GPUs based on the RDNA 3 architecture adopts 3D stack design
AMD seems to have invested heavily in stack cache and chiplet technology in both CPU and GPU. In terms of CPU, the EPYC processor code-named Milan-X, based on the Zen 3 architecture and equipped with 3D V-Cache technology, has been officially released. In terms of GPU, the Instinct MI200 series based on the CDNA 2 architecture and using MCM multi-chip packaging was also launched.
AMD introduced Infinity Cache technology on the RDNA 2 architecture GPU, and the GPU achieves higher access efficiency and bandwidth. The existing cache design reaches a maximum of 128MB, providing a bandwidth of 2 TB/s. On the RDNA 3 architecture, the cache capacity will be doubled. It is estimated that Navi 33 will be 256MB and Navi 31 will be 512MB. Since Navi 31 uses an MCM multi-chip package, there are two small chips for calculation, so each calculation module is still 256MB.
According to Twitter user @greymon55, Infinity Cache will also switch to 3D stack design, and GPUs based on the RDNA 3 architecture is likely to adopt this technology. This means that in the future AMD’s entire product line, including Ryzen, EPYC, and Radeon chips, will be equipped with 3D vertical cache technology. At present, the GPU based on the CDNA 2 architecture is the first product to use the MCM multi-chip package, but there is no information showing that the cache is equipped with 3D vertical cache technology.
There have been rumors that AMD’s RDNA 3 architecture GPU will have a considerable leap in performance. The Radeon RX series graphics cards equipped with Navi 3x cores are 3 times faster than existing products. AMD will also further polish ray tracing and FidelityFX Super Resolution (FSR) technology. It is expected that AMD Radeon RX 7000 series graphics cards will be launched at the end of 2022. It is believed that the competition between next-generation GPUs will be more intense.
