Announcement

Collapse
No announcement yet.

ARM Announces Cortex-A76 Processor, Mali-G76 & Mali-V76

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

  • #11
    Originally posted by Space Heater View Post
    Hard to get excited about new Mali graphics when ARM refuses to support open drivers.
    For me it's not even a matter of having open drivers. Having fully functional drivers on a modern kernel is all I ask.

    Comment


    • #12
      I'm hoping the Panfrost project pans-out. Lol. The project lead is doing all the work on an armv7 chromebook. But they have a proof of concept and are planning to get it to work with mesa. I wish my skills were good enough to help on a project like that.

      To anyone wanting an arm box with decent power behind it, there's the tegra series (very expensive though), and the latest suscessful chipset from rockchip RK3399. Boards with that chip are the rock64pro and the TBA Odroid-N1 which should be announced for sale in the coming weeks I guess. I have that chipset powering my Chromebook and it's pretty snappy. It's not gonna replace a box used for compiling large projects or video editing/transcoding, but for what it is, it's quick. Good for browsing the interwebs, hosting a high speed NAS, etc.

      I don't really expect to see this new arm CPU IP in the wild any time soon. Manufactures use the crap out of what they already paid for and only turn to new hardware when what they're currently using won't cut it. So don't expect to see this CPU for 3 to 5 years. Don't expect to see Linux support for that processor for another 2+ years after that.

      Comment


      • #13
        schmidtbag well no one ever made a explicitly the laptop class ARM SoC. The A76 should be approximately the same performance as QC Falcon core's but more power efficient. Still there are no announcements anyone is working on a laptop class SoC but the QC S855 will probably be based on the A76's, announced by the end of the year and availability in first commercial products in 1st Q of 2019. Would be nice to see a hexa A76 (&two A55`s DinamIQ cluster & who knows me by we will.
        The A76 is a 13 stage long pipe design meaning it can go beyond the 3 GHz easily but doing that in any power limited device would be insane thing to do. ARM did a master work in lowering the latencies for instructions and lowering latencies and increasing true output for caches (the area ARM used to lag behind the X86 a lot).

        The current FinFET structure transistors meat their sustainable saint maximum levels at 2.1~2.2 GHz (while ideal performance/power/leakage delta is around 1.65~1.75 GHz) disregarding of vendor/foundry or lithography process (nm). So going beyond that for anything more than a short burst or a limited amount of core's used is insane in anything that have limited DTP (including servers) as it leaks & spills as heel. ARM actually did a fantastic yob with A76 producing wider (bigger) OoO core which is able to scale up performance very good to the area it takes. So A76 is 65% faster than an A73 while it's roughly 2x the size (which is a big thing regarding two to four intrusion per clock design comparison), when you count in the switch from the 14 nm to the 7nm lithography you get 1.8x performance per same DTP. You didn't sean that much increase in x86 space for a whole decade. While ARM OoO designs are getting bigger/wider & have higher performance the x86 part's are remaining the same (regarding the design) only scaled down to new manufacturing node but ARM managed to retain (the most part) of it's power efficiency advantage and it's scale up proportional to the manufacturing node advancement.

        Best regards.

        Comment


        • #14
          Originally posted by dwagner View Post
          Strange that they do not even mention any Spectre mitigation attempts. One would think that every customer wants to know if a new core is still susceptible to these dire security flaws.
          Every ARM Ltd customer has very likely been informed. They'll in turn let their customers (us) know.

          Half kidding. I would have expected some words about that. But perhaps it was just not reported by journalists?

          Comment


          • #15
            Originally posted by debianxfce View Post

            Android kernels are not modern. For the octa core Amlogic S912 SOC you need to use the 3.16 kernel derived from android if you want to have hardware accelerated graphics and audio. The Amlogic Meson project is an attempt to have support in the mainline kernel and no support for never Mali gpus exists. Youtube video software rendering works with the Amlogic S912 but the mainline kernel has no audio support.
            some android kernels are modern, just not the majority of kernels shipping on devices... most Oreo devices ship linux-4.4, but there is the odd board that's using linux-4.9 (so not that old)... hikey is one such board...

            and there are also several branches for 4.9 and mainline tracking in AOSP. examples;

            Contribute to aosp-mirror/kernel_common development by creating an account on GitHub.

