AMD Ryzen 9000X3D rumored to reverse the 3D stacking hierarchy — the L3 cache block allegedly sits below the CCDs

News
By
published

So that's where the higher clock speeds and better thermals stem from.

Ryzen CPU with 3D V-Cache
(Image credit: AMD)

In two weeks, AMD will introduce its Ryzen 9000X3D chips, which will rival the best CPUs on the market. Hardware leaker HXL on X has clarified, to some extent, the architectural differences between the first and second generations and why AMD can achieve better thermals and higher clock speeds with Ryzen 9000X3D. It is claimed that AMD is reversing its 3D die-stacking structure, with the SRAM block now placed beneath the CCDs.

When AMD first launched its Ryzen 5000X3D chips, the concept was to stack a block of L3 cache atop the CCDs. A large issue with this model was heat dissipation due to using extra silicon over the CCDs for structural integrity. This thermal constraint also restricted AMD from bumping the clock speeds on these processors. However, that seems to be addressed with Ryzen 9000X3D, as the CCD reportedly sits atop the cache die. If true, AMD's new design will genuinely be an engineering marvel - from both architectural and packaging standpoints.

Currently, we have no diagrams of these CPUs and are unsure how AMD will manage the potential bump in the y-axis compared to Ryzen 9000 non-X3D. However, such a change was expected since Zen 5-based CCDs have a considerably smaller internal L3 block size than Zen 4. This meant that any potential external dies (through 3D stacking) would overlap with the CCDs and cause heat dissipation problems. High Yield's chip analysis indicated that a plausible solution was to use 2-Hi stacking of 3D cache, but AMD has surprised us all with a completely different approach.

Image 1 of 2
Zen 5 relative L3 cache size
(Image credit: High Yield)

It will be intriguing to see how AMD accomplishes such a layout, and we hope Team Red blesses us with an in-depth overview. Based on this leak, Ryzen 9000X3D chips will have the CCD in direct contact with the IHS without any extra silicon for improved thermal performance.

This spells trouble for Intel since Core Ultra 9 285K, the flagship Arrow Lake chip, fails to beat even last generation's Ryzen 7 7800X3D. The Ryzen 9000X3D series will debut on November 7, and retailer listings have put the Ryzen 7 9800X3D in the ballpark of $484-$525.

See more CPUs News
Hassam Nasir
Hassam Nasir
Contributing Writer

Hassam Nasir is a die-hard hardware enthusiast with years of experience as a tech editor and writer, focusing on detailed CPU comparisons and general hardware news. When he’s not working, you’ll find him bending tubes for his ever-evolving custom water-loop gaming rig or benchmarking the latest CPUs and GPUs just for fun.

27 Comments Comment from the forums
  • helper800
    I cannot wait until they get CCD density improvements at some point. Greater than 8 core CCDs will be when I jump back in on an X3D part coming from the 5800X3D.
    Reply
  • Alvar "Miles" Udell
    This spells trouble for Intel since Core Ultra 9 285K, the flagship Arrow Lake chip, fails to beat even last generation's Ryzen 7 7800X3D.

    In games, it's nearly 30% faster in non gaming applications than said 7800X3D.

    Reply
  • DaveJ
    What I would really like is an AMD 9800G-X3D that would be a great chip to own!
    Reply
  • endocine
    Alvar Miles Udell said:
    In games, it's nearly 30% faster in non gaming applications than said 7800X3D.
    That probably has something to do with all the ecores, that chart, when compared to something with equivalent core counts, like a zen5 7950 and 9950 is about the same, where the 285K has 24 real cores, and those AMD parts running at the same perf in the chart only have 16(32 with SMT) and are on an older TSMC node. If you are not gaming and want productivity, why buy the intel part when you can have homogeneous cores with AVX512, and when you are gaming, will still beat the 285K, and use less power doing it, on a platform with an upgrade path that is also more stable.
    Reply
  • usertests
    helper800 said:
    I cannot wait until they get CCD density improvements at some point. Greater than 8 core CCDs will be when I jump back in on an X3D part coming from the 5800X3D.
    Maybe with Zen 6, or maybe AMD will take forever to increase cores for yield reasons.

    I think what you really want is more than 8 cores in a CCX (rather than something like a 16-core Zen 5c chiplet with two CCXs).

