News for hackers. Stuff that matters.
Re:Ryzen Z1's Tiny iGPU by ben-schaaf on Monday February 26, 2024 @10:46PM (#39519931)
Timings are much less important for GPUs than bandwidth; the cores are slow and doing the same work on similar memory. A 3090 has ~270ns of memory latency - <70ns is normal for a CPU. See https://chipsandcheese.com/2021/05/13/gpu-memory-latencys-im...
Re:Ryzen Z1's Tiny iGPU by jorvi on Tuesday February 27, 2024 @02:02AM (#39520965)
Maybe I’m misunderstanding the article, but there’s multiple places where they mention memory latency being the main bottleneck. At any rate, thanks for sharing! Interesting stuff.
Re:Ryzen Z1's Tiny iGPU by lynguist on Tuesday February 27, 2024 @04:00AM (#39521786)
No, it’s really bandwidth. Big GPUs have on the order of 800 GB/s memory bandwidth and big CPUs only 80 GB/s! Except for Apple. They really need to up the bandwidth.

GPU performance is proportional to both bandwidth and compute.