NVIDIA announced that the TensorRT AI inference acceleration framework is now available on GeForce RTX graphics cards, with performance directly double that of DirectML.TensorRT is an inference optimizer launched by NVIDIA, which can significantly improve the operating efficiency of AI models. This time, NVIDIA introduces TensorRT to the RTX platform, so that all RTX graphics card users can enjoy faster AI performance.
According to NVIDIA, TensorRT not only supports native Windows ML, but can also automatically select the appropriate hardware to run each AI function and download the corresponding execution provider, greatly simplifying the development process.
In terms of performance,NVIDIA showed the comparison data between TensorRT and DirectML: in ComfyUI, users using TensorRT can get up to 2 times performance improvement; in DaVinci Resolve and Vegas Pro, the performance improvement reaches 60%.
In addition, TensorRT for RTX brings other benefits, such as 8x smaller library file sizes and on-the-fly optimizations for each GPU. These improvements not only improve performance, but also optimize the storage and management of the software.
TensorRT for RTX will be officially launched in June. NVIDIA is also continuing to expand its AI SDKs and is currently providing support for more than 150 AI SDKs. Five new ISV integrations will be launched this month, including LM Studio, Topaz Video AI, Bilibili, AutoDesk VRED and Chaos Enscape.