    DaveJ said:
    What I would really like is an AMD 9800G-X3D that would be a great chip to own!
    Best you can hope for is Strix Halo (soldered) with added 3D V-Cache (not just the confirmed 32 MiB MALL cache, AKA Infinity Cache for the iGPU).
    Reply
  • Guardians Bane
    DaveJ said:
    What I would really like is an AMD 9800G-X3D that would be a great chip to own!
    With a full Halo version of a 40cu iGPU... If that would actually work, it def would be an awesome APU!!! Imagine a mini PC with that inside. I'm into minis and am excited for the Halo versions. Pricey most likely, but there gonna be awesome!
    Reply
  • abufrejoval
    DaveJ said:
    What I would really like is an AMD 9800G-X3D that would be a great chip to own!
    I'm afraid you'd just create problems faster than you can solve them...

    First of all: aggregating functionality and performance means aggregating heat, which is less troublesome to eliminate with more surface area on spread out dies.

    In a thin and light notebook you accept that and compromise on performance, but if you're aiming for better gaming muscle, the returns have a hard time justifying that effort.

    And SRAM simply uses too much real-estate with 6-8 transistors per bit of cache to completely replace a frame buffer, which means you'll still need something with sufficient sustainable bandwidth and capacity.

    That can be multi-channel DRAM on a shared die carrier à la Mx or Lunar Lake, it could be HBM or GDDRx, but it will also need space and cooling and there is no chance it will fit into an AM5 socket.

    You'd need to question your motives: why do you want an APU?

    If it's about not needing a dGPU, it's all the troubles I mentioned above.

    If it's about being able to transition from a very low power environment up to great gaming with a dGPU-on-demand, you'd be paying a lot extra for iGPU transistors that won't get used in gaming.

    Again on a notebook, people pay extra for that flexibilitiy.

    On a desktop there will still be some interested, myself included (because it runs 24x7), but AMD is financially so much more healthy than Intel, becasue they won't do bespoke small niche designs: if it doesn't have scale, they won't go for it.
    Reply
  • TheHerald
    endocine said:
    and when you are gaming, will still beat the 285K, and use less power doing it, on a platform with an upgrade path that is also more stable.
    Uhm, yeah, only that did not happen. The 9950x used 40% more power for identical gaming performance on Tom's hardware 14 game test.
    Reply
  • abufrejoval
    helper800 said:
    I cannot wait until they get CCD density improvements at some point. Greater than 8 core CCDs will be when I jump back in on an X3D part coming from the 5800X3D.
    Well, they could do a "desktop" part using Zen 5c Turin dense CCDs, which have 16 cores each, but of course they won't clock as high as gamers might want.

    Now those most likely won't have TSV dock points allocated for V-Cache, so they won't match what you want.

    But in your case, going with a dual CCD SoC seems like the much easier solution than hoping for performance and density.

    In a way AMD is very clearly demonstrating that you simply can't have both, but that they offer you a choice.

    The 5800X3D is a very good chip, so perhaps you should concentrate on if CPU performance is actually limiting you in any way.

    My impression is that reviewers are on a gaming performance craze comparing games at THD medium settings and triple digit FPS, while the actual games have long ceased to be CPU bound at all.

    So you might be better off replacing your GPU or just enjoying your games.
    Reply
  • usertests
    abufrejoval said:
    But in your case, going with a dual CCD SoC seems like the much easier solution than hoping for performance and density.

    The 5800X3D is a very good chip, so perhaps you should concentrate on if CPU performance is actually limiting you in any way.

    So you might be better off replacing your GPU or just enjoying your games.
    At some point in the future, AMD will move up core counts for each tier, e.g. giving 24-32 cores instead of 16, 10-12 cores instead of 8, and perhaps eliminating the 6-core. One way they could do that is to increase the number of cores in a standard/non-dense CCD. If they do, the cheaper CPUs with only one CCD will get more cores. If the CCX also increases beyond 8 cores, that keeps it simple and prevents a potential source of latency. That's assuming it's technically appealing for AMD to do that. If not, then a 16-core, dual-CCX die like Zen 4c/5c could be the easier option.

    @helper800 is obviously in no hurry to upgrade, isn't getting a 9800X3D, and may end up skipping AM5 entirely. By the time core counts increase (maybe Zen 6 or Zen 7), there could be decent reasons to upgrade from the 5800X3D. You seem to have a bad habit of telling people what they can't wish for.
    Reply