NVIDIA rolls out RTX upgrades for local generative AI workflows
The tech giant has announced a package of RTX software and performance updates aimed at improving speed and reducing memory use for local generative artificial intelligence tasks.
NVIDIA said it is introducing a set of RTX optimisations across GeForce RTX, NVIDIA RTX PRO and NVIDIA DGX Spark systems, targeting faster performance and lower memory consumption for on-device generative AI workloads.
The company said the updates include PyTorch-CUDA optimisations and native NVFP4 and NVFP8 precision support in ComfyUI, alongside RTX Video Super Resolution integration for real-time 4K video upscaling.
Additional enhancements include NVFP8 tuning for Lightricks’ LTX-2 model, a Blender-guided 4K video workflow and RTX acceleration for Nexa.ai’s Hyperlink application. Improvements to small language model inference are also being delivered through llama.cpp and Ollama, NVIDIA added.
According to the company, ComfyUI performance is up to three times faster with up to 60% lower video memory usage when using NVFP4 on RTX 50 Series graphics processors, and around twice as fast with 40% lower memory use using NVFP8. Open-weight checkpoints using NVFP4 and NVFP8 are being made available for models including LTX-2, FLUX.1, FLUX.2, Qwen-Image and Z-Image.
For search and language workloads, NVIDIA said Hyperlink enables local video content search on RTX-powered PCs, with significantly faster indexing and response times compared with central processing unit-based systems. The company also reported inference speed gains over recent months for llama.cpp and Ollama, with further updates due to appear in upcoming software releases.
Related reading
- NVIDIA unveils AI inference memory platform powered by BlueField-4
- NVIDIA releases open AI models and datasets to speed development
- PayPal unveils new analytics tools for advertisers at CES
NVIDIA added that NVIDIA Broadcast 2.1 extends support for Virtual Key Light effects to a wider range of RTX graphics cards, while DGX Spark systems have received performance updates delivering faster results and new deployment playbooks.
The company said several of the updates are available immediately, with further video and application features due to roll out over the coming weeks.
The Recap
- NVIDIA launches RTX upgrades to speed 4K AI video generation.
- Up to 3x faster generation and 60% less VRAM usage.
- Video workflow and RTX Video node available next month.
