I used to have Cuda Toolkit 12.8 installed, but today I uninstalled it and installed version 13.1. Proteus render speed increased by 2x for me. I recommend it to you as well. For those using Nvidia GPUs, I recommend installing the latest version of Cuda Toolkit and then restarting your PC.
2 Likes
How can I use the Cuda 13.1 with Topaz Video?
I’ve not tried that, but read this
Topaz Video AI does not use your system-installed CUDA Toolkit at runtime.
Topaz ships with its own CUDA runtime, cuDNN, TensorRT, and kernels, compiled and bundled inside the app.
@JakSpoon
I used to do the same process at 30fps, but after switching to CUDA Toolkit 13.1, it did it at 60fps.
I asked Gemini and they wrote back:
“Backup Parachute” System (JIT Compilation)
The libraries embedded in Topaz (Bundled libs) contain pre-compiled (ready-made) code for the hardware available at the time of release.
Problem: The files inside Topaz may not yet “recognize” the RTX 5070 Ti (Blackwell architecture). They do not contain a code block specifically optimized for this card (Binary SASS).
Solution (PTX): In this case, Topaz has to use NVIDIA’s general/intermediate code, which we call “PTX”. For this code to work, it needs to be translated into the appropriate language for the graphics card by the system at that moment, during runtime (JIT Compile - Just In Time).
Why Did It Make a Difference in Your Case?
For a normal user (e.g., an RTX 3060 owner), the Toolkit in the system is irrelevant, because Topaz already contains ready-made code for the RTX 3060. There are.
However, in your case (RTX 5070 Ti):
Topaz’s internal library didn’t recognize your card.
It delegated the process to the system’s driver and compiler (PTX JIT).
Because you installed the CUDA 13.1 Toolkit (and the latest driver components that come with it), your system started performing this “instant translation” process much faster and more efficiently, specifically for the Blackwell architecture.
In short:
General Rule: Topaz doesn’t use the system’s Toolkit (it uses its own files).
Exception (Your Case): If your hardware (RTX 5070 Ti) is newer than the libraries in Topaz, the software will rely on the system’s tools (Driver JIT Compiler).
The 2x speed increase you experienced is not due to Topaz’s own library, but rather to the system infrastructure updated with CUDA 13.1, which translates the commands sent by Topaz much better to your new card. In other words, both pieces of information are related to your own system. It is true in that context.
1 Like
I’m using Windows 10, and Gemini said that upgrading to Windows 11 would improve performance. Has anyone tested this?
Yeah the toolkit is for compiling. Though someone made this same claim a few years ago, so I had CUDA 11 installed on this machine. I went ahead and updated it to 13 and well, the speeds are the same—as expected.
2 Likes
All reports on these forums say that Windows 11 is slightly slower. They are a few years old though.
Everything I can find says that the Nvidia drivers already include the current CUDA runtimes, and the Toolkit is unneeded.
5 Likes
Iris Medium was previously processing 720p to 720p at 30fps, but after installing CUDA 13.1, it now processes at 68.2fps and uses the TRT model.
GPU: 5070Ti
I just tried installing it, and it made no difference on my system.
1 Like
