In the RTX 50 (Blackwell) generation the programmable shader can process neural networks.

Judge as of 2025-01-07, when the RTX 50 series (Blackwell) was announced at CES. True: Nvidia technical documentation or reliable coverage shows that RTX Blackwell lets programmable shaders run neural networks (for example 'neural shaders' or 'RTX Neural Shaders', with Cooperative Vectors or similar APIs giving shader code access to Tensor Cores), and this is presented as a feature of this generation. Mostly true: the feature exists in Blackwell, but the neural-network work runs on Tensor Cores invoked from the shader rather than on the shader cores themselves, or the feature is also partly available on earlier RTX generations, so 'this generation' slightly overstates it. Mostly false: shaders could already process neural networks to essentially the same extent in earlier generations, so the claim presents an old capability as new, or the feature was only announced as future software and not supported on RTX 50 at launch. False: no such capability exists in the RTX 50 generation.

Set before the research began.

1. E1 NVIDIA Blackwell GeForce RTX 50 Series Opens New World of AI Computer Graphics NVIDIA (nvidianews.nvidia.com) · 6 January 2025 Own record

   “Alongside GeForce RTX 50 Series GPUs, NVIDIA is introducing RTX Neural Shaders, which brings small AI networks into programmable shaders, unlocking film-quality materials, lighting and more in real-time games.”

2. E2 NVIDIA RTX Neural Rendering Introduces Next Era of AI-Powered Graphics Innovation | NVIDIA Technical Blog NVIDIA Technical Blog (developer.nvidia.com) · 7 January 2025 Own record

   “Alongside the new GeForce RTX 50 Series GPUs powered by NVIDIA Blackwell architecture, NVIDIA is introducing RTX Neural Shaders, which bring small neural networks into programmable shaders, enabling a new era of graphics innovation.”

3. E3 NVIDIA RTX Neural Rendering Introduces Next Era of AI-Powered Graphics Innovation | NVIDIA Technical Blog NVIDIA Technical Blog (developer.nvidia.com) · 7 January 2025 Own record

   “For DirectX developers, Microsoft will be adding Cooperative Vector support to DirectX in the near future. This will enable neural shaders to take advantage of accelerated performance with RTX Tensor Cores.”

4. E4 V1.0 NVIDIA (images.nvidia.com) · 28 January 2025 Own record

   “With the new Cooperative Vectors API for DX12 and Vulkan, Tensor Cores can be accessed through any type of shader, including pixel and ray tracing, in a graphics application allowing for a host of neural technologies.”

Nvidia's CES press release and developer blog (E1, E2) present RTX Neural Shaders as part of the GeForce RTX 50 launch and say they bring small neural networks into programmable shaders. The Blackwell architecture whitepaper (E4) says the Cooperative Vectors API lets any type of shader access Tensor Cores. NVIDIA's material describes the neural-network work in shaders as using Tensor Cores accessed from shader code [E3][E4] (E4 is dated after the announcement), rather than describing it as running on the shader cores alone, which matches the criteria's description of mostly true. E3 adds that DirectX support for Cooperative Vectors was still to come "in the near future" as of the announcement. All of this evidence is Nvidia's own official material.