The graphics cards from AMD could be very different in the future and at least after a newly discovered patent switch to an MCM design (multi-chip module).
Notebookcheck.net highlighted the discovery of the patent by the hardware leaker @ davideneco25320 on Twitter, and it’s sure to be interesting read that offers a potential glimpse of how AMD is changing its GPU design to keep rising graphics card prices under control and better to keep competing with Nvidia (and indeed Intel for that matter).
https://t.co/8K9XmmiOjOChiplet gpu from amdJanuary 1, 2021
The simple idea, in simple terms, is to use MCM or multiple chips (“chiplets”) on one board – as AMD is already doing with its Ryzen processors – as opposed to the current monolithic design (single chip).
Moving to MCM could bring a number of benefits in terms of ensuring better returns as graphics cards continue to perform and their design becomes increasingly challenging to identify and implement while keeping costs reasonably low. As we’ve seen lately, GPUs, or certainly the more powerful ones, have already become very expensive.
However, there are serious problems changing an MCM model in how graphics cards work. However, the AMD patent describes how these delicate problems can be solved.
New way forward
The main stumbling blocks to an MCM design lie in the fact that games are specifically programmed to work with a single GPU. Therefore, this new approach – effectively using multiple GPUs on a single board – is problematic in this regard. In addition, it is difficult to implement parallel workloads across multiple chips and at the same time keep the memory contents synchronized across them.
AMD’s solution in the patent is to connect these GPU chiplets via passive, high bandwidth networking, with one of these chiplets being the primary GPU, so to speak, that is directly connected to the CPU – which means that the processor (and the operating system ) would see this The graphics card is just a single (monolithic) entity in terms of coding software or games for it.
In order to solve the memory problems mentioned above, each GPU chiplet has its own last-level cache, which is connected to one another in such a way that the coherence of all chips is guaranteed.
When could this new design actually take place? It’s possible AMD is considering MCM technology for next-gen RDNA 3 graphics cards, but that could be optimistic, and maybe later – maybe RDNA 4 – would be a more likely prospect.
Of course, all of this is so much guesswork at this point, and we can’t read too much into a single patent anyway. After all, such design concepts are often exploratory or experimental in nature.
However, it shows the direction AMD is looking to travel in, or at least seriously considering, while also shedding light on possible solutions to the main drawbacks traditional monolithic designs face. As you move forward, this type of graphics card could become increasingly difficult to manufacture while keeping yields at a good enough level (or in other words, keeping costs down).
Rumor has it that AMD isn’t the only company thinking the way you’d expect, with Nvidia exploring the use of MCM itself for hopper graphics cards, and Intel with Xe HP Arctic Sound. Keep in mind that Intel is expected to compete with Nvidia and AMD in the heavyweight division this year with the launch of its Xe-HPG card.
Source link : https://www.techradar.com/news/amd-could-have-a-cunning-plan-to-cut-the-cost-of-its-future-graphics-cards/