The most powerful RDNA 3 GPU starts to show, which is a monster

AMD engineers have domestically agreed on the design of the most powerful future Navi 31 chip in the RDNA 3 range. Expected at the end of 2022, it is more than optimistic.

AMD’s next generation graphics cards begin to make a name for themselves… And it promises // Source: AMD via WCCFTech

We know, teamsAMD We have been working for many months to develop the RTNA3 graphic architecture. The preliminary leaked map allows us to get the first idea of ​​what we find out there today … and let’s just say that AMD is not disconnected.

The next generation flagship graphics card is registered.

– Greymon55 (@ greymon55) October 29, 2021

Having said that, and Mentioned by WCCFTech, This map shows an important step taken internally in the development of RDNA 3 chips, new release Graphics cards AMD is no more. The latter should not be issued before the second half of next year. Because, if AMD agrees with itself on this design, it will have to carry out more electrical tests and then send the first engineering models to its partners.

An architecture that develops in depth

From a technical standpoint, we know that AMD has dropped the computing units (CUs) of its older GPUs in favor of workgroup processors (WGP) anyway. But this is not the only change we see clearly.

The configuration of the Navi 31 GPU released by WCCFTech reveals the responsibility of connecting two GCDs (graphics core tie) and one MCD (multi-cache tie) to the new Infiniti fabric interface. Each GCD has 3 shader engines (blue on the map) and each shader engine has 2 shader rows (light blue). So the chip has 6 shader engines, which in turn connects a total of 12 shader rows together.

Each shader array is made up of 5 task force processors (60 in total on the GPU), and each WGP has 8 SIMD32 units. These SIMD32 units are assembled and refers to WCCFTech, which produces at least 7,680 cores through the graphics core die, so the Navi 31 chip has a total of 15,360 cores. We finally learned that a GPU must contain a total of 8 memory controllers in 32 bits. A 256 bit bus.

Source: Allrock via WCCFTech

Want to see what Nvidia has in store. Because, if AMD wants to rumor that it can beat its competitor in the rasterization (traditional calculation process) shows, Nvidia can regain the advantage in ray tracking with its next chips. The company will adopt the MCM (Multi-Chip Module) design from 2022. Ampere with GPUs is currently more than twice the potential performance.