Announcement

Collapse
No announcement yet.

Intel Core X-Series Lifts, Linux Benchmarks Forthcoming

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

  • #21
    Originally posted by jrch2k8 View Post
    Plus i have a feeling the cache latency in TR will be a whole lot lower with quad channel because i seriously believe those CCXs were meant that way from the beginning and were forced into dual channel high latency to keep ryzen good enough at a spine break price to push intel into panic + lower platform costs specially for B350 boards. i also hope Vega brings a worthy contender to the 1070+ too, so we can have better prices overall
    I think TR will actually have worse latency, but the "infinity fabric" will run more efficiently. Keep in mind it is basically just two Ryzen 7s slapped together, which also includes doubling the CCXs.

    It's really no different than anything else in computers. Take RAM for example: when you have too little, things are slow because it has to clear and re-write data. You can keep adding more RAM and as long as it's being properly utilized, performance will remain good. But there is such thing as "too much", where you start to lose performance since the system takes longer to fetch the data from the correct location. This is why caches on the CPU get smaller as you go lower.

    Bigger isn't always better. Get what you need, not what you want to brag about.

    Comment


    • #22
      Originally posted by karasu View Post
      "Up to..."
      LOLOLOLOLOLOLOLOL Raid keys for NVMe raid beyond RAID0. This is fucking ridicolous. (for windows users anyway, Linux can RAID fine probably as it does not need silly "HW-ish raid" features from the board)

      Comment


      • #23
        Originally posted by schmidtbag View Post
        I think TR will actually have worse latency, but the "infinity fabric" will run more efficiently. Keep in mind it is basically just two Ryzen 7s slapped together, which also includes doubling the CCXs.

        It's really no different than anything else in computers. Take RAM for example: when you have too little, things are slow because it has to clear and re-write data. You can keep adding more RAM and as long as it's being properly utilized, performance will remain good. But there is such thing as "too much", where you start to lose performance since the system takes longer to fetch the data from the correct location. This is why caches on the CPU get smaller as you go lower.

        Bigger isn't always better. Get what you need, not what you want to brag about.
        yeah, totally correct.

        But in this case it will depend how is the quad channel is actually used in the infinity fabric, if:

        1.) as on ryzen both CCX having a deathmatch to map the ram available bandwidth at ram speed(cheaper but high latency)
        2.) the infinity fabric distribute at double rate the bandwidth to each CCX(expensier and prolly requires a better binned silicon but should improve latency)
        3.) each CCX get a dedicated channel bypassing the fabric for latency
        4.) other method AMD could use.

        My point is Ryzen is very heavily constrained to be cheap enough for consumer market while TR/Epyc have more cost room for efficiency, so i think is more natural to expect improved latency doing a smarter use of the fabric since you have enough cost room and is targeted for a higher tier market.

        Sure if they just brute force slapped togheter 2 unmodified ryzen 7 through a fabric then you are probably 100% right here of course

        Comment


        • #24
          Michael, can I request a test of the x264 H.264/AVC encoder run on the CPU (OpenCL support disabled)? Interested to see how it scales with the number of cores.
          OpenBenchmarking.org, Phoronix Test Suite, Phoronix Global, Linux benchmarking, automated benchmarking, benchmarking results, benchmarking repository, test profiles

          Comment


          • #25
            As others have pointed out, Threadripper is two cpu's stitched together. Predictable, but don't expect much. As for PSU, it's not a linux issue, so reviews are not required, unless some company sends you samples.

            I have a request of Apache benchmark, and nginx. But it's not really a request, just a "it would be nice". And some BTRFS stuff.

            Comment


            • #26
              Throw everything you got at it!
              Hi

              Comment


              • #27
                Originally posted by devius View Post

                A very small step... only 100MHz more on the base frequency, but that is offset by the increase of 21W in the TDP, so maybe it isn't a step at all?
                You'll easily notice the step though. On your electricity bill...

                Comment


                • #28
                  Originally posted by devius View Post

                  A very small step... only 100MHz more on the base frequency, but that is offset by the increase of 21W in the TDP, so maybe it isn't a step at all?
                  Probably all 7700k users will overclock at least 100 MHz. Why else would you buy a K chip?

                  Comment


                  • #29
                    Originally posted by MuPuF View Post
                    Very nice! Looking forward to the Timed Linux compilation results What else is a CPU used for anyway?
                    Video encoding, AI, and other HPC stuff. It's interesting to see if the ordinary apps for lamers begin to support multiple cores now. I can give tons of examples of everyday Linux apps that don't even support the simplest form of threading.

                    A good example is publishing business. My friend uses Inkscape, Scribus, LaTeX, and some other stuff. For example when he downloads a set of assets (e.g. 40 GB), most of the programs in the pipeline only use one core. Even generating thumbnails of images takes ages. Uses only one core.. What he does is, comes up with trivial workarounds: thumbnails - open multiple file managers and scroll down to a different part of the folder to start another generator, image format transformers - GNU parallel or GNU make, PDFs - generate multiple single page documents and combine with pdftk, slow image filters in inkscape/gimp - start multiple instances.

                    Transition to a multicore mindset in apps probably won't take long, but many projects are stubborn and refuse to convert. Currently, to an ordinary user, a 5 GHz single-core is faster than a 4.5 Ghz 1024-core with 100% perfect linear scalability. The apps just aren't there yet. Case in point: in Kdenlive, go to settings -> environment. Take a look at 'processing threads'. The description says '> 1 is experimental'. Geez..
                    Last edited by caligula; 20 June 2017, 08:43 PM.

                    Comment

                    Working...
                    X