So hearing / seeing a lot of awesome feedback here about 20.5 Karma XPU and 2-3x speed increase over Houdini 20 when using RTX 4xxx. I have 3090 with 24GB RAM… Thing of downgrading/upgrading to 4070 Ti Super. It has about 30% less CUDA cores than 3090 it seems and obviously less RAM, but from what I understand the speed increases in Karma XPI are significant due to 4xxx hardware optimizations,

Question - are there any other areas of Houdini, such as maybe Copernicus, Dynamics, or Vulcan viewport that also utilize 4xxx hardware better and hence would be faster on 4070 than 3090?

Thank you

I would assume that anything that uses the GPU stands to benefit from a more powerful one. What that means in real-life, practical terms is difficult to say. How hard will you be pushing Houdini?

Sorry that wasn’t my question. My question was were there any specific areas, other than karma xpu, that would leverage new capabilities of 4.xxx cards. Was more of a SideFX question.

Karma XPU was optimized for the Nvidia ADA Shader Execution Reordering found in the Nvidia ADA Lovelace GPU, so my understanding is that it only affects shader efficiency and rendering.

For more info...

Nvidia ADA Shader Execution Reordering [developer.nvidia.com]

To give you a better idea which GPU is best for 3D, look at the Blender Open Data page, it shows the results for the Blender benchmark, higher is better...

Blender Open Data [opendata.blender.org]

RTX 4070 TI Super (Windows): 7291.23
RTX 4070 TI Super (Linux): 7577.32

RTX 3090 (Windows): 5491.67
RTX 3090 (Linux): 5712.01

Hope this helps

LukeP
So hearing / seeing a lot of awesome feedback here about 20.5 Karma XPU and 2-3x speed increase over Houdini 20 when using RTX 4xxx. I have 3090 with 24GB RAM… Thing of downgrading/upgrading to 4070 Ti Super. It has about 30% less CUDA cores than 3090 it seems and obviously less RAM, but from what I understand the speed increases in Karma XPI are significant due to 4xxx hardware optimizations,

I have the same gpu and personally I'd never downgrade RAM! A little fast render or sim here and there is nothing compared to the problems you'd have with too little memory.

There are multiple optimizations in Karma XPU for Optix8. The ADA cards give about a 1.5 to 2 times improvement. But there are additional optimizations that we take advantage of that also improve performance on older cards. So you should still see a speed boost on 3090 cards in most scenes (it's obviously dependent on the scene). You won't see as huge an improvement on older cards, but there still should be performance improvements.

Karma XPU is using RTX specific features that other parts of Houdini do not use.

mark
There are multiple optimizations in Karma XPU for Optix8. The ADA cards give about a 1.5 to 2 times improvement. But there are additional optimizations that we take advantage of that also improve performance on older cards. So you should still see a speed boost on 3090 cards in most scenes (it's obviously dependent on the scene). You won't see as huge an improvement on older cards, but there still should be performance improvements.

Karma XPU is using RTX specific features that other parts of Houdini do not use.

Thank you kindly. So in summary 4070 should be in general faster than 3090 in karma xpu but in other areas of Houdini where OpenCL is used - 3090 would be faster?

LukeP
Thank you kindly. So in summary 4070 should be in general faster than 3090 in karma xpu but in other areas of Houdini where OpenCL is used - 3090 would be faster?

Here is an OpenCL benchmark...

OpenCL Benchmark [browser.geekbench.com]

Great list. Thank you. So 3090 and 3070 are very close on OpenCL. Guessing that means close to same performance with non karma xpu stuff.

4070 seems to win when it comes to ray tracing optimization where 3090 seems to win due to more ram.

Tough call.

Like I said, it's a toss between VRAM, rendering power, and price, I choose the RTX 4070 TI Super because I felt that it was the best bang for my buck, 16 GB of VRAM filling most of my needs, but if you need more VRAM, than either stick to the RTX 3090, which is still a great card BTW, but you'll miss on render power, the RTX 4090 is also an option, of course, but they cost double the RTX 4070 TI Super!

There are rumours about an RTX 5xxx series coming out by the end of the year, the RTX 5090 could have as much as 32GB of DDR7 VRAM and be 70% faster than the RTX 4090, you might want to wait for more info on those, but so far it's only a persistent rumour, no official announcement from Nvidia!

I just read the chart, the RTX 4070 TI Super is 15.67% faster than the RTX 3090 for OpenCL, it might not be enough of a reason to buy an RTX 4070 TI Super, and the RTX 3090 has more ram, the only reason to get a RTX 4070 TI Super would be much faster ray-tracing, but you get 16 GB of VRAM instead of the 3090 24 GB, it really depends on the kind of work you do!