Double precision floats (IEEE-754 binary64) are useful for various applications. Although often slower than 32-bit single precision floats, they are supported by many desktop GPUs as well as some mobile ones. They are exposed either directly or via extensions and optional capabilities in OpenGL, DirectX, Vulkan, CUDA and OpenCL.

There are many scientific areas that have use cases that require double precision floating point formats such as physics, GIS, simulations or astronomy.

Since WebGPU exposes GPGPU features on the Web via compute shaders, it will be beneficial to expose 64bit floats when available/emulatable, e.g. via an extension, similarly to how 16bit floats are exposed.

Extension requirements:

• f64 type available in WGSL shaders representing IEEE-754 binary64
• Expose 64bit vertex formats
• Allow read and write storage buffers to be typed as 64bit floats