Comments Locked

15 Comments

Back to Article

  • Dolda2000 - Sunday, June 2, 2024 - link

    >6:30pm PT/02:30 UTC
    Surely, 18:30 PDT is 01:30 UTC?
  • Ryan Smith - Sunday, June 2, 2024 - link

    Yes, it is. This is what I get for copying the text from the last time AMD had a keynote at 6:30pm PT (which was in the winter)
  • kpb321 - Sunday, June 2, 2024 - link

    That Ryzen AI 9 HX 370 die shot in the slide seems to have two types of cores? The top 8 cores definitely look different than the bottom 4 cores. I assume this means something like Ryzen 5 and 5c cores both showing up. I'd think 4 regular and 8 compact cores would make sense but the 4 bottom cores seem a lot more densely packed so might be the other way round.
  • Dolda2000 - Sunday, June 2, 2024 - link

    I was also surprised by the dieshot. If anything, I think the two core types seem equally densely packed, so I don't really get it.

    Almost more importantly though, I can't believe they actually went with the stupid rumored new naming scheme. :P
  • Dolda2000 - Sunday, June 2, 2024 - link

    The images of 192C Turin seems to reinforce this. The 16C chiplets seems to be huge compared to the Zen 4c chiplets on Bergamo. Weird.
  • kpb321 - Sunday, June 2, 2024 - link

    Yeah. I don't know. If you include the rectangles above and below the bottom 4 cores they seem to take the same overall space as the top cores which doesn't make sense for 5 and 5c cores. Maybe those are not actually part of the lower cores and just the square is the entire 5c core compared to the square and bigger rectangle being the the regular Ryzen 5 cores. Then what are the rectangles around the lower cores if not part of the cores? What ever is happening, it's pretty obvious that it isn't 12 identical layouts.

    The top 8 cores seem to be mirrored around what ever the bar is between them and the bottom 4 cores seem to have 2 of those bars. If that's cache would certainly point to the bottom 4 core being higher performance with a lot more cache per core.
  • nandnandnand - Monday, June 3, 2024 - link

    You can find a better image in the slide deck. It looks like a render rather than an actual die shot, and the widths align perfectly. Maybe they did not draw these to scale and have confused us as a result.

    It's unfathomable that Zen 5 and Zen 5c would be the same size, so I'm calling it a bad artist's rendition. We can't trust anything in the image right now.
  • nandnandnand - Monday, June 3, 2024 - link

    There are 4x Zen 5 and 8x Zen 5c cores.
  • Dolda2000 - Sunday, June 2, 2024 - link

    They announced that Zen 5 is shipping as soon as July, but no prices? Is there another event before then where they're going to announce that?
  • yankeeDDL - Monday, June 3, 2024 - link

    AMD's advantage over Intel in the server space is mind-blowing.
    Regarding AI today NVIDIA is the undisputed king; if the MI350 can really make a dent into a multi-billion-dollar market, it could be a game changer in terms of revenue.
  • CrystalCowboy - Monday, June 3, 2024 - link

    With the growing importance of AI, will the traditional desktop Ryzen be overtaken by the mobile-derived Ryzen AI line? Can traditional CPU and AI chiplets be intermixed? Will there be add-in AI cards for desktops with traditional CPU/APUs? If so, would these be PCIexpress or some other form factor?
  • Dolda2000 - Monday, June 3, 2024 - link

    NPUs are mainly there for power efficiency in mobile form factors. On the desktop, there's little reason to nut just use a GPU instead.
  • nandnandnand - Monday, June 3, 2024 - link

    Intel will add NPUs to desktop CPUs first. And that might make sense, as many millions of Intel desktop CPUs are destined for office PCs with no discrete GPU.
  • Dolda2000 - Sunday, June 9, 2024 - link

    >with no discrete GPU
    I don't necessarily think a discrete GPU will be needed. In many cases, even the iGPU is more powerful for inference workloads than the NPU. It's just less power-efficient for that purpose, but again on desktop that doesn't matter as much.
  • nandnandnand - Monday, June 3, 2024 - link

    There are both M.2 and PCIe AI accelerator add-in cards if you really wanted one. Example:

    https://www.cnx-software.com/2024/05/23/edgecortix...

    Most people would grab a consumer GPU instead.

    If AnandTech's information is correct, AMD is reusing the I/O chiplet from Raphael because it was an easy option, and there might not be any room to add a large NPU. So same RDNA2 graphics, no NPU... for now.

Log in

Don't have an account? Sign up now