CPU Usage: Predominantl using efficiency cores for rendering

   1213   4   1
User Avatar
Member
176 posts
Joined: March 2008
Offline
Hi,

So giving H19.5 a whirl and noticed that Karma (CPU) was doing something a bit odd with the efficiency vs performance cores.

When rendering to disk the efficiency cores were fully saturated, however the performance cores were only reaching 30-50% with occasional spikes to 100%. I would have thought rendering would take full advantage of the performance cores? This was just rendering a single frame.

I've attached a screenshot of what I was seeing in the activity monitor. xPU seemed slightly better, but there wasn't much in it.

H19.5.303

Cheers,

Julian

Attachments:
karma_rendering.png (422.5 KB)

Enivob
Member
2282 posts
Joined: June 2008
Offline
At a quick glance, "The efficiency cores save power and free up performance cores to concentrate on the most demanding tasks." Perhaps change the power profile of your machine? It makes sense to use the power-saving cores first. Not all of the rendering process is multi-threaded. I see similar usage on my non-M1 system. Half the cores are saturated to near 100% while the others reveal some headroom, not under full load.
Using Houdini Indie 19.5.
Ubuntu 64GB Ryzen 16 core.
nVidia 3050RTX 8BG RAM.
User Avatar
Staff
308 posts
Joined: Aug. 2019
Offline
Hi Julian, could you provide a scene file that reproduces the issue? I made a simple file with a rubber toy, a dome light, a camera, a Karma Render Settings node and a USD Render Rop. When clicking on Render to Disk or Render to MPlay buttons in the USD Render ROP I see full core utilization on an M1 Pro. I also checked to see if low power mode made any difference with the core utilization, however it still used all cores when it was enabled.
Edited by johnmather - July 23, 2022 23:47:12
User Avatar
Member
176 posts
Joined: March 2008
Offline
johnmather
Hi Julian, could you provide a scene file that reproduces the issue? I made a simple file with a rubber toy, a dome light, a camera, a Karma Render Settings node and a USD Render Rop. When clicking on Render to Disk or Render to MPlay buttons in the USD Render ROP I see full core utilization on an M1 Pro. I also checked to see if low power mode made any difference with the core utilization, however it still used all cores when it was enabled.

As always seems to be the way, I managed to reproduce this a couple of times yesterday on different occasions, however today it seems to be using all the cores.

The only things I could think of was was that perhaps the OS was trying to do something in background (a bit fuzzy on how it decides to split work), causing the render to go on the E cores or it was throttling. Tried running the render a couple of times in a row and still appears to be using all the cores.

Will keep an eye out for it but going to assume it was just some temporary glitch on MacOS' part.

Enivob
At a quick glance, "The efficiency cores save power and free up performance cores to concentrate on the most demanding tasks." Perhaps change the power profile of your machine? It makes sense to use the power-saving cores first. Not all of the rendering process is multi-threaded. I see similar usage on my non-M1 system. Half the cores are saturated to near 100% while the others reveal some headroom, not under full load.

Still bit fuzzy on how macOS decides what work goes where - it seems to be mainly up the programme rather than the user? (Apple has a full page on it which I should probably take the time to read sometime). In this case low power mode was off and the laptop was plugged in (although I don't think that makes a difference for AS laptops).
User Avatar
Member
176 posts
Joined: March 2008
Offline
Hey hey,

So ran into this issue a couple of times on build 19.5.423. This time I actually have a hip file to go with it.

Reloading the scene and restarting the renderer seemed to fix this, so not sure how reproducible this is going to prove to be - I replicated it a couple of times, but also had it work fine off the bat, within 5 minutes of each other.

Attachments:
cores.hiplc (709.0 KB)
efficiancy_cores.mov (8.8 MB)

  • Quick Links