• mateoboudoir@alien.topB
    link
    fedilink
    English
    arrow-up
    1
    ·
    10 months ago

    It’s always neat to see occasional IGP vs. dGPU comparisons like this, if for nothing else then as a sort of barometer of IGP progress.

    As always, the numbers indicate that the major Achilles’ heel of IGPs remains memory bandwidth. The 780M even seems like it could perform a tier better than implied here - in the ballpark of the legendary GTX 1060/RX 480 GPUs - if only it could fetch more data more quickly. The revelation isn’t exactly surprising, as there already exists basically a 780M in dGPU form in the RX 6400 (also 12 CUs, just paired with GDDR6 memory) that performs right in that range.

    • stilljustacatinacage@alien.topB
      link
      fedilink
      English
      arrow-up
      1
      ·
      10 months ago

      Hmm. This is the first time it’s occurred to me, but I wonder what sorts of performance we might see from iGPUs once Direct Storage is more mature. I know pulling texture data from SSD could never be as fast as from RAM - but I wonder if reducing SSD -> RAM -> GPU to just SSD -> GPU will have meaningful savings.