In a few months, the new architecture of graphics cards from AMD will debut. We are talking about RDNA 3, which will replace the previous RDNA 2 and will face the new generation of NVIDIA in the form of the Ada Lovelace architecture and GeForce RTX 4000 systems. We do not know much about the architecture itself, except that the manufacturer plans at least three graphics chips: NAVI 31, NAVI 32 and NAVI 33. According to unofficial information, the last of the chips is to offer performance similar to the Radeon RX 6900 XT, and the top NAVI 31 is to significantly boost the GPU’s capabilities compared to the previous top. We also expect a strong increase in performance in the context of Ray Tracing and (perhaps) the use of artificial intelligence algorithms to significantly improve the FSR technique. One of AMD engineers, in turn, confirmed important information about the upcoming NAVI 3X systems.
Just a few days ago, the CEO of AMD, Dr. Lisa Su, confirmed that before the end of 2022, a new GPU architecture in the form of RDNA 3 will debut. It has long been said that it will partially use the MCM design, similar to what was presented on the occasion of AMD graphics accelerators Instinct MI250 Series (CDNA 2). According to an entry on LinkedIn (the annotation from the site has already been removed) Brian Walters, one of the engineers at AMD, is currently working on new graphics chips, including NAVI 31, NAVI 32 and NAVI 33. The first two chips also announced the use of two separate technological processes of TSMC – 5 nm and 6 nm.
We would like to remind you that according to the information available so far, the main computational matrices in the NAVI 31 and NAVI 32 systems are to take advantage of the lower, 5 nm technological process. In turn, an additional MCD block – Multi Cache Die – is to be produced using 6 nm lithography. The situation is different in the case of the NAVI 33 system – this one is to be entirely manufactured using the 6 nm TSMC technological process. Interestingly, the profile also includes information about the work on the next generation of GPUs for the HPC / AI market – AMD Instinct MI300 Series. The new chips, certainly based on the CDNA 3 architecture, are to be produced in 6 nm lithography (CDNA 2 uses the 7 nm TSMC process).