRP3. Each time BRP3 starts, Process Tamer changes it priority from idle level (I guess it is the same as nice 19 for Linux) to Normal. But can go further and set it to HIGH or even Runtime. Just tell us what happens then... :)

Dirk
Dirk
Joined: 4 Jun 08
Posts: 35
Credit: 88,264,743
RAC: 0

Alright, was a bit hectic

Alright, was a bit hectic (I'm about to become an uncle) but I've got the CPU cooler installed. Now my CPU is finally being cooled properly (full load at 3.4Ghz and it maxed out at 56c). Good timing too because next week promises to be very hot and humid.

I went and installed ubuntu, installed the recommended nvidia drivers but found out they were a 270.xx release. I've had a look at how to install 260.xx manually but I'm way too tired right now, is there a way to do it manually with that driver update utility thingy?

I've yet to install boinc but I'll do that after I get those nvidia drivers up and running. Anyways, I hope I can get some good sleep tonight because I'm dead tired.

Now that I think about it, perhaps I should just use the 270.xx driver and install boinc to see which app I get. Oh well, tomorrow...

Dagorath
Dagorath
Joined: 22 Apr 06
Posts: 146
Credit: 226,423
RAC: 0

I'm not sure what you mean by


I'm not sure what you mean by "that driver utility update thingy".

I would go with the 270.xx driver. You might have heard me say the 270.xx driver broke the nvidia-settings GUI but that started working for me again (though I hate not knowing why) so I have nothing bad to say about 270.xx. There are reports from a few claiming their GPU downclocks with 270.xx but you can monitor that later to see if you're affected that way.

If it happens that the nvidia-settings GUI doesn't work for you then have a look at this post for commands to query the fan speed and GPU temp. I use the last command in that post and shrink the terminal size to about 1" X 1" and leave it run all the time. You can add to that command to make it monitor the clock freqs too. If you wanna go crazy with it you can redirect the output to a file and graph the data with GNUplot.

Welcome to Linux and a world of free software :)

Shafa
Shafa
Joined: 31 May 05
Posts: 53
Credit: 627,005,014
RAC: 0

I did a few experiments with

I did a few experiments with CPU priority of BRP3 (changing the nice value for BRP3), bud did not see any noticable improvement at all...

Ubuntu 10.04, GTS250, Athlon2 4core, gpu driver 270.29.
4 CPU units and 2 BRP3 units at the time...

There is no reason to install 260.xx if a system is working fine with 270.xx - moreover if I remember well, I had some troubles with a combination of 260.xx + E@H + cuda...

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.