Join the discussion

Write your take first — we'll ask for email only when you're ready to publish.

  • Hacker News
  • surely someone in apple would have run some common neural net architectures through the rtl simulator before tapeout and checked that the performance meets expectations??
  • I guess any processes affected by this will solve this in software as time goes on?
  • This goes way over my head and I don't understand most of it lol. I noticed you're still in the middle of getting your B.S. degree and you're already writing things like this...amazing.
  • This is kinda nuts.

    Can this mostly be fixed in software, or are M3 users basically stuck unless apps work around it themselves?

    Going from 10 to 24 tok/s just from a workaround is a pretty crazy jump.

  • Just checking here- this systemverilog is a hypothetical telling of what you think is going on? Or do you have the actual source of the RTL?
  • Not all systems affected M1 and M5MAX are OK

    https://x.com/anemll/status/2098454204478366132?s=20

  • "Apparently the memory controller's throughput has a dominant harmonic with wavelength 2048 in tensor-dimension space."

    That got a laugh out of me.

  • Nice investigation.

    It is always surprising to me when a nice round number like 1MiB results in the “bad performance” configuration (although it happens).

    Are you sure erratum is the right word in this context? I usually see it used to describe the notice that a document has an error in it.

Explore Birbla archives