Announcement

Collapse
No announcement yet.

Intel Pursuing AVX-512 Optimized Crypto Algorithms For The Linux Kernel

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

  • Intel Pursuing AVX-512 Optimized Crypto Algorithms For The Linux Kernel

    Phoronix: Intel Pursuing AVX-512 Optimized Crypto Algorithms For The Linux Kernel

    Intel engineers have posted the initial Linux kernel patches providing AVX-512 optimized versions of common crypto algorithms. The AVX-512 optimized versions do pan out and promise to offer huge speed-ups but are disabled by default at this stage over the negative CPU frequency/performance impact that running AVX-512 can have on CPU cores / shared threads...

    http://www.phoronix.com/scan.php?pag...-Crypto-Kernel

  • #2
    I remember Linus complaining about Intel's AVX-512 stuff. Are these patches related to that?

    Comment


    • #3
      Originally posted by [email protected] View Post
      I remember Linus complaining about Intel's AVX-512 stuff. Are these patches related to that?
      Nope, this is simply implementing what already is in Crypto on the kernel that stops at AVX2.

      Linus and basically everyone else simply hate AVX-512 because Intel went full retard on its design and is a f****ing mess but that is unrelated to this

      Comment


      • #4
        Would avx512 be also beneficial to raid xor functions? Or are they already fast enough for normal use cases?

        I'm wondering how the industry will start handling raif scenarios ... flash instead of disks ...

        Comment


        • #5
          Originally posted by pegasus View Post
          Would avx512 be also beneficial to raid xor functions? Or are they already fast enough for normal use cases?

          I'm wondering how the industry will start handling raid scenarios ... flash instead of disks ...
          I think they have already moved to distributed object storage platforms like Ceph?

          Comment


          • #6
            Even if distributed, you still have to calculate parity or erasure coding somewhere. I remember intel optimized libs for ceph for this, even saw gpu offload for erasure coding So there must be some value in having nicely tuned xor somewhere in the kernel ...

            Comment


            • #7
              Sounds like a bad idea in the kernel. AVX-512 is only good for intensive concentrated work taking up all the CPU time for 10s of milliseconds.

              Comment


              • #8
                Originally posted by carewolf View Post
                Sounds like a bad idea in the kernel. AVX-512 is only good for intensive concentrated work taking up all the CPU time for 10s of milliseconds.
                Thats why you benchmark to verify if this is true or not.

                I do think however it makes sense to blacklist this feature for the initial set of Intel CPU's which features AVX512, because on those CPU's it really was a power virus that slowed down the other cores. iirc the later CPU's resolved this issue?

                Comment


                • #9
                  Originally posted by mdedetrich View Post

                  Thats why you benchmark to verify if this is true or not.

                  I do think however it makes sense to blacklist this feature for the initial set of Intel CPU's which features AVX512, because on those CPU's it really was a power virus that slowed down the other cores. iirc the later CPU's resolved this issue?
                  it will probably perform better in dedicated benchmarks. The problem is the negative effects on everything else.

                  Comment


                  • #10
                    Strange side channel attack leaking the keys used, but only on intel cpus, discovered in 3 2 ...

                    Comment

                    Working...
                    X