Hi, long story short: I would like to be able to tell for sure that VRAM limit has been reached for a given XPU render.

I'm running benchmarks and I'm using
$HFS/bin/iinfo -J image.exr
to get render stats of my renders.

My current opinion is that those stats are inaccurate when it comes to VRAM consumption when its limit is exceeded.

I have purposely created VRAM killer scenes and I ran iinfo on those renders.

I have found that both GPU and CPU are listed there and when VRAM is exceeded, they report VRAM consumption as "10GB" or something small, but if you look at the device contribution to the render it's CPU 100%, GPU 0%...

...or so I thought. Turns out the GPU contribution is not always 0%. It can be 5%, 15%, 27%. And I'm not sure at this point if it's truly a viable way to judge VRAM limit.

Is there maybe a different key in the stats? or should I increase verbosity and get some sort of warning from husk's stdout somehow? My preferred way would be to keep using iinfo.