Announcement

Collapse
No announcement yet.

NVIDIA Releases 470.57.02 Linux Driver, DLSS SDK Adds Official Linux Support

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

  • #31
    Sadly this release breaks wayland, back to the beta driver I guess.

    Comment


    • #32
      Originally posted by avem View Post
      You continue to lie as you've been doing for ages. No, Valve chose AMD because AMD is the only company which has a high performance x86 uArch and a fast GPU arch.

      Intel: only CPU.
      The GPU benchmarks for Tiger Lake look pretty comparable to AMD's APUs. The deal-breaker with Tiger Lake is that it's simply going to be too expensive. If Intel had more 10 nm superfin production capacity, they might be able to cut Valve a sweetheart deal and plant their flag in the console market, but that's not the current reality.

      Originally posted by avem View Post
      NVIDIA: only GPU.
      Nintendo Switch is using one of their ARM-based SoCs. That said, Switch's SoC is old and the newer SoCs have been AI-focused, at the expense of GPU performance.

      Originally posted by avem View Post
      Valve couldn't have chosen anyone else.
      MediaTek, Qualcomm, Samsung... basically any high-end tablet/laptop SoC would be an option. Just that none have (AFAIK) quite Vega8 levels of performance, at least until Samsung ships their RDNAx-based SoC (but that'll probably have the pricing issue, I'm guessing).

      Imagination is still out there... I don't know if anyone is using their high-end GPU (other than Apple using licensed derivatives). I know that MediaTek has traditionally been a customer, but seems now to be in ARM/Nvidia's orbit.

      Originally posted by avem View Post
      Going the ARM route is a suicide considering Steam has no ARM support or ARM games in its catalogue.
      I take your word for it. So, you're saying that Steam doesn't exist on Android or iOS?

      Comment


      • #33
        Originally posted by coder View Post
        The GPU benchmarks for Tiger Lake look pretty comparable to AMD's APUs. The deal-breaker with Tiger Lake is that it's simply going to be too expensive. If Intel had more 10 nm superfin production capacity, they might be able to cut Valve a sweetheart deal and plant their flag in the console market, but that's not the current reality.
        That was too late for the first Valve deck. That chip only came samples in October 2020. Valve had samples of the AMD chip early 2020 as in Jan 2020 Dec 2019. Now Intel for the next generation of Valve deck if it comes may be in place. Yes the time to QA a product. Without covid-19 and silicon supply issues there is odds that the Valve deck would have been released Nov-Dec 2020.

        Originally posted by coder View Post
        MediaTek, Qualcomm, Samsung... basically any high-end tablet/laptop SoC would be an option. Just that none have (AFAIK) quite Vega8 levels of performance, at least until Samsung ships their RDNAx-based SoC (but that'll probably have the pricing issue, I'm guessing).

        Imagination is still out there... I don't know if anyone is using their high-end GPU (other than Apple using licensed derivatives). I know that MediaTek has traditionally been a customer, but seems now to be in ARM/Nvidia's orbit.
        https://9to5google.com/2021/06/29/sa...formance-leak/
        Being made for mobile devices the samsung SoC are not going to be super fast.
        https://en.wikipedia.org/wiki/Tegra
        Its really samsung/amd socs are going to give Nvidia socs a headache.

        https://blogs.nvidia.com/blog/2021/0...e-rtx-arm-gdc/
        Not exactly. As we are getting more arm socs with PCIe 4.0 the dGPU connected to arm cpu is coming a option. Yes this is MediaTek and Nvidia doing a joint showing. Welcome to fun future where Nvidia laptop GPU with competitor ARM CPU will be able to compete with arm Tegra offerenings

        Of course AMD with open source drivers if you had a AMD GPU and a sane enough PCIe implementation on arm you have been able to this for a while and of course being open source drivers your solution is not restricted to AMD selected platforms. Yes Nvidia offering is restricted to Nvidia selected platfoms.

        Originally posted by coder View Post
        I take your word for it. So, you're saying that Steam doesn't exist on Android or iOS?
        https://store.steampowered.com/mobile
        It exists but its not what you call usable. As in you can buy games and chat on Android and IOS but you need a Linux, Windows or Mac PC to play them.

        Comment


        • #34
          Originally posted by coder View Post
          The GPU benchmarks for Tiger Lake look pretty comparable to AMD's APUs. The deal-breaker with Tiger Lake is that it's simply going to be too expensive. If Intel had more 10 nm superfin production capacity, they might be able to cut Valve a sweetheart deal and plant their flag in the console market, but that's not the current reality.


          Nintendo Switch is using one of their ARM-based SoCs. That said, Switch's SoC is old and the newer SoCs have been AI-focused, at the expense of GPU performance.


          MediaTek, Qualcomm, Samsung... basically any high-end tablet/laptop SoC would be an option. Just that none have (AFAIK) quite Vega8 levels of performance, at least until Samsung ships their RDNAx-based SoC (but that'll probably have the pricing issue, I'm guessing).

          Imagination is still out there... I don't know if anyone is using their high-end GPU (other than Apple using licensed derivatives). I know that MediaTek has traditionally been a customer, but seems now to be in ARM/Nvidia's orbit.


          I take your word for it. So, you're saying that Steam doesn't exist on Android or iOS?
          1. I haven't seen the recent comparisons between TGL IGP and Vega 2 found in AMD's APU, so I cannot comment on that. Please add citations. Recent reviews show that Ryzen 5600G/5700G decimate Intel but testers tested the IGP in the 11 desktop series processors and they feature quite anemic GPUs compared to mobile Intel offerings. But remember that RDNA 2.0 is leaps and bounds better than Vega found in these APUs, so it's far from obvious whether Intel is competitive.

