Updated GPU limitations?

I’m ready to hit go on getting 3 new updated workstations with the 5090’s in them, and coming from 7 years of rendering with CPU I just want to get my expectations in line, is there a list of absolute limitations when GPU rendering? Like I think radial brushed still doesn’t work, stuff like that. I am going with the Threadripper 7960x processors, so we won’t be totally dead in the water as far as rendering goes if we hit a snag with the GPU’s.

Was going to go with the new Intel 285k but looks like it doesn’t have enough PCIE lanes if we happened to add a second GPU in a couple years.

Hi, @matt.gerard -

For me, a big deal breaker is rounded edges are broken. When you are a certain distance away from the geometry, any rounded edges you’ve set will not draw/render. This hurts when dealing with CAD data where it is impractical to model edge breaks into it. As of 2024.3 this issue remains unresolved, but I am assured it will be resolved “soon.”

That is a big one. I wouldn’t say its mission critical to my stuff, but when I use it, it really does help.

I wouldn’t care about PCIE lanes. I’ve two GPUs and I’m still on a i9-9900K with PCIe 3.0 and not that many lanes as well. Doesn’t impact GPU performance since all info is loaded in the GPU and the amount of data it transfers back to CPU is just the rendered bitmap. My CPU and DDR4 does make it a bit slower to get a scene into the GPU but that’s all and barely noticeable.

Same with the PCIe versions, 3.0, 4.0 and now 5.0 gives only a very tiny difference in actual GPU performance. In games where constantly textures are going forward the GPU the fps difference is just 1% to max 4% in some games. With renders it will be way less since nothing is transferred once the scene is inside the GPU memory.

Threadrippers are great if you still have some things that love high amounts of cores. I’m also orientating a bit these days and I wonder about the upcoming AMD 9950X3D which will be March I think. That would be no competitor against the Threadripper but I would certainly prefer it over the Intel 285K.

An offical list of GPU limitations you find on this page in the yellow box around the bottom of the page: GPU Mode

I watched some clips about the Founders Edition 5090 and it’s really a great piece of engineering if you look at the different PCB’s so they were able to make it just 2 slots and 2/3 of the card is basically see through with cooler fins. If you think about having 2x5090 in one case I think those or for example the liquid cooled GigaByte 5090 are the only options since other cards are close to 4 slots.

The Founders Edition does seem really noisy from what I heard but if you’re going for a liquid cooled one it’s good to keep in mind your case would be able to fit the radiator(s). I’ve currently the liquid cooled 4090 from GigaByte and placed it vertically and behind it is the 3090. That fits but only because the 3090 I have ain’t that wide.

If you want to have two cards horizontally, just make sure your mainboard has the PCI slots separated enough. My board doesn’t so the extension cord > vertical was my only option.

I didn’t have issues with round corners myself as Dan mentioned but it could be because I don’t use a lot of nurbs models. Not sure if it could help but maybe if you convert the nurbs from the scene it would work without an issue.

The Brushed Radial node seems to work again in 2025.1 Beta:

seems, that this issue is fixed in the latest 2024.3

1 Like

thanks for the replys everyone, I’m getting these machines built by Boxx, so whatever MB’s they are using will support 2 GPUs and cooling. Which will be liquid cooled. That makes me a little nervous, as the HP workstations are all air cooled, and cooled quite well i must say. They are pretty quiet as well.

I’m still torn on the Intel vs AMD, as the intel would save me quite a bit of cash. I’m going to make a field trip today to the local MicroCenter and see if they have any towers with the 285k in it and ask if I can run the KS Benchmark on them to see where it sits. Either way it will be faster than the dual xeons I have now.Even the old school i914xxxx series are faster in the benchmark than my xeons!

And true about the PCIE lanes, I do forget that the nature of rendering with GPU is one large transfer to the GPU then it just chugs away. All the stuff written is about gaming, so constant updating of the VRAM is a lot more intense. That right there would bias towards the Intel as a cost performance comparison. Might even wager a glance back to the Ryzen high end line.

Ugh. just want to pull the trigger and be done! There’s never a perfect time to buy new technology :slight_smile:

I can really imagine you want it being done, I must say I hoped Intel would bring some actual new things past October, but they were a bit disappointing. The reason I think for me the 9950X3D is interesting because it combines really nice performance in games with a nice amount of cores and high frequency. More competitive to the 285K in business applications than the current top consumer one of AMD. But like you say, always hard to pick the right moment.

Today I read some rumours AMD will maybe sooner announce their latest processors but those are rumours :slight_smile:

I’m sure Boxx will know what they are doing with the cooling. I think the custom cooling solutions often look also awesome but wouldn’t dare starting such project myself if I had the funds. What I really like about my liquid cooled 4090 is that it’s really silent and also stays really cool. It’s also clocked a bit higher than other cards. Since the temperatures stay so low it’s also easy to overclock it some more of lower the max. power usage to save some energy while keeping the same speed.