No announcement yet.

Intel Launches The Xeon D 2100, Up To 18 Core SoCs

  • Filter
  • Time
  • Show
Clear All
new posts

  • #31
    RISC-V SoC or SiFive Freedom U540 SoC unleashed - it runs Doom with Radeon card


    • #32
      Originally posted by wizard69 View Post
      You completely missed the point of my post!
      I'm just pointing out that currently the issue is in the applications not multithreading themselves anywhere near enough. The OS and hardware was ready since last decade, it's so widespread that it's practically irrelevant.

      If you are only concerned with running a single application then a modern operating system isnt of any value at all to you.
      Huh? For any high-performance application you very much care. The OS does have all the thread-management infrastructure, and schedulers for CPU time, I/O and so on. If you are on a multicore system you need some kind of OS to do this job for you, or you must do so on your own, and that's not a piece of cake.

      No consider a modern highly threaded app like Eclipse that most certainly benefits from more cores over a high click rate.
      *That runs like shit even on octacores.

      And no, I don't see why an IDE should need any significant amount of processing power. It's a glorified text editor with auxiliary package manager or XML/similar parser functionality.

      The OS works in conjunction with Java too spread the processes and threads over many cores to give the user good performance.
      *Java duplicates a bunch of OS functions for the sake of making a cross-OS application, but by itself does not multi-thread your program.

      It may do something very rudimentary, yes, if you use pre-made libs/frameworks that are multi-threaded, but you can use libraries/frameworks on any language.

      Multi-threading happens when the programmer lays out a project where his program splits up itself in multiple threads where each does stuff on its own. This division must happen at the logic level, on the drawing board, you can't just take a blob and split it brutally.

      Multi-threading requires man-hours and skill. It is not granted by a framework unless you are doing something with 0 innovation in it (the nth "great webapplication for company accounting" for example) where you can just take a framework that is multi-threaded already and assemble the programming equivalent of Lego blocks to create your program.

      In a nut shell the operating system is extremely important in leveraging those cores and conversely the cores enable modern OS features.
      The OS is just exposing the capability and providing the infrastructure to deal with many threads running independently.

      If the applications don't multi-thread themselves, there is little the OS can do. Which is my point.

      Again you are hung up on the idea that consummers use a singke program and that is it. On the other hand reality is far different. Even a person surfing the net in a browser may have other apps running these days, a streamming app perhaps, most likely an E-mail app and whatever else. The days of DOS are far behind us!!!
      That's not what multi-threading is about. That's multitasking. Even a single-core system can run multiple different programs at the same time, since pretty much any OS worthy of the name is multitasking.

      the way many people run games they would be just as well off with a modern version of DOS.
      DOS is not multi-threading either, it would not allow to use more than a single core effectively.

      The only reason they started making multi-core CPUs is because they could not make faster single-core ones. So no, using only a single core would severely limit the processing power the game has access to.

      Multithreading is NOT the same thing as multitasking. Multitasking is easy, as it is dealt with by the OS. Multithreading is hard, as it requires the program itself to split itself up in many smaller "programlets" that are dealt with as they were separate programs in a multitasking OS.

      A failed attempt at paralleling an app means nothing in the context of todays user where multiple apps in use and highly threaded apps exist. Even a modern web browser beƱefits highly from cores with the corresponding OS suppport.
      "highly threaded" yeah right. Even browsers "highly thread" by just using a thread per web-page rendered, or splitting the media rendering on its own thread or similar. That's kindergarden-grade multithreading.

      Ask yourself why the Apple phones/tablets have the CPUs with best IPC (instructions per clock, it's a measure of how powerful is the processor in executing single-thread programs), or why Intel is still doing their damn best to provide consumer hardware that has the best IPC possible.
      Last edited by starshipeleven; 02-07-2018, 05:13 PM.


