Question TR 2990WX, 3D Rendering, NUMA, RAM Stick Location Confusion

[meteion]

Any thoughts are appreciated I know this is a long post and I'm also not super knowledgeable on these things, but please bear with me!

Asus PRIME X399-A EATX sTR4
Corsair Vengeance 32 GB (2x16gb) DDR4-3200 CL16
Corsair Vengeance 64 (2x32gb) DDR4-2666 CL16
Radeon RX 580

Question: Is my RAM affecting 2990WX performance?

I recently replaced my TR 1950X that I use for 3D rendering in a software called e-on Vue with a 2990WX. It primarily uses CPU rendering so having the 16 cores of the 1950X worked out for me quite well in the past.

The 1950X performed well enough, but seeing that it was 7 year old tech at this point I figured I'd try to max out my workstation to the CPU with the most cores that would fit my MOBO (before having to build a whole new machine altogether because of the socket). I went with the 2990WX thinking double the amount of cores would mean (almost, since the slower clock speed) double the speed, but I've been pretty disappointed so far...

Vue comes with an internal and external renderer. The internal renderer allows you to see the render happen on screen, whereas the external renderer simply renders the image and saves it to a location. For some reason with the internal renderer, the 2990WX is only using 1 of the 4 NUMA nodes (16/64 logical cores), effectively utilizing only 25% of its power. This is resulting in slower render speeds than the 1950X (which was also throttled here for reasons unknown to me but was still faster)!

With the external renderer, however, it utilizes all 64 logical cores, but then takes longer than the 1950X to set up each render . So, the render time itself is halved, but the time in between renders is longer than what I experienced before, resulting in external render speeds which are basically the same. My best guess is that this has something to do with the 2990WX only having four memory controllers resulting in latency... Should I have gone with the 2950X this whole time?

Then grasping at straws I started to wonder if my RAM had anything to do with this.

So, here I realized I didn't properly install the 4 sticks in the 8 DIMM sockets. I don't remember in what order, but initially I had them all on one side of the CPU (instead of 2 on one side and 2 on the other in the right sockets). This made the CPU prioritize NUMA Node 2 (instead of 0) when using the internal renderer. (I also did Cinebench tests with this random RAM setup, getting full CPU utilization and ok-ish scores for the 2990WX, hovering around 900, with the official 2990WX benchmark being at 1200).

After swapping the sticks around so that they were in the right DIMM sockets, NUMA node 0 is what's prioritized now, but now I'm getting significantly lower Cinebench scores (490, basically half). What happened?

(EDIT: this was a result of BIOS EZ tune settings- when I restored it to default, Cinebench test is normal, if not better than before)

Right now, I'm trying to see if the 2990WX is still viable, but I might be reaching the limits of my expertise.

I'm currently waiting to hear back from e-on support about their internal render, because I feel like I remembered there was a time when even my 1950X used all 32 logical cores with it. Maybe they updated it and something changed, but the 2990WX being limited to a single NUMA is pretty upsetting.

 

[kanewolf]

Any thoughts are appreciated I know this is a long post and I'm also not super knowledgeable on these things, but please bear with me!

Asus PRIME X399-A EATX sTR4
Corsair Vengeance 32 GB (2x16gb) DDR4-3200 CL16
Corsair Vengeance 64 (2x32gb) DDR4-2666 CL16
Radeon RX 580

Question: Is my RAM affecting 2990WX performance?

I recently replaced my TR 1950X that I use for 3D rendering in a software called e-on Vue with a 2990WX. It primarily uses CPU rendering so having the 16 cores of the 1950X worked out for me quite well in the past.

The 1950X performed well enough, but seeing that it was 7 year old tech at this point I figured I'd try to max out my workstation to the CPU with the most cores that would fit my MOBO (before having to build a whole new machine altogether because of the socket). I went with the 2990WX thinking double the amount of cores would mean (almost, since the slower clock speed) double the speed, but I've been pretty disappointed so far...

Vue comes with an internal and external renderer. The internal renderer allows you to see the render happen on screen, whereas the external renderer simply renders the image and saves it to a location. For some reason with the internal renderer, the 2990WX is only using 1 of the 4 NUMA nodes (16/64 logical cores), effectively utilizing only 25% of its power. This is resulting in slower render speeds than the 1950X (which was also throttled here for reasons unknown to me but was still faster)!

With the external renderer, however, it utilizes all 64 logical cores, but then takes longer than the 1950X to set up each render . So, the render time itself is halved, but the time in between renders is longer than what I experienced before, resulting in external render speeds which are basically the same. My best guess is that this has something to do with the 2990WX only having four memory controllers resulting in latency... Should I have gone with the 2950X this whole time?

Then grasping at straws I started to wonder if my RAM had anything to do with this.

So, here I realized I didn't properly install the 4 sticks in the 8 DIMM sockets. I don't remember in what order, but initially I had them all on one side of the CPU (instead of 2 on one side and 2 on the other in the right sockets). This made the CPU prioritize NUMA Node 2 (instead of 0) when using the internal renderer. (I also did Cinebench tests with this random RAM setup, getting full CPU utilization and ok-ish scores for the 2990WX, hovering around 900, with the official 2990WX benchmark being at 1200).

After swapping the sticks around so that they were in the right DIMM sockets, NUMA node 0 is what's prioritized now, but now I'm getting significantly lower Cinebench scores (490, basically half). What happened?

(EDIT: this was a result of BIOS EZ tune settings- when I restored it to default, Cinebench test is normal, if not better than before)

Right now, I'm trying to see if the 2990WX is still viable, but I might be reaching the limits of my expertise.

I'm currently waiting to hear back from e-on support about their internal render, because I feel like I remembered there was a time when even my 1950X used all 32 logical cores with it. Maybe they updated it and something changed, but the 2990WX being limited to a single NUMA is pretty upsetting.

Your non-uniform memory hurts your performance. See page 1-6 of your user's manual.