Nowhere to be found so far… Looking to an equivalent to the KDE System Monitor or Windows Task Manager for the GPU instead of what is running in the CPU.
On this brand new 12.2 install I’ve tried to implement as many sensible system monitoring tools as possible.
When my FF suddenly slowed and was intermittently freezing, I glanced down at the KDE monitoring widgets installed and was amazed that CPU and RAM usage was entirely within normal parameters but my nVidia GPU usage was maxed out! Normal is to only use approx 170mb of the 356mb total VRAM.
Nothing that I would think should impact VRAM was running, only running numerous FF tabs/windows and Konsole, connected to the Internet (WiFi).
So, that made me really curious, what could be causing a spike in VRAM usage?
So far am finding zero apps for both Windows and cross-platoform, many apps promise “monitoring” but they only display sensor information, the usual stuff nVidia tools already do.
The reason you can’t find such a thing is that what runs on a GPU are
not classical processes. There is no operating system on the graphics
card which could monitor a process and report back to the operating
system running on your CPU what happens. It is in principle nothing than
a super parallel (programmable) pipeline which processes whatever it is
feed with optimization for maximal throughput, any internal monitor
would slow that down.
The only things that you can monitor are the sensors since that is what
the graphics card provides to the rest of the world.
Maybe in the future as GPU computing becomes more and more used such
things will be implemented in graphics hardware so that it can be
monitored from outside (maybe there is even already such hardware which
I don’t know about).
–
PC: oS 12.2 x86_64 | i7-2600@3.40GHz | 16GB | KDE 4.8.4 | GeForce GT 420
ThinkPad E320: oS 12.2 x86_64 | i3@2.30GHz | 8GB | KDE 4.9.1 | HD 3000
eCAFE 800: oS 12.2 i586 | AMD Geode LX 800@500MHz | 512MB | KDE 3.5.10
On 09/30/2012 01:15 AM, Martin Helm wrote:
> (maybe there is even already such hardware which
> I don’t know about).
and, my guess is what ever is now, or comes in the future will be made
by the GPU maker…you know all the in GPU’s speedy magic tricks are
very closely held secrets…which is why the proprietary drivers are so
much better (usually) than the open ones…so, i don’t expect them to
open up enough for non-proprietary sensors to work…
Am 30.09.2012 11:16, schrieb deano ferrari:
>
> It all depends on what you’re trying to monitor, but maybe these are of
> interest to you:
>
> ‘GPU usage monitoring (CUDA) - Unix and Linux’
> (http://tinyurl.com/9r62ona)
>
> 'Monitoring Nvidia GPU usage in Linux ’
> (http://www.matrix44.net/blog/?p=876)
>
>
I learned something new and silently admire the master
I have to play with the nvidia-smi on my PC, thanks for the pointer.
–
PC: oS 12.2 x86_64 | i7-2600@3.40GHz | 16GB | KDE 4.8.4 | GeForce GT 420
ThinkPad E320: oS 12.2 x86_64 | i3@2.30GHz | 8GB | KDE 4.9.1 | HD 3000
eCAFE 800: oS 12.2 i586 | AMD Geode LX 800@500MHz | 512MB | KDE 3.5.10
Thx,
Without actually trying, based on the description and screenshots IMO the CUDA code looks promising but the script that runs on the non-CUDA kernel looks no different than everything else available.
For the non-CUDA kernel I do instead recommend the KDE plasmoid widget also mentioned in the comments of the above reference.
May try the CUDA code, was hoping to avoid again using special drivers. At least IIRC today the CUDA driver doesn’t have to be manually compiled.