          2. Again ARM is not an option.

          3. Again ARM is not an option.

          4. Again ARM is not an option, since the Steam App is a shell for their website. It cannot be used to launch or play any Windows games on your Android/iOS device. When you mentioned it I didn't understand what to think because it just sounded ridiculous if not idiotic.

          Comment


          • #35
            Originally posted by avem View Post
            1. I haven't seen the recent comparisons between TGL IGP and Vega 2 found in AMD's APU, so I cannot comment on that. Please add citations. Recent reviews show that Ryzen 5600G/5700G decimate Intel but testers tested the IGP in the 11 desktop series processors and they feature quite anemic GPUs compared to mobile Intel offerings. But remember that RDNA 2.0 is leaps and bounds better than Vega found in these APUs, so it's far from obvious whether Intel is competitive.
            https://www.youtube.com/watch?v=BAs2iWX7dsE 11 series IGP the Xe 96EU ones and the Ryzen 5xxxG/H vega 8 are fairly down right close to each other for graphical performance . Tiger Lake IGP is quite a uplift compared to past intel igpu offering but there is still artefact problems with Intel drivers. So Intel was not exactly ready for supply time for the steam deck. The RDNA 2.0 in the steam deck is suspected to be 15-20% uplift over past Vega 8 but please do take this with a serous grain of salt as we don't have the benchmarks confirming this but if it is Intel Tiger Lake igpu would not have competitive option vs RDNA 2.0 APU that AMD was offering Valve.

            Please note AMD did have close to 100 percent free artefact free drivers they could demo for RDNA 2.0 they could demo over year ago. Basically intel could not tick the boxes not with the hardware they could make.

            Intel 11 series its not just 1 iGPU there is 3 of them.
            https://www.intel.com.au/content/www...rocessors.html Its 3 iGPU designs. Xe 32EU and the UMD 720 options get kicked into the ground by a Vega 8. The Xe96EU puts up a very good showing against vega 8. There are quite a few reviewers who have missed the third option the Xe 96EU also a more expensive option.

            https://ark.intel.com/content/www/us...-with-ipu.html
            Yes that youtube video where the Intel iGPU is close is the Xe 96EU to Vega 8 not the weaker Xe 32EU or UMD 720 igpu options.

            By the way the tiger lake branding is killed off by intel what was Tiger lake branded is now Rocket Lake.

            Originally posted by avem View Post
            2. Again ARM is not an option.

            3. Again ARM is not an option.

            4. Again ARM is not an option, since the Steam App is a shell for their website. It cannot be used to launch or play any Windows games on your Android/iOS device. When you mentioned it I didn't understand what to think because it just sounded ridiculous if not idiotic.
            Arm is not 100 percent not a option Long term. The time frame of the steam deck having to get to production it was not option. Wine hangover work may see x86 binary on arm be good enough in time.

            Comment


            • #36
              Originally posted by oiaohm View Post
              11 series IGP the Xe 96EU ones and the Ryzen 5xxxG/H vega 8 are fairly down right close to each other for graphical performance .
              Yes, this is what I was referring to. You can search this site for Michael's i7-1165G7 benchmarks to see the big Xe iGPU in action, but I didn't find an article that compared it with AMD APU graphics performance.

              Originally posted by oiaohm View Post
              By the way the tiger lake branding is killed off by intel what was Tiger lake branded is now Rocket Lake.
              No, but your confusion is understandable. Tiger Lake is the 11th Gen laptop/NUC-oriented product line (built on 10 nm SF), while Rocket Lake is the somewhat equivalent (but based on an earlier micro-architecture) for the desktop (built on 14 nm+++++).
              Also, neither of these names count as "branding". They're internal code names. But, they get onto roadmap slides and in certain documents, and then the tech press reports on them. So, that's how we get to know them. You won't find them being used in advertising aimed at normal end users, however.

              Comment


              • #37
                Originally posted by coder View Post
                Yes, this is what I was referring to. You can search this site for Michael's i7-1165G7 benchmarks to see the big Xe iGPU in action, but I didn't find an article that compared it with AMD APU graphics performance.
                https://www.youtube.com/watch?v=BAs2iWX7dsE

                Coder I understand Michael not doing it. As noted on the 2021 video that even today the big Xe IGPU has higher cases of artefacting. This artefacting also true under Linux.

                Performance of a big Xe IGPU to a Vega 8 is close the quality of output is not. This is a direct side effect if the Xe design been newish so the drivers are lacking maturity. AMD has had a lot more time to de-quirk their implementation.

                You think about it if someone buys a i7-1165G7 thinking its graphics is going to be like a Vega 8 they are going to be very disappointed at this stage. This is one of the cases that something looks good in benchmarks but in real usage is horrible, The common Xe artefacting generates flickering effect that end up being detracting in game play. So the Xe problem is in the face detracting artefacting not something like we just missed a texture on something. Hopefully Intel will get their Xe drivers sorted before Xe DGPU land in a big away.

                I would suspect that all the Xe problems are driver solve-able. Remember these Xe artefacting is very like the first generation GCN artefacting that was resolved in time as drivers matured.

                I have to say one of the worse form of artefacting is what the Xe does where it renders the right texture some of the time and sometimes does not render that texture resulting in flicker. Yes earily GCN drivers from AMD had the same problem. First generation of a total new design from Nvidia when they do it also has the habit of having this problem. This issue seams to be part of the GPU early adopter tax and this also can lead to miss leading benchmarks.

                Please note the artefacting normally means that when fixed the GