236
Valve 'Fremont' APU breaks cover on Geekbench, most likely for a future console (www.club386.com)
submitted 1 year ago by to c/steamdeck@sopuli.xyz
36 comments fedilink hide all child comments
 

New steam controller expected to release alongside it. This also should push back estimates of when to expect a Steam Deck 2, since valve is probably putting most of their hardware efforts towards this instead.

you are viewing a single comment's thread
view the rest of the comments
[–] 8 points 1 year ago* (4 children)

apple silicon has specific changes to amd64, and some extra instructions to help with x86 translation afaik (1 of the main things i think is how memory is laid out is different somehow? so every memory access needs extra clock cycles to accomplish in standard arm64)

but in general, rosetta2 is incredible: very little performance loss… i thiiink it’s something like 2-10%, which is made up for by the cheaper cores

from what i understand, graphics is more of a problem… metal just isn’t well supported - i think the standard thing is you have to do direct x <-> dxvk <-> vulkan <-> moltenvk <-> metal, which is a huge overhead and is far from perfect in its translations

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 4 child comments
    •
  • [–] 7 points 1 year ago* (1 child)

    1 of the main things i think is how memory is laid out is different somehow? so every memory access needs extra clock cycles to accomplish in standard arm64

    It's down to "memory ordering", as different cores interact with RAM there's rules that govern how those cores see changes made by other cores. ARM systems are "weak", so rely on developers to be explicit about the sharing, while x86's "Total Store Order" is considered "strong" and relies on the hardware to disentangle it all so software can make assumptions and play fast and loose.

    You can do software emulation of strong memory ordering on a weak system, but it's slow. What Apple did was provide a hardware implementation of strong ordering in their ARM chips, and Rosetta enables that when running x86 code, so users don't encounter that slowdown.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 1 child comment
    •