At the end of July, data from AMD’s presentation became public, in which it promised to bypass the NVIDIA A100 computing accelerator in single precision calculations using its own solution named Instinct MI100 on a chip with a new CDNA architecture. As it became known today, products with CDNA architecture will be presented by the company on November 16.
We came across this information quite by accident while studying the October edition of the AMD corporate identity book, which anyone can get from the official section company website for partners. The book was updated to reflect the new Radeon logo style, which was unified with the font used to denote Ryzen processors, and the information about Instinct accelerators was not particularly striking.
And yet, AMD felt it necessary to warn the casual reader that information about the computational accelerators of the Instinct family is under a nondisclosure agreement until November 16 this year. Tellingly, following the example of NVIDIA, the company has excluded associations with gaming video cards from the official name of accelerators – “Radeon” in this sense has become an atavism.
In addition to the flagship Instinct MI100, the family will include the Instinct V640 and Instinct V620 accelerators. Most likely, all three models will be able to combine into computing clusters with similar ones. Senior Instinct MI100 will allow combining from four to eight accelerators in one system. There is even an image of a fragment of the casing of the cooling system of this product and its advertising banner.
AMD explains that Instinct products can be called both processors and GPUs (graphics processing units), but their name cannot be written in capital letters alone. In the future, the CDNA architecture will evolve independently, regardless of RDNA game graphics.
If you notice an error, select it with the mouse and press CTRL + ENTER.