Image: AMD

AMD is currently working a next-generation successor to its RDNA 3 architecture called RDNA 4. That isn’t really a surprise, nor is the fact that it’ll likely be built using a smaller and much more advanced node process, but leakers are now claiming to know some of the specifics.

According to leaker Greymon55, who has made a habit out of sharing information on NVIDIA and AMD’s quickly approaching Lovelace and RDNA 3 GPUs as of late, red team will be leveraging 3-nanometer and 5-nanometer processes for its RDNA 4 products.

These are believed to be multi-chip-module designs that do away with the traditional monolithic dies for what are hopefully admirable gains in efficiency and performance. AMD’s RDNA 3 lineup of graphics cards, presumably the Radeon RX 7000 Series, is expected to be the first to include MCM-based models when they launch in 2022.

Red team’s RNDA 4-based Radeon RX 8000 Series is a bit further off and not expected to launch until late 2023. But perhaps the great graphics card drought will have finally resolved by then.

Greymon55 lists two nodes for the RDNA4 series, which obviously refers to the I/O part connecting the chiplets and the graphics tiles themselves. For NAVI4X MCM chips this means 3nm graphics tiles and 5nm I/O dies. Respectively, RDNA3 MCM GPUs will see 5nm and 6nm processes.

Source: Greymon55 (via VideoCardz)

Don’t Miss Out on More FPS Review Content!

Our weekly newsletter includes a recap of our reviews and a run down of the most popular tech news that we published.

Join the Conversation

9 Comments

  1. There’s a bit of shrink addiction in the world of high performance silicon… Is this strictly necessary? Need is the mother of invention… Wonder what ideas could come out if you are stuck with say 22 28nm, or 14nm? Seemingly no attempts since you know you can just shrink it for now. Im just thinking bigger process might actually be better at modifying them in a way that may make them equivalent, but way cheaper.

  2. [QUOTE=”Uvilla, post: 40060, member: 397″]
    There’s a bit of shrink addiction in the world of high performance silicon… Is this strictly necessary? Need is the mother of invention… Wonder what ideas could come out if you are stuck with say 22 28nm, or 14nm? Seemingly no attempts since you know you can just shrink it for now. Im just thinking bigger process might actually be better at modifying them in a way that may make them equivalent, but way cheaper.
    [/QUOTE]
    Pretty sure Intel has beat this horse

  3. [QUOTE=”Brian_B, post: 40069, member: 96″]
    Pretty sure Intel has beat this horse
    [/QUOTE]
    Yeah, just ask Intel how far you can take 14nm. It’s actually pretty far, just not enough to compete with smaller nodes

  4. Well…we will see. The Chinese are in a pickle so they might get some out of the box thinking going… Or be self reliant at small process which is also a heavy lift.

  5. [QUOTE=”Uvilla, post: 40094, member: 397″]
    Well…we will see. The Chinese are in a pickle so they might get some out of the box thinking going… Or be self reliant at small process which is also a heavy lift.
    [/QUOTE]
    I don’t know that China is particularly known for out of the box thinking. Their style is more to wait on someone else to think out of the box, then just copy it and mass produce it.

    They are making a major push to be self reliant when it comes to electronics though, so it does bear watching. They have started their own CPUs and GPUs, but to date none of them have really threatened to take any performance crowns.

  6. [QUOTE=”Brian_B, post: 40095, member: 96″]
    I don’t know that China is particularly known for out of the box thinking. Their style is more to wait on someone else to think out of the box, then just copy it and mass produce it.

    They are making a major push to be self reliant when it comes to electronics though, so it does bear watching. They have started their own CPUs and GPUs, but to date none of them have really threatened to take any performance crowns.
    [/QUOTE]
    They’re trying to do [I]everything[/I], and being the ‘world’s manufacturer’ has provided them with quite a bit of know-how. And their neighbors and largest business partners are the world’s innovators. With those ingredients and a billion citizens, although hampered a bit by lack of natural resources, they’re making a pretty good ‘go’ at it.

    With respect to cutting-edge semiconductors like many cutting-edge technologies for which ‘shoulder-surfing’ can only get them so far, they have a ways to go, but in general they also have strong motivators. Chiefly among them seems to be that they’d like their indigenously-produced products to be desirable both to the world’s middle classes as well as their own.

    I’ll say that in terms of general consumer electronics, it seems that Chinese companies can produce anything that’s needed with indigenous supply, outside of integrated circuits.

  7. [QUOTE=”Uvilla, post: 40060, member: 397″]
    There’s a bit of shrink addiction in the world of high performance silicon… Is this strictly necessary? Need is the mother of invention… Wonder what ideas could come out if you are stuck with say 22 28nm, or 14nm? Seemingly no attempts since you know you can just shrink it for now. Im just thinking bigger process might actually be better at modifying them in a way that may make them equivalent, but way cheaper.
    [/QUOTE]

    You can only optimise so far and a lot of high end chips are already pretty big in die size making them harder to make and smaller yiields which is bad for pricing.

    However there are still a whole lot of chips made on bigger nodes that don’t require state of the art technology to work.

  8. I’ve always said that GPU’s are really where chiplets will shine.

    They are great on CPU’s but provided you can feed them with fast interconnects they should be AMAZING on GPU’s allowing you essentially to connect SP’s/Cores/Whatever units from different chips as if they were on the same die, without any kind of SLI/Crossfire nonsense.

    I’m looking forward to learning more about these.

    Too bad Nvidia nabbed up all of TSMC’s early 3nm supply 🙁

Leave a comment