Announcement

Collapse
No announcement yet.

NVIDIA Announces Grace CPU For ARM-Based AI/HPC Processor

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

  • #71
    Originally posted by coder View Post
    There are knowledgeble folks who would sincerely disagree on the latter point, as well as whether ARM is really even RISC. Rather than take a position on that, I just want to point out that the advantages of AArch64 include:
    • simpler ISA -> simpler, more energy-efficient decoder
    • fixed-sized instruction word -> wider front-end
    • larger GP register file -> less spilling
    • relaxed memory-consistency -> greater instruction-reordering flexibility

    These are undeniable, though you can certainly debate the impact each has on performance.
    I meant more that CPUs internally are more RISC in nature, as big intstructions always gets decoded into smaller very simple RISC like instructions and go onto pipeline. X86 by itself or newer ARM architectures are not by itself RISC.

    Comment


    • #72
      Originally posted by coder View Post
      I think that's not what they meant. I imagine they had in mind that one OS kernel should be managing hybrid ISA CPU cores, which share a global pool of RAM. This would be an interesting project, but I'm not sure we really have anything like it, today.
      That sounds horrible to code in and extremly challenging to compiler. Honestly CUDA more behaves like that already.

      Comment


      • #73
        Well...WELL !! The AGE of ARM ( with Nvidia at the helm ) just got a WHOLE LOT more interesting !!

        Nvidia and Mediatek announced today a partnership to bring Nvidia RTX GPUs to Mediatek ARM based SoCs.

        " According to the announcement, the two companies are partnering to create a “reference platform” that supports “Chromium, Linux, and NVIDIA SDKs.” More specifically, MediaTek’s Arm-based systems on a chip (SoCs) will be combined with Nvidia’s RTX GPUs to bring ray-tracing graphics and more to Chromebooks.
        MediaTek is the world’s largest supplier of Arm chips, used to power everything from smartphones, Chromebooks, and smart TVs. We look forward to using our technology and working with NVIDIA to bring the power of GPUs to the Arm PC platform for gaming, content creation, and much more. GPU acceleration will be a huge boost for the entire Arm ecosystem.

        Rick Tsai, CEO, MediaTek "

        https://9to5google.com/2021/04/13/me...x-chromebooks/

        Comment


        • #74
          The ARM shilling in this forum is getting annoying.

          NVIDIA wants to lock everyone in and keep competitors out by ignoring industry standards. There's nothing commendable about this disgrace.

          Comment


          • #75
            Originally posted by numacross View Post
            The A1100 is not socketable, it's BGA
            That is not quite right.
            https://www.ironwoodelectronics.com/...a-sockets.html
            All BGA products can be put in a socket. Is is a common produced AMD socket no its not. Are all systems with a A1100 inside BGA soldered on the answer some do have third party sockets. Yes there are through hole designs for the A1100 for test boards as to mount a BGA socket though hole is way better than surface mount pads.

            Think about it there are times you want to validate chips function before they get exposed to the heat of soldering so you can work out if a problem is because of your soldering process or if you have got a lemon batch for some reason.

            People incorrectly presume just because something is BGA packaged that it does not have socket. 99% of everything that is BGA has a socket for making chip testing boards. Socket costs more money and requires more PCB space than using the BGA soldered on.

            Comment


            • #76
              Originally posted by Jumbotron View Post
              Nvidia and Mediatek announced today a partnership to bring Nvidia RTX GPUs to Mediatek ARM based SoCs.
              This is basically like how Samsung licensed RNDA from AMD.

              Also, it's one of the likely outcomes people were talking about, when the Nvidia/ARM acquisition got announced. For ARM customers using Mali GPUs, it's probably now a much shorter step to upgrade to a Nvidia-designed unit.

              It also sounds like it brings decent ARM mini-PCs a step closer.

              Comment


              • #77
                Originally posted by oiaohm View Post

                That is not quite right.
                https://www.ironwoodelectronics.com/...a-sockets.html
                All BGA products can be put in a socket. Is is a common produced AMD socket no its not. Are all systems with a A1100 inside BGA soldered on the answer some do have third party sockets. Yes there are through hole designs for the A1100 for test boards as to mount a BGA socket though hole is way better than surface mount pads.

                Think about it there are times you want to validate chips function before they get exposed to the heat of soldering so you can work out if a problem is because of your soldering process or if you have got a lemon batch for some reason.

                People incorrectly presume just because something is BGA packaged that it does not have socket. 99% of everything that is BGA has a socket for making chip testing boards. Socket costs more money and requires more PCB space than using the BGA soldered on.
                Of course there are sockets able to take BGA chips, but as you said they are mainly used for development and testing since they are quite expensive. The bigger ones are actually prohibitively expensive due to the torque tooling used in them.

                Comment


                • #78
                  More details from Nvidia's GTC event on it's partnership with Mediatek. Nividia straight up says that they are going to expand the ARM eco-system to PC desktops as well as laptops.

                  The Age of ARM is truly and verifiably here.

