Do we have any performance data from the new 5090 RTX cards?

[HowardDay]

I'm considering buying at least one of these monsters to speed up my rendering - and I'm wondering if the Chaos team has gotten hands-on performance data yet!
..I hope the news is good.

[Nejc Kilar]

Ha, the problem is, even if we did there is an embargo on sharing such information Sorry!

[nicolas_fuminier]

We will have to way until January 30th if i'm correct.

As far as I understand, with the spec, the price and the electrical consumption of the 5090, the only good news is the 32Gb of Vram. But i'm no expert and we will have to wait benchmarks to be sure.

[HowardDay]

Yeah, that's what I was afraid of - Looking at some released numbers it seems like an average 30% faster overall which...really isn't worth it at this point, especially for the price.

[nicolas_fuminier]

Yeah, that's what I was afraid of - Looking at some released numbers it seems like an average 30% faster overall which...really isn't worth it at this point, especially for the price.

I've sadly come to the same theoretical conclusion as you.

The extra 8GB of Vram is a real plus for us, but the price seems prohibitive for this addition alone.
For the rest, 30% more price for 30% more electricity consumption for 30% more performance seems almost a non-starter.

[nicolas_fuminier]

HowardDay

From Puget System and the Vray GPU Facebook goup :
https://www.pugetsystems.com/labs/ar..._V-Ray_Blender

with RTX (can't find source for CUDA)
If the 4090 was 2 x 3090, a 5090 is 3 x 3090 with 32go of VRAM.
I'll need to do the math to see what is the best move here compare to the future market price of the 3090, the 4090 and the 5090.
Not as disappointed as expect, but clearly not a must have either.

[akos_kiss]

An almost 50% uplift compared to the previous gen is great news! But you are right, the price/power consumpiton/ performance trio needs a bit more math to see if its the right choice.

edit. yeah my bad. i had to refresh my memory.... the 3000 to 4000 series was a huge jump in performance.

[nicolas_fuminier]

I'd love to hear what Muhammed_Hamed has to say about this new generation of cards (when he is allowed to), especially the 5090, if you've been able to get your hands on it and start technical testing.
Especially now that Vray7 is advocating CUDA over RTX. Is this a step in the right direction for hardware development at NVIDIA?
I have the impression that RTX is being promoted more than CUDA for generation 50, but the information is still sketchy.

[charles_messenger]

The big thing I’ve been wanting to know about this series of GPU is if there will be some kind of vram pooling via PCIe. I’ve kept with the 3090 cards for NVlink and vram doubling.

I remember hearing way back at an Nvidia presentation that they were saying “in the future” PCIe would be fast enough to support sharing vram between cards without the need for NVLink.

[Vizioen]

The big thing I’ve been wanting to know about this series of GPU is if there will be some kind of vram pooling via PCIe. I’ve kept with the 3090 cards for NVlink and vram doubling.

I remember hearing way back at an Nvidia presentation that they were saying “in the future” PCIe would be fast enough to support sharing vram between cards without the need for NVLink.

I wouldn't count on it in the near future.

[Vizioen]

Also this guy did a benchmark test, but I would take it with a bag of salt as they don't have the proper drivers.

https://www.youtube.com/watch?v=fOjrvXxSe0A

And there's this benchmark: https://benchmark.chaos.com/vray/v6/...x=1&order=desc but that doesn't seem correct either, 4090 has like 3x that score.

[charles_messenger]

Also this guy did a benchmark test, but I would take it with a bag of salt as they don't have the proper drivers.

https://www.youtube.com/watch?v=fOjrvXxSe0A

And there's this benchmark: https://benchmark.chaos.com/vray/v6/...x=1&order=desc but that doesn't seem correct either, 4090 has like 3x that score.

pugent systems was saying there was currently an issue with the 5090s and rendering with the CUDA engine

https://www.pugetsystems.com/labs/ar..._V-Ray_Blender

[Muhammed_Hamed]

Hello everyone,

We do have an RTX 5090 now, although we need wait until the 30th for proper drivers. I will publish a blog post in the upcoming days about the performance and all relevant information
>> The RTX engine in the current V-Ray builds work well with the new Blackwell GPUs, no changes are required to support the new generation
>> The CUDA engine has an issue where it shows reduced performance as some of you noticed in the Puget thread, this will be solved with the upcoming driver around mid February. For now if you have one of the new cards, please stick to the RTX engine.
>> The RTX 5090 seems to be 40% faster than the RTX 4090 with the RTX engine, while consuming 25% more power. I will share more testing in my post soon

Best,
Muhammed

[Muhammed_Hamed]

The big thing I’ve been wanting to know about this series of GPU is if there will be some kind of vram pooling via PCIe. I’ve kept with the 3090 cards for NVlink and vram doubling.

I remember hearing way back at an Nvidia presentation that they were saying “in the future” PCIe would be fast enough to support sharing vram between cards without the need for NVLink.

We asked Nvidia, there is a chance we might see such a feature in the future. They seem to be working on that still

[Muhammed_Hamed]

I'd love to hear what Muhammed_Hamed has to say about this new generation of cards (when he is allowed to), especially the 5090, if you've been able to get your hands on it and start technical testing.
Especially now that Vray7 is advocating CUDA over RTX. Is this a step in the right direction for hardware development at NVIDIA?
I have the impression that RTX is being promoted more than CUDA for generation 50, but the information is still sketchy.

The RTX performance is better because of the compiler issue with the CUDA engine, it should be fixed around mid February
Redshift, Octane and other engines don't work at all with the new GPUs. Our RTX engine works well without needing any changes

Best,
Muhammed