Slow Renders.DaVinci is only using 8 cores when rendering.

I have been noticing this for a while but it has finally become a really big problem and is significantly slowing down my renders.

When I am rendering I am seeing that DaVinci is only using 8 logical cores. And while they are all maxed out I have a feeling that it is really bottlenecking my performance, especially when it comes to outputting.

Anyone have any ideas on what might be causing this?

Here are the specs on the machine I am noticing has the problem. It is running 15.1.1

Supermicro 7048GR-TR
E5-2697 v4 x2
8 x 16GB DDR4 2400 ECC Registered Memory
3x GTX 1080
On 10Gbe Network connected to Nexis E4

I am working mostly with DNxHD footage and rendering out DNxHD footage when having problems in the deliver page.

DNxHD renders are working on 32 logical core for me, so it's probably a system issue, if that's any comfort.

Strangely that does make me feel better. . . I am assuming from your post that you have hyper-threading turned on?

[Peter Chamberlain]

Can you try rendering dpx and report.

So I was able to do a little more testing this morning. With some good although frustrating results.

DPXs do seems to more evenly distribute across all of the cores but strangely some of my test renders with DNxHD also seem to evenly distribute. I can't figure out why sometimes it pegs just 8 cores to 100% and other times it will distribute across more cores. I will continue to try and test.

The bigger story here is this, I am starting the 2nd season of a show and using 15 (right now v 15.1.1) we are replicating the workflow from Season 1 almost exactly (Season 1 was done in 14.?). Outputs that used to take 10 min on 14 are now taking 45 min in 15. We are using the exact same hardware and codecs. Trying to figure out if the programs has changed and now does not as efficiently utilize resources / requires new/different more powerful resources. Or if something has gone wrong with our systems hardware or configuration.

[Peter Chamberlain]

You can load v14.x into the same system and compare, v15 should be better

Sadly I am not able to roll back that particular machine, but I did a little testing on my personal systems, I should have known better than to doubt the goodness of v15. You are totally right, 15 is on par or better than 14 in all of the tests that I have done.

I still a little disappointed in the performance that I am seeing, so the good news is that if the program performance hasn't changed it must be something with my hardware or my workflow. One thing we have done differently is that we have started using the Resolve Color Space transform sometimes over a LUT, so maybe it is that? I will continue to test and will follow up if I find anything relevant.

AFAIK the windows schedular can't handle more than 64 threads, you have a dual 2697 v4 18c/36t which makes 72 threads. Have you tried to switch of hyper threading? (intel's hyper threading doesn't bring that much anyway).

So we did some more playing around and it seems like the problem was linked to having "Dynamic Project Switching" turned on. Even though it looked like having this turned on did not effect system resources the renders just chugged. Turning this off speed up the renders significantly.

Is this just something obviously that I should have known? I know that having that option turned on loads the other projects into memory that you have open, but many time we were only using like 20% of total memory and still had really low render times.

Does having "Dynamic Project Switching" turned on allocate more then just memory to other projects?

one way I am able to see the 8 core limitation is with H.264 renders. Anyone else having this problem?

15.1.1

IMG_20181211_142020.jpg (592.28 KiB) Viewed 1305 times

H.264 export in MP4 container threads evenly across 16 physical cores for me on 15.2.1.

Cool hoping this is a 15.1.1 problem then. . .

So great news, 15.2.2 with the nVidia encoder is cranking through H.264 outputs without problem.

I am still noticing that Render Cacheing is still only using 8 cores and this seems to be the case no matter what codec I am cacheing to.

Is anyone else seeing this as well or are people seeing scaling across all cores when render caching?

So great news, 15.2.2 with the nVidia encoder is cranking through H.264 outputs without problem.

I am still noticing that Render Cacheing is still only using 8 cores and this seems to be the case no matter what codec I am cacheing to.

Is anyone else seeing this as well or are people seeing scaling across all cores when render caching?

Try this https://bitsum.com/portfolio/coreprio/ it often helps on Threadrippers with more than 16c/32tr.

I am on a Xeon. So not totally sure that is applicable.

Are you seeing the processor load all cores when rendering to the render cache?

I am on a Xeon. So not totally sure that is applicable.

Are you seeing the processor load all cores when rendering to the render cache?

Yes but I have 1950x with 16c/32tr, the windows scheduler can still handle that.
CorePrio is free, so you can just try it, when it doesn't work for you, don't use it.

Or switch to Linux..

Gotcha so you are seeing performance scale across all cores? Would you be able to post a screenshot so that I can compare to what I am seeing? Not sure if this is a problem with Windows, DaVinci or just my computers in general.

From my reading on that program I am hesitant to drop it on my machines, I was kind of just hoping that DaVinci would scale properly on it's own.

I think it's some weird system issue. I also use a 1950X and have never had these kinds of problems, but every now and then someone will post about it on the forum. I would try making an image of the system with Veeam for backup, and doing a total reinstall, just to see if it behaves the same after that.