Announcement

Collapse
No announcement yet.

AMD To Unveil Next-Gen Server Processors On 10 November

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

  • #11
    Originally posted by orzel View Post

    But (am I naive?), that all depends on actual use, doesn't it ? I don't see the point of limiting tb/ht. That would have made sense before throttling and power management. But today ?

    I mean if you don't do much, the cpu will not be warm/noisy/high consuming. But if you require more power, you'd be happy to have the cpu ready to provide it, and not limited.

    I dont disable tb/ht, and most of the time my cpu is completely silent. Or rather "under the (low) noise of the hard disk". (my gpu is fanless)
    There are some potential exploits via the Spectre route for some HT-type enabled processors. If you're ultra paranoid you can disable it at the sacrifice of multithread performance. How much performance you lose depends on the CPU, application, etc. I doubt you'd get much power efficiency gain doing so versus not over-clocking, using the proper power management profile, etc., but I'd certainly bow to hard numbers that show otherwise.

    I've yet to see any in-the-wild (actively used malware) that exploits HT speculative execution vulnerabilities, however, because there's so much low hanging fruit out there already that targeting specific CPU model groups is going to be hit or miss. It's far more easy to target OS vulnerabilities, social engineering, and especially misconfigurations where you're practically guaranteed penetration.

    That said, usually it's just a matter of picking the right power management profile to get reasonable performance v. energy efficiency & thermal management. It's pretty extreme to actively disable certain functions entirely.

    Comment


    • #12
      Originally posted by coder View Post
      I think you'd do better by dialing back the power limits than clock speed, if your BIOS supports it. Recent Intel BIOS typically lets you adjust PL1 and PL2 (which is the "turbo" limit). Not sure how much control AMD gives you over the equivalents.
      Absolutely, its PPT (Package Power Tracking) to limit the overall W drawn by the CPU. I've set mine to 71W for my 5900X, and for normal browsing or light / medium gaming there is no noticeable impact on clock speeds. Only when running something like Blender, which I use my GPUs for anyhow does it have an impact.

      Comment


      • #13
        At 96 cores, they will exceed the core count of every Xeon Phi that was released (but not thread count yet, even with Bergamo's 256). Then maybe up to 256 cores, 512 threads with Turin.

        There will probably be at least another 3 major nodes after TSMC N3, so maybe we'll see 1024 cores before everything is forced to go full 3D. On a 'C' variant if not regular cores.

        Comment


        • #14
          Originally posted by torsionbar28 View Post
          First thing I do on a new machine nowadays, is disable Turbo Boost, and disable Hyperthreading. Machine runs very cool and quiet now. I'm perfectly happy with a 3.4 Ghz base clock. I have no need for a 5 Ghz boost that heats the room and makes my pc sound like a leaf blower.
          Why not just cap your fan RPM and undervolt the CPU? That way, you'll still get [short] bursts of high performance without the noise and heat.

          Comment


          • #15
            Originally posted by fitzie View Post
            amd should have one announcement meeting a year. server cpus, desktop cpus, mobile cpus, desktop gpu and datacenter gpus. what is the point on dragging this all out? they have way too many pr people giving themselves busy work.
            That would be very difficult on reviewers. Many reviewers review most or all of those product lines, like this site or AnandTech.

            Comment


            • #16
              Originally posted by Mark Rose View Post

              That would be very difficult on reviewers. Many reviewers review most or all of those product lines, like this site or AnandTech.
              Originally posted by r1348 View Post

              They're different products released at different times.


              I'm talking about announcements. Not release dates. They can still release products separately. I feel like I'm in groundhog day reliving zen 4 announcements, and we still have zen4 apus and 3d zen4 to go.

              Comment


              • #17
                Originally posted by jaxa View Post
                At 96 cores, they will exceed the core count of every Xeon Phi that was released (but not thread count yet, even with Bergamo's 256).
                And didn't KNL have 2x AVX-512 FMA per core? I think Zen4 has only the equivalent of 1 FMA per core, though it can also do 512-bits add/cycle, plus 512-bits of load or store. Not to mention much higher clock speeds and higher memory bandwidth than even KNL's MCDRAM.

                Originally posted by jaxa View Post
                There will probably be at least another 3 major nodes after TSMC N3, so maybe we'll see 1024 cores before everything is forced to go full 3D. On a 'C' variant if not regular cores.
                With these crazy-high core counts and CXL.mem for expandability, I wonder if they might even drop support for dual-CPU, at some point. I think one of the main use cases for multi-CPU is just to scale up memory capacity.

                Comment


                • #18
                  Originally posted by schmidtbag View Post
                  Why not just cap your fan RPM and undervolt the CPU? That way, you'll still get [short] bursts of high performance without the noise and heat.
                  Theoretically, power management could achieve better efficiency, if it knew the envelope it's working within. You can do things like not spin up turbo as high, and thereby sustain it for enough additional time to get more work done. In practice, they might indeed behave similarly.

                  Comment


                  • #19
                    Originally posted by fitzie View Post
                    I'm talking about announcements. Not release dates. They can still release products separately. I feel like I'm in groundhog day reliving zen 4 announcements, and we still have zen4 apus and 3d zen4 to go.
                    Okay, so just don't read them after you already think you know enough. Problem solved.

                    I'm like that with rumor articles. I don't read every one that comes along, or pay any particular one much heed.

                    These companies are going to announce stuff on their own schedule, and also to mess with their competitors. That's just how it is. Luckily, there will not be a test, so just ignore it if you're not interested!

                    Comment


                    • #20
                      Originally posted by torsionbar28 View Post
                      First thing I do on a new machine nowadays, is disable Turbo Boost, and disable Hyperthreading. Machine runs very cool and quiet now. I'm perfectly happy with a 3.4 Ghz base clock. I have no need for a 5 Ghz boost that heats the room and makes my pc sound like a leaf blower.
                      A better solution (for the users) is to have another cpu model that is not on steroids. Something like ryzen 5700x was to 5800x or intel 12700 to 12700k. Cheaper, a tad slower, and sips less power. Maybe some other kind of optimization is possible but I doubt it. I am not sure what is there for the producer, they can maybe target another market segment, or maybe just sell lower quality silicon in this way.

                      Comment

                      Working...
                      X