AMD recently released a patent for distributing the rendering load across multiple GPU chiplets. To maximize the use of shaders in games, a game scene is broken up into small blocks and distributed to the chiplets. This uses two-level chiplet binning.
AMD issues a patent for its use of GPU chiplets to enhance the use of shader technology.
The latest AMD patent provides more information about the company’s future plans for cutting-edge GPU and CPU technology. 54 patent applications were revealed to have been submitted for publication at the end of June. Which of the more than fifty published patents will be used in AMD’s plans is unknown. The applications covered by the patents describe the company’s methods moving forward.
One application, patent number US20220207827, which was spotted by community member @ETI1120 on the website ComputerBase, discusses crucial image data in two stages to effectively distribute the loads from a GPU’s rendering over many chiplets. This was first submitted by CPU to the US Patent Office at the end of the previous year.
The shader unit, also known as the ALU, completes the same task and names each pixel’s color when image data on a GPU is rasterized using conventional methods. Then, the textured polygons that are present at that particular pixel in the game scene are directly mapped onto that pixel. Last but not least, the task as formulated will maintain unusual principles and only vary through other textures located at various pixels. Single Instruction – Multiple Data, or SIMD, is the name of this approach.
The majority of modern games require a GPU for more than just shading. But after the initial shading, a number of post-processing elements are added. For instance, anti-aliasing, shadowing, and occlusions of the gaming environment are effects that the GPU would add. However, shading and ray-tracing work together to create a new calculation technique.
The amount of load generated by the computer is exponentially increased into thousands of computing units when discussing the GPU controlling the graphics in current games.
This computing load scales up to several thousand computing units in games on GPUs in a somewhat ideal way. In contrast to processors, this requires applications to be specifically written in order to add more cores. This action, known as binning, is created by the CPU scheduler by breaking up the work from the GPU into smaller tasks that can be handled by the compute units. The rendered game image is then divided into distinct blocks, each of which has a specific number of pixels. A component of the graphics processor computes the block, which is then synched and created. After that, pixels that need to be calculated are added to a block until the graphics card’s sub-unit is finally called upon. The memory bandwidth, cache sizes, and shader computing power are taken into account.

In the patent, AMD explains that there is a problem because the dividing and joining require a complete and thorough data connection between all of the GPU’s components. The process is slower because external data links have higher latency than die-based links.
Because CPUs can distribute the task across multiple cores, making it suitable for chiplets, they have easily made the transition to chiplets. Due to their lack of flexibility, GPU schedulers are comparable to those of entry-level dual-core processors.

AMD is aware of the issue and makes an effort to address it by changing the rasterization pipeline and distributing the work among several GPU chiplets, much like CPUs. For this, the company is introducing “two-level binning,” also referred to as “hybrid binning,” which is a cutting-edge binning technology.


Instead of processing the split directly into blocks of pixels, hyper binning divides the processing into two stages. The first step is to solve the equation, taking the original three-dimensional environment and turning it into a two-dimensional image. Vertex shading is the stage that is finished before rasterization, and it takes up very little space on the GPU’s first chiplet. When the game scene is finished, the binning process starts, developing into coarse bins and processing into a single GPU chiplet. After that, routine tasks like post-processing and rasterization can start.
When AMD plans to start using this new procedure and whether it will be approved are both unknown. However, it provides a preview of more effective GPU processing in the future.