Another :) Corona not using all threads topic

[Namik Pirkic]

Hello all :)

I have a strange issue with my render node.

I'm using Pulze Render Manager to send the scenes from my WS to the Node for rendering so that I can have the WS free for work.

In some of the scenes that I send over to the Epyc node with the Render Manager (most of the time I'm using Pulze RM, but from time to time I tend to use Backburner as well), I can really see the swift render times (Corona using all cores at 100%), while in some cases the CPU usage drops only to one CPU or it's still using both CPUs but not all the cores, OR (this is the most frequent issue), it is using all the cores but the performance is jumping from 0% to 100% every few seconds and there is an idle drop between the peaks, so the performance graph is looking like a hair comb. Priority is set to normal although I have tried changing it to High and above but it seems that nothing helps.

Some scenes render really really fast (around 10-15 minutes for 4K render to finish at 3% noise threshold), while some (usually) simple scenes, take 30-45 minutes to render in 4K resolution on the Epyc node, while on my WS which has 8 times fewer threads, they render even faster.

When rendering on my WS (Ryzen 5950x, Win 11 Pro, 128GB Ram), the Corona is using 100% CPU power all the time.

When using DR, both WS, and Rendernode are using 100% CPU power. I tend to stay away from the DR bcs I like to have my WS free for work while the Node is supposed to do all the rendering alone.

Forgot to mention, I'm using the latest Corona 10 with HF1. Both machines have more than enough ram, both machines are using Win 11 (the Epyc is using Win 11 WS).

Before posting this I did search the forum for known issues of this kind and according to the forum posts, issues like this were supposed to be resolved a long time ago.

Tested the Epyc node with Cinebench, Old Corona benchmark, New Corona Benchmark, Vray Benchmark, all of them are using 100% of CPU power.

Went to the Corona System Settings (please see the image below), and everything seems to be in order.

Could anyone throw some insight?

Really appreciate all the feedback :)

Thanks,

Namik.

[Avi]

Hi, I just want to clarify what seems to be working and what is not working.

Working Fine:

• On Workstation (Ryzen 5950x, Win 11 Pro, 128GB Ram): Corona uses 100% CPU power all the time when rendering directly on the workstation.
• Using Corona DR: When using Corona DR between the workstation and the render node, both machines are utilizing 100% CPU power.
• Benchmark Tests: The Epyc node is fully utilized during benchmark tests like Cinebench, Corona benchmarks, and Vray Benchmark, using 100% of CPU power.

Not Working Fine:

• Using Pulze Render Manager with Epyc Node: When sending scenes to the Epyc node using Pulze RM the following issues are observed:

• Sometimes only one CPU or not all cores are utilized.
• Performance fluctuates from 0% to 100% in a repeating pattern, creating a "hair comb" appearance in the performance graph.
• Some scenes render very quickly, while others take an unusually long time, even when compared to the workstation with fewer threads.

Can you ans few questions and also try some troubleshooting steps:
1. What version of 3ds Max you are using?
2. What version of Pulze you are using? Are you using stable version or beta version?
3. When using Corona DR, everything is ok right?
4. Can you check if this is a scene specific issue or is consistent across all kinds of scene, even a simple scene like some boxes and plane?
5. When using Pulze RM, can you check network speed as well along CPU, RAM and disk usage.
6. Power settings on your Epyc node is set to High instead of power saving mode?
7. Look for any background processes that might be interfering with rendering performance.

It might take a bit long to investigate this issue so it might be best to start a support ticket with us. You can do it here: https://support.chaos.com/hc/en-us/requests/new

[Namik Pirkic]

Hello Avi :)

Working fine - Correct.

Not working fine - Using Pulze or any other Render manager (for example, Backburner - tested it thoroughly before moving to Pulze).

- I'm using 3DS Max 2024;

- I'm using the latest stable version of Pulze RM;

- When using Corona DR without Pulze, everything is okay (although I didn't have time to test all the scenes with Corona DR only);

- This issue is appearing in every third or fourth scene (sometimes it is a simple scene like a product shot, sometimes is "normal" interior scene, and sometimes it is a large exterior scene);

- When using Pulze RM, the network speed is okay, the scene is being transferred really fast, RAM usage is okay (the system is using the same amount of ram as it is was being used when I tested the scene on standalone rendering on the WS), and Disk usage - I'm not sure what you mean. I have more than enough free disk space on the node (above 250GB of free space);

- Render node power settings are all maxed out, it is not on a power-saving plan;

- I hardly doubt that there is any background process that could be doing some interference (on the Render node, I only have Win 11 WS, 3DS Max 2024, Corona, and Pulze installed).

Regards,

Namik.


PS, I have also tried pasting this "int progressive.boxScramblingPassLimit = 1000" into the dev rollout, and this is the graph (please see the attachment)

Also, upon checking 3ds max process priority, I could see that it has been sat to "below normal" automatically.

PS 2:

I have tried to tweak 3DS Max priority settings with the Process Hacker and managed to leave it automatically at "High", but still, the CPU usage looks jagged, with lots of drops, etc. Please see attachment number 2 (I have marked with red important stuff on the screen: priority, CPU usage, and the estimated rendering time). As for the estimated rendering time, my WS (Ryzen 5950x renders this scene in the same amount of time. Targeted noise threshold is 3%). So, my guess is that the dual Epyc setup should eat this scene in 2-3 minutes at least?

PS 3:

Now, what I did was to only take one asset from the scene and copy/paste it to the new fresh 3DS Max scene, setup the lights, set it to 2000px resolution and 3% noise threshold. Copied the scene directly on the Node and ran in there (double click on the scene file and just press render), and now the results are even weirder. Now the CPU usage is constantly at 100% but the render time estimation is longer (I really can not believe that I'm seeing this). Please see attachment number 3.

[Avi]

Hi Namik,

Thanks for this info. I have started a ticket 144877 for you as this can take some time to investigate.

Let's continue this conversation on the ticket.