[Mesa-dev] GPU (and system) monitoring

[Eero Tamminen]

Hi,

On 21.11.2017 02:39, Gordon Haverland wrote:
> On Mon, 20 Nov 2017 13:25:26 +0100
> Nicolai Hähnle <nhaehnle at gmail.com> wrote:
> 
>> ... umr does
>> that for the amdgpu kernel module
> 
> I downloaded the source tree, compiled umr, mounted the debugfs to use
> umr, and ran
>    umr --top
> 
> switching to sensor mode, you could see loads.  And I collected a few
> second of information to a log file.  Then I umounted the debugfs.
> 
> The load values can change from quite high to quite low from sampling
> interval to interval.  I think some kind of averaging would probably
> need to be done, just taking a single load value chosen at random would
> not be useful.
> 
> In a sense, temperature is going to be an average related to power
> (load), so maybe just having the temperatures may be enough?

1. Temperature is affected by the temperature of the surrounding media, 
power usage less so
2. Temperature sensor might not be exactly where current load is using 
most power (e.g. ALUs vs. memory)


> For now, I think I should collect a longer series of data, and then see
> how the (moving) average behaves with respect to how many data points go
> into the (moving) average.


	- Eero