AMD advances shader optimization in games with new GPU chipset patent


Written by admin



AMD recently published a patent for spreading the rendering load across multiple GPU chipsets. The game scene is divided into separate blocks and distributed among the chiplets to optimize the use of shaders in games. For this, two-level chiplet binning is used.

AMD Publishes Patent on GPU Chiplets to Better Use Shader Technology

A new patent published by AMD provides more insight into what the company plans to do with next-level GPU and processor technologies in the coming years. At the end of June, it was discovered that 54 patent applications had been submitted for publication. It is not known which of the more than fifty published patents will be used in AMD’s plans. The applications discussed in the patents detail the company’s approaches in subsequent years.

One application, which was seen by community member @ETI1120 on the ComputerBase website, patent number US20220207827, discusses critical image data in two steps to effectively offload rendering loads from the GPU to multiple chiplets. The CPU originally filed an application with the US Patent Office late last year.

When image data is rasterized on the GPU by standard means, the shader unit, also known as the ALU, performs the same task and assigns a color name to individual pixels. In turn, textured polygons found at a particular pixel in a particular game scene are mapped directly to the pixel. Finally, the formulated problem will support atypical principles and differ only in other textures located in different pixels. This method is called SIMD, or Single Instruction – Multiple Data.

For most modern games, shading is not the only task that the GPU performs. But instead, several post-processing elements are included after the initial shading. Actions that the GPU can add, for example, will be anti-aliasing, shading, and occlusion of the game environment. However, ray tracing happens in tandem with shading, creating a new calculation method.

Speaking of the GPU driving graphics in today’s games, the load generated by a computer is exponentially increasing to thousands of computational units.

In GPU gaming, this compute load scales perfectly to several thousand compute units. This differs from processors in that applications must be specially written to add additional cores. The CPU scheduler creates this action by dividing the work of the GPU into more digestible tasks handled by compute units, also called binning. The image from the game is rendered and then divided into separate blocks containing the specified number of pixels. The block is computed by the GPU division, where it is then synchronized and created. After this action, pixels awaiting calculation are included in the block until the video card sub-block is finally used. Shader processing power, memory bandwidth and cache sizes are taken into account.

Source: AMD via ComputerBase.

AMD explains in the patent that splitting and merging requires a full and complete data connection between all GPU elements, which poses a problem. Off-chip data links have an increased level of latency, which slows down the process.

Processors easily migrated to chiplets due to the ability to transfer a task across multiple cores, making it accessible to chiplets. GPUs don’t have that flexibility, so their scheduler is comparable to an entry-level dual-core processor.

Source: AMD via ComputerBase.

AMD recognizes the need and is trying to find answers to these problems by changing the rasterization pipeline and sending tasks between multiple GPU chiplets, similar to CPUs. This requires cutting-edge binning technology, which the company introduces “two-level binning,” also known as “hybrid binning.”


In hyperbinning, the division is processed in two separate steps instead of being processed directly into pixel-by-pixel blocks. The first step is to calculate the equation by taking the 3D environment and creating a 2D image from the original. The stage is called vertex shading and is completed before rasterization, and on the first GPU chiplet this process is minimal. Once completed, the game scene begins to be binned, turning into rough bins and being processed into a single GPU chiplet. Then you can proceed to routine tasks, such as rasterization and post-processing.


It is not known when AMD intends to start using this new process or if it will be approved. However, it gives us a glimpse into the future of more efficient GPU processing.

News Sources: ComputerBase, Free Internet Patents.

#AMD #advances #shader #optimization #games #GPU #chipset #patent



About the author


Leave a Comment