NVIDIA works on a 'tile' rendering technology that spreads the load between two GPUs

NVIDIA CFR technology is filtered, a kind of system that divides the frame into 'tiles' and allows two graphics to share the burden of rendering it.

The hardware industry is moving towards element communication systems to maximize performance. NVIDIA is known for its graphics cards and they use graphics interconnect systems using SLI y now NVLink. SLI technology is very outdated and NVLink technology is the replacement. It seems that NVIDIA is developing a rendering system based on 'tiles'

This technology that might seem new, is not so new. Tile-based rendering is one of NVIDIA's long-held secrets. It seems that a fairly primitive version of this technology (so to speak) was already in use at Maxwell. Now the company has evolved it and would be working on a multi-GPU solution.

No products found.

NVIDIA would be working on a new multi-GPU system

CFR would be what this new technology is called. CFR would be the abbreviation of 'checkerboard frame rendering' or 'checkered frame rendering'. A technology that would have already been implemented using current graphics drivers. Its integration has been done in secret and there will be no documentation for developers to implement it.

It appears that CFR would divide a screen into small squares or tiles, resembling a chessboard. Tiles with even numbers will be rendered by one GPU and odd numbers by the other. A very different technology to AFR (alternate frame rendering) where the memory of each GPU has a copy of all the resources necessary for the rendering of the frame.

CFR technology was allegedly less micro-stuttering relative to AFR. Only the DirectX API is used for this technology, it cannot be used in OpenGL and Vulkan. Furthermore, this technology is exclusive to the Turing architecture, since it requires NVLink to function.

There is still a lot of its popularization, since there would be quite a few compatibility problems. The documentation regarding this technology is quite scarce. These two factors mean that this technology has not yet been presented by NVIDIA. 

Source: TPU