I’ve been using Topaz Video AI for years. Recently, I added an Nvidia RTX PRO 6000 graphics card to my system. However, when running Topaz Video AI with the RTX PRO 6000 alone, it’s actually slower than my previous RTX 5080. During processing, the GPU utilization and CUDA core usage consistently stay below 40%. Additionally, when I install both GPUs and enable Topaz’s multi-GPU support, the overall speed is even lower than single-GPU mode, and the software’s GPU scheduling appears quite inefficient.
Could you please advise how I can resolve this low GPU utilization issue?
Server, Workstation and Q-Max, three different kind of coolers and power consumption.
The RTX 5080 has 30 different versions.
And your CPU has only 24 PCI-Lanes.
The two GPUs alone need 32 PCI-E lanes.
So, this is what gemini said about.
Even though the RTX 6000 Blackwell has superior raw hardware, it gets bottlenecked in your specific setup. Because the Ryzen 9 9950X is a consumer CPU, installing both cards forces your motherboard to split the PCIe lanes into an x8 / x8 configuration, starving the massive RTX 6000 of data bandwidth. Combined with the RTX 5080’s significantly higher clock speeds and gaming-optimized drivers, the consumer card will easily beat the restricted enterprise giant in any software that cannot fully utilize all 24,000+ workstation cores simultaneously.
To unleash the full potential of your RTX 6000 Blackwell without it being bottlenecked, you should run it on a dedicated workstation platform rather than a consumer desktop setup.
Switching to a system powered by an AMD Ryzen Threadripper (Pro) or Intel Xeon W processor is highly recommended. These enterprise-grade platforms provide a massive amount of PCIe lanes, allowing you to run both your RTX 6000 and RTX 5080 simultaneously at full x16 / x16 bandwidth without any performance penalties.
Alternatively, if you want to keep using your current Ryzen 9 9950X setup, you should operate the RTX 6000 as a standalone GPU (completely removing the RTX 5080) to ensure it gets the full 16 PCIe lanes all to itself.
I recently found that TOPAZ VIDEO has an issue with VRAM usage. When I run STARLIGHT 2.5, the software can only utilize 24GB of VRAM on my graphics card (which actually has 96GB of VRAM). This forces it to frequently exchange data with the RAM. How can I remove the 24GB VRAM usage limit of the software? This would significantly speed up the processing.
Not sure on that situation, or where the limit cap would be getting implemented from. I can check with the devs to see if they know and get back to you.
