For example, GPU architecture, or ISA extensions which might accelerate certain algorithms. Such questions might fall between the cracks of theory - not abstract enough - and practice - because it is not applicable except to a very few. Those few are actual hardware designers who are not likely to be publicly hinting what their next generation architectures will look like, or students of architecture.

I am personally interested because of my background in graphics drivers and related performance issues, but, from some of what I've seen so far, this is not of interest to many on the SE: Computer Graphics site.

  • 2
    $\begingroup$ I wouldn't presume that GPU architecture is interesting for hardware designers only. The better you know your system, the more you can tailor your software to it. $\endgroup$ – David Kuri Jan 28 '16 at 7:21
  • 1
    $\begingroup$ Some feedback I've seen on my own and other's questions about theoretical limits to HW performance questions is basically "run benchmarks because it's too complicated". I certainly see both sides, but these answers have gotten enough upvotes that I wanted to clarify before I start asking more questions oriented to getting myself up to speed on GPU architecture. $\endgroup$ – Daniel M Gessel Jan 28 '16 at 14:01
  • $\begingroup$ Okay, I see your point, thanks for clarifying. My opinion is that it's reasonable to ask for details about hardware architecture. A question that can't be answered, e.g. because interplay between hardware pipeline and algorithm steps are too complicated, is not a good one. $\endgroup$ – David Kuri Jan 28 '16 at 15:02
  • 1
    $\begingroup$ For me, the difference is "how fast will this HW go" vs "are there clear architectural limits". The latter often has answers that are interesting (to me). For dealing with the former, I would be interested in seeing an open benchmark suite. $\endgroup$ – Daniel M Gessel Jan 29 '16 at 16:23

You must log in to answer this question.

Browse other questions tagged .