                  " About halfway through his presentation, Huang presented a slide entitled "Expanding Arm Ecosystem Beyond Mobile," which mentioned MediaTek's collaboration. According to NVIDIA, it will be licensing its GeForce RTX 30 Ampere IP for use in MediaTek MT819x SoCs. "

                  "We're announcing a partnership with MediaTek to create a reference system and SDK for Chrome OS and Linux PCs," said Huang during the presentation. "MediaTek is the world's largest SoC maker. Combining NVIDIA GPUs and MediaTek SoCs will make excellent PCs and notebooks."

                  " While we might be most familiar with Ampere in desktop and laptop applications (where it draws a considerable amount of power), Ampere can also be scaled down for low-power situations. We see this with NVIDIA's Orin SoC which includes Hercules Arm CPUs and Ampere GPUs in a TDP ranging from 5 to 45 watts. "

                  https://hothardware.com/news/future-...or-chromebooks

                  Comment


                  • #79
                    After pondering a bit more about Nvida's announcement at GTC that they are partnering with Mediatek to bring RTX Ampere GPUs to Mediatek' ARM SoCs it does make some sense. I think Nvidia ALSO sees the Age of ARM happening and also saw themselves getting squeezed out of the ARM consumer market.

                    Apple has their own GPU IP and Core for their ARM based M1. Qualcomm has long had their ATI based IP and GPU core called Adreno (an anagram of "Radeon"). Samsung is now licensing RDNA IP from AMD for new GPU cores for Samsung's Exnyos ARM based SoCs. Which, by the way, may show up this Fall in Google's custom ARM SoC, code named "Whitechapel" which will debut in the upcoming Pixel 6. And Broadcom tends to just go with ARM's own Mali GPU.

                    So that leaves Nvidia with little choice but to go with Mediatek. But as Mediatek sells more ARM based SoCs than anyone else, even Qualcomm, it's not a bad deal for Nvidia to get into the ARM consumer world as they are in the x86 world.

                    Comment


                    • #80
                      Originally posted by numacross View Post
                      Of course there are sockets able to take BGA chips, but as you said they are mainly used for development and testing since they are quite expensive. The bigger ones are actually prohibitively expensive due to the torque tooling used in them.
                      There are many different designs of torque tooling. You do get BGA sockets in basically the same designs as Land grid array used by Intel with a pinning difference to allow for the solder balls. Yes a Land grid array chip can be used as a BGA just apply solder. Of course there is a catch the force applying heat spreader is not include in the BGA chips price so the BGA socket look more expensive but if you are using cpu + socket cost here it basically the same the heat-spreader cost moves from the CPU to the socket cost.

                      The metal heat-spreader on a modern cpu is doing two jobs even that people call it a heat-spreader its a heat and force spreader. Using a force spreader on top of chip provides a really good way to dependably torque down a chip.

                      Basically there are multi designs of BGA chip sockets. I have used one of the Land Grid array like BGA sockets on a large FPGA that only comes in BGA package where it is happens to be a risk of getting fried if particular things go wrong. This is not a testing BGA socket this a production grade socket that is only expected to be undone and redone under 1000 times in production. Testing grade BGA sockets they are required to a few million times in endurance.

                      More expensive torque tools are using in big sockets of BGA when they are designed multi times a day undone and redone for testing.

                      Funny enough the big thing that make testing BGA sockets more expensive is not exactly the torque down bit. Its extra bits so they have to be cleaned less those solder balls under the BGA are not your best friend. Yes the problem of the grim the solder balls can be basically ignored for a production 1000 or less insert because clean socket can be cleaned in the middle. Testing BGA sockets also don't want have a exact matched force spreader so multi models of the same chip can simple go into the same socket.

                      There are two difference classes of BGA sockets. Testing and production sockets both have there place. Some cases production soldering a chip on even if it only comes in bga form is bad news. Production BGA sockets look very much like you land gate array sockets with basically the same costs but slightly differently distributed.

                      We have seen from china maker do BGA to adaptor to Land grid array with a custom force spread with BGA silicon dia straight to heatsink and this was to put a batch of left over intel BGA chips into standard intel motherboards. Yes this was a non soldered assembly between the BGA and adaptor. So the idea that you are needing more complex/expensive than your standard AMD/Intel land gate array down force system to torque a large BGA is wrong.

                      The pinning in the BGA socket to deal with the solder balls properly is slightly more expensive than than a land gate array socket but its also more durable but we are talking under 2 USD dollars for something as many pins as a epyc cpu.

                      numacross you are not alone in thinking that BGA sockets have to be massively expensive. Most people don't have a clue that there are production BGA sockets and testing BGA sockets with a massive difference in cost between them that purely link to how many inserts and removes are expect to be performed in their life time and how adaptable they are to other chips other than the intended one for the design.

                      Of course a Land grid array socket used under CPU is not exactly cheap when you compare it to soldering that land grid array chip straight on-to the board if you are doing something custom in volume as well. This is also why BGA sockets are not common because a lot of BGA chips are being used in places where there is no intention to repair/replace/upgrade so might as well save the cost and not have a socket. Directly soldered on does gain some reliability as well over socket option due to not having to worry about working loss.

                      Comment

                      Working...
                      X