Microsoft's DirectX update could slash ray tracing memory by up to 90 percent

Microsoft's latest DirectX update could significantly reduce the memory required for ray tracing by up to 90 percent. This development was detailed by the company on its DirectX developer blog, highlighting the significant changes to DirectX Raytracing since its release in 2018. The two primary features introduced are Clustered Geometry and Partitioned TLAS, which work together to optimize ray tracing processes.

Traditionally, when objects in a game deform or stream in at a new level of detail, the GPU has to rebuild the object's acceleration structure from scratch. However, with Clustered Geometry, models are divided into smaller chunks of up to 256 triangles. These clusters are built once and reused, meaning when an object changes, developers can reassemble existing clusters or rebuild only a few of them.

Partitioned TLAS takes a similar approach to the entire scene by dividing the world into regions. Only the parts where something moves get updated each frame, thereby optimizing performance and memory usage.

Microsoft showcased these improvements using a Radeon RX 9070 test sample. Switching from conventional rebuilds to Clustered Geometry resulted in a remarkable increase in frame rate, from 1.2 fps to 79 fps. The per-frame build time was drastically reduced from 788ms to just over 3ms, and the memory required for these structures shrank by approximately 90%, from roughly 7.5GB to 774MB.

This reduction is solely attributed to ray tracing data, not the game's overall VRAM footprint. Microsoft, however, cautioned that developers should not expect such dramatic improvements across the board. Clustered Geometry implementation requires substantial temporary scratch memory, which developers must plan for accordingly.

Interestingly, Nvidia had already been implementing a similar concept called RTX Mega Geometry since last year. In Alan Wake 2, for instance, RTX Mega Geometry led to both improved frame rates and image quality. Earlier this month, Gears of War: E-Day utilized RTX Mega Geometry to ray trace full-detail Nanite geometry, instead of using static, low-poly proxies typically employed by Unreal Engine 5.

The difference now is that Clustered Geometry and Partitioned TLAS are integrated into DirectX, providing a unified approach that works across different vendors. Importantly, no additional hardware is required for this feature to function; Nvidia supports it on every RTX card, and Intel covers its Arc B-series GPUs and recent Core Ultra chips.

However, like Nvidia, AMD's support for this feature is currently limited, with only the Radeon RX 9000 series having been tested so far.

While everything is still in preview mode and requires Windows Developer Mode to activate, the potential impact of these updates could be substantial. Developers are still some way off from incorporating these features into their games. Additionally, Nvidia's version of this feature is only compatible with GeForce cards boasting at least 12GB of VRAM, making it inaccessible for lower-end systems.