Earlier today, Intel’s Raja Koduri shared one of the API-level features of the Xe-HPG (High-Performance Gaming) graphics cards using the yet unreleased 3DMark Mesh Shader Feature test. Mesh shading is one of the core features of DirectX 12 Ultimate. It makes the graphics pipeline more flexible and multi-threaded and also allows for early culling of primitives, greatly improving performance even using modest hardware.
We already know that Intel’s upcoming gaming products will support hardware-accelerated ray-tracing, reportedly as well as NVIDIA’s RTX cards, at least on par with Turing if not Ampere. The inclusion of Mesh Shaders means that we’re likely looking at hardware that supports the latest version of DX12 with support for the other accompanying technologies such as Variable Rate Shading (VRS), Sampler Feedback, and more.
The Xe-HPG graphics cards were originally supposed to roll out last year but were delayed to 2021. Intel’s first discrete GPU lineup for the gaming market will be fabbed on an external process node (most likely TSMC’s 7nm) and come with GDDR6 memory. The core counts are expected to max out at 4,096 shaders or 512 EUs. In comparison, the fastest iGPU powering the higher-end Tiger Lake SKUs top out at 96 EUs. There has been no word on Intel’s implementation of ray-tracing, but we’re likely going to see dedicated hardware that speeds up ray-triangle intersection testing and BVH traversal similar to NVIDIA’s RTCores.
We expect Intel to go with a more hardware-centric approach (like NVIDIA) where the entire ray-tracing pipeline is offloaded to dedicated hardware. AMD uses a hybrid approach where the Ray-Accelerators are used for ray-box/ray-triangle intersections while BVH traversal and shading is handled by the vector units. This is the primary reason why the Big Navi GPUs perform rather terribly in most ray-traced titles.
