Announcement

Collapse
No announcement yet.

Intel Dramatically Speeds Up NSS With AVX2

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

  • #11
    Anecdote of when NSS is a bottleneck: trying to start FF in a fresh VM. There is not enough entropy, NSS insists on /dev/random instead of /dev/urandom, and so FF start blocks until NSS has enough random data.

    Comment


    • #12
      Originally posted by bastiaan View Post
      Although I agree it is a good thing that Intel is participating in this project, I'm also a little skeptical. If you have a modern Haswell CPU, is NSS processing really going to be a noticable bottleneck for your average browsing session? I don't think so. Perhaps we are indeed looking at a technology demo with little impact.
      Probably useful server side.

      Comment


      • #13
        Originally posted by RealNC View Post
        What happened to AVX1? No support?
        well the first release of AVX is FP operations using 256 wide instructions and AVX2 handle the integer operations using 256 wide instructions, so assuming the algorithm is integer only it would be trivial for AVX2 but not so much for AVX since you need an additional step and cast the operands as floats/doubles before move the vector to the closest cache[L1/L2] to later recast the output vector back to integer.

        now if the algorithm can operate with FP then it can use AVX just fine.

        abut FMA4/3 im not entirely sure either can handle integer vectors.

        so as a thumb rule AVX1 Floating Point only and AVX2 Integer only

        Comment


        • #14
          Originally posted by erendorn View Post
          Probably useful server side.
          That's what my first thought was, since that's an environment where you would be doing a lot of signing and verifying.

          Comment


          • #15
            Originally posted by bastiaan View Post
            I doubt it, except under special circumstances. NSS is fast enough to run smoothly on low power ARM chips (as part of Firefox mobile, for example), and a Haswell Atom will surely be at least on par with the performance of the fastest ARM chip.
            Apparently the new Atoms won't be based on Haswell, but on 'Silvermont', which won't have AVX2. So it is a moot point.

            Comment


            • #16
              Originally posted by curaga View Post
              Anecdote of when NSS is a bottleneck: trying to start FF in a fresh VM. There is not enough entropy, NSS insists on /dev/random instead of /dev/urandom, and so FF start blocks until NSS has enough random data.
              last time i checked many distros had random as a symlink to urandom
              slackware never had

              Comment


              • #17
                Originally posted by gens View Post
                last time i checked many distros had random as a symlink to urandom
                slackware never had
                how about using frandom as an alternative ?

                https://wiki.archlinux.org/index.php/Frandom

                would that help ?


                this is broadly used in custom kernels & ROMs on android smartphones over at xda-forums

                Comment

                Working...
                X