Új hozzászólás Aktív témák

  • Yutani

    nagyúr

    válasz Szeszkazán #26016 üzenetére

    AMD's biggest problem with GCN is the inability of most game engines fully utilizing it's shaders. We've seen what can happen with good programmers with Doom Vulkan, using Shader Intrinsics and Async Compute to max out GCN, RX 480 ~20% faster than 1060, which actually matches it's paper TFlops specs. In most games for GCN, there's a lot of SIMDs with idling ALUs doing nothing so in effect, it's not a 5.8 TFlop GPU since that rating is the max theoretical perf assuming all ALUs are working.
    Going with NCU's variable width SIMD layout will solve this issue, so we should get a more closer and truer comparison of paper spec TFlops to gaming performance. ie. A Vega of 12.5 TFlops will absolutely wreck Fury X.

    Jobb később megtudni, mint soha. Most már értem az eddig érthetetlen túlfogyasztást az Nvidia GPU-khoz képest.

    #tarcsad

Új hozzászólás Aktív témák