      • #33
        Originally posted by starshipeleven View Post
        ... why Intel is still doing their damn best to provide consumer hardware that has the best IPC possible.
        Because Intel was ignorant and invented ventilators to PC decades ago, as higher IPC core means power going to the roof

        I don't know what is your definition of consumer hardware, as everything is consumer hardware to me

        Ask Michael how much thousands of kilowatts he use per month to run all these benchmarks, just because of all of that IPC

        Not only that Intel was ignorant, but these consumers were even more ignorant... top notch IPC core they wanna OC even more, for a couple percent boost in performance, power usage doubling - what kind of sport that is?

        At the end we get Meltdown/Spectre just because that consumer world is so crazy and wanna perf no matter what
        Last edited by dungeon; 02-07-2018, 06:03 PM.


        • #34
          Originally posted by schmidtbag View Post
          Though I personally don't know of any SoCs that include both RAM and storage (not even EEPROM or CMOS), I'm pretty sure there are a handful of ARM SoCs that have embedded RAM, such as some OMAP and Exynos products. For example, take a look at the ODRIOD-XU4. This platform has 2GB of RAM, but there aren't any chips on the board. There are other products by Hardkernel (like the ODROID-C2) which do explicitly point out their RAM chips, so it's not like the manufacturer was being negligent or lazy.
          Neat. Looking at the SoC's specs it's confirmed, this thing has integrated RAM in the SoC

          I guess Samsung can get away with it.


          • #35
            Originally posted by dungeon View Post
            Because Intel was ignorant and invented ventilators to PC decades ago, as higher IPC core means power going to the roof
            Yeah. because multicore processors don't have the same TDP.

            I don't know what is your definition of consumer hardware, as everything is consumer hardware to me
            You buy many 12-core Xeons that cost 1000+$ apiece for your 1000+$ multi-socket motherboard in your HTPC?

            Not only that Intel was ignorant, but these consumers were even more ignorant... top notch IPC core they wanna OC even more
            Can they add more cores to their processors to increase performance?


            • #36
              Originally posted by wizard69 View Post

              Generally a System On Chip is everything excluding RAM and secondary storage. The vast majority of SoC that have RAM included do so as multi chip modules of one sort or another. Note that multi chip modules dont always use SoC in their implementation.

              By the way i dont think we are far away from seeing multi chip modules implementing all of the system RAM within the module. That is a APU style SoC wired directly to 8 or 16 GB of RAM (HBM) in a multi chip module. This wouldnt be much different than some of todays GPUs.
              I hope to see such solution from Intel in next 2 years.


              • #37
                I don't believe sub-100 watt TDP. Add 10%-25% to the TDP value that intel gives to get actual TDP.


                • #38
                  Originally posted by tildearrow View Post
                  I have a question. Why are "enthusiast" consumer processors (like the 7980XE) clocked higher than high-core-count server/workstation processors?
                  You get a lot better perf/watt if you go with lower clocks. It's the same reason why mobile phones don't go too high - they need to be power efficient.

                  In the data center it counts, because you have thousands of CPUs and perf/watt is crucial for running 24/7... cost savings add up.


                  • #39
                    Originally posted by wizard69 View Post
                    This is so true for many server duties. But modern operating systems are also well supported by multiple cores. It might be hard for the younger members to understand just how bad it was to run a modern operating system on a single core machine. The first dual chip machines (two cores) had a huge impact on OS usability be that OS is Windows, Linux or MacOS. Core literally where a turning point for anybody using an OS beyong the trivial.
                    I 'member the old days when the first multi CPU x86 systems sprung up. Linux was still pretty terrible at handling multiple CPUs ('member the BKL/big kernel lock?), and it took a few years for the BSDs to catch up.


                    • #40
                      Originally posted by Yndoendo View Post
                      Will not touch Intel until at least Meltdown is fixed on silicon. AMD is currently your safer bet.
                      Of course then you're essentially limited to 8 Cores per NUMA node, not ideal for Software Defined {Storage,Network}.