Announcement

Collapse
No announcement yet.

Using AVX2 With Android's Bionic Library Can Yield Much Better Chromebook Performance

Collapse
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • #31
    Originally posted by carewolf View Post
    Because what he said made no sense, so I tried to parse it into something that did.

    Besides using AVX(1) as just enhanced 128bit SSE4 was how I used it the most until AVX2 became widespread.
    Sorry, I missed notification on this thread.

    Yes coder I was talking about integer arithmetic with vectors, which is also what this thread is probably about. But I will admit that I did not know that it increased registers to 256-bit for floats (not going to excuse myself here, I admit I was half wrong!), I simply never used that, just like carewolf.

    Comment


    • #32
      Originally posted by starshipeleven View Post
      Are you sure? Afaik the Atoms are still a separate name within the main arch name for low end chips https://en.wikichip.org/wiki/intel/m...tures/goldmont
      Atoms in Goldmont were Denverton, and Penitum/Celeron were Apollo Lake, for example.
      1. Click the link I posted.
      2. Note that it's part of the "Gemini Lake" product line.
      3. Look up "Gemini Lake", noting that it's built around "Goldmont+" Atom-derived cores.
      Intel has gone a weird direction, with Atom branding. They seem to have used it for certain market niches, but their Atom-derived cores are used more broadly.

      Let's take a Celeron example. The ODROID-H2 has a Celeron J4105:

      https://www.hardkernel.com/shop/odroid-h2/
      https://ark.intel.com/products/12898...p-to-2-50-GHz-

      Goldmont+ is actually faster than Core 2, and some benchmarks suggest it's clock-for-clock competitive with early Core i-series laptop CPUs.
      Last edited by coder; 02-11-2019, 09:36 PM.

      Comment


      • #33
        Originally posted by starshipeleven View Post
        Even if AMD offered amazing AVX2 on low end hardware, the software still could not assume it was there as most low end hardware is Intel and does not support it.
        https://en.wikipedia.org/wiki/CPUID

        Comment


        • #34
          Originally posted by Weasel View Post
          I will admit that I did not know that it increased registers to 256-bit for floats (not going to excuse myself here, I admit I was half wrong!), I simply never used that, just like carewolf.
          Thank you.

          Comment


          • #35
            Originally posted by coder View Post
            I meant that it makes little sense invest in feature that is not present in low end hardware if you are mostly targeting low end hardware.

            We were talking of adoption speed of new features. If most of the hardware you are targeting does not support feature X you are not likely to invest in it

            Comment

            Working...
            X