            Contribute to aosp-mirror/kernel_common development by creating an account on GitHub.



            personally, I find it super annoying that my Google Pixel (OG) shipped with linux-3.18 - so old. grrrr!. ... I ended up pulling a lot of code from newer kernels, in order to support features that will never be found on such an old kernel... it would've been better if google shipped linux-4.4 on the OG pixel.... I imagine pixel 3 will ship linux-4.9 though ~ so maybe the next-gen of flags ships will be using a fairly modern kernel...

            Comment


            • #16
              Originally posted by RelaxTrolls View Post

              some android kernels are modern, just not the majority of kernels shipping on devices... most Oreo devices ship linux-4.4, but there is the odd board that's using linux-4.9 (so not that old)... hikey is one such board...

              and there are also several branches for 4.9 and mainline tracking in AOSP. examples;

              Contribute to aosp-mirror/kernel_common development by creating an account on GitHub.

              Contribute to aosp-mirror/kernel_common development by creating an account on GitHub.



              personally, I find it super annoying that my Google Pixel (OG) shipped with linux-3.18 - so old. grrrr!. ... I ended up pulling a lot of code from newer kernels, in order to support features that will never be found on such an old kernel... it would've been better if google shipped linux-4.4 on the OG pixel.... I imagine pixel 3 will ship linux-4.9 though ~ so maybe the next-gen of flags ships will be using a fairly modern kernel...
              The Pixel 3 will run a newer kernel for sure, but it's most likely going to be just 4.0 :P

              Comment


              • #17
                Originally posted by Vistaus View Post

                The Pixel 3 will run a newer kernel for sure, but it's most likely going to be just 4.0 :P
                4.0?! Hell no. that's absolutely wrong, what I said is most likely correct... why?

                - 4.0 is EOL (for that reason alone, it won't be used.)
                - pixel 2 ships linux-4.4, which is the first 6yr extended LTS kernel (so pixel 3 will ship at least that kernel, but most likely 4.9...)
                - most of the EAS integration is happening on 4.9-eas-dev branch && there are stable 4.9 branches in AOSP now.. I would imagine that for the pixel 3, google will be wanting to be using EASv1.5+... the code could be backported to linux-4.4 (which I've done for my pixel's 3.18 kernel, but it's a pain && it seems like the android + eas integration is already great on 4.9).
                - there are features starting to be used in android that require 4.4+ kernels.

                there's absolutely no reason 4.0 would be used. i'm not sure where you get the impression that it would be...?

                Comment


                • #18
                  Originally posted by Apokalypz View Post
                  To anyone wanting an arm box with decent power behind it, there's the tegra series (very expensive though), and the latest suscessful chipset from rockchip RK3399. Boards with that chip are the rock64pro and the TBA Odroid-N1 which should be announced for sale in the coming weeks I guess. I have that chipset powering my Chromebook and it's pretty snappy. It's not gonna replace a box used for compiling large projects or video editing/transcoding, but for what it is, it's quick. Good for browsing the interwebs, hosting a high speed NAS, etc.
                  Well for RK3399 SBC I've heard of this NanoPC-T4 , but I'd like to buy from the EU if possible. http://www.friendlyarm.com/index.php...product_id=225

                  Comment


                  • #19
                    Im kinda hoping that ARM can start to support GPU drivers for Linux. That is in the same way Intel does. In the end we can wish all we want for ARM hardware but not having drivers justvsucks. It sucks enough that many users wont bother.

                    As for availibility remember these are designs that get integrated in vendor SoC so an entire round of engineering has to happen. It will be interesting to see how these stack up against Apples A12 when released. By the way 7nm frees up a lot of die space, this means more cores, speciàl function units and other features come free. It should be interesting to see what gets packed into these new SoC.

                    Comment


                    • #20
                      if one wanted to make a vulkan driver for a yet unsuported or new gpu what would the mnium requiered ? ; how many components could one do away with ; would something analoge to pal be necessary to support one , low level api ? ( lets say the os gui uses vulkan 1,1 not gl stream or such )

                      Comment

                      Working...
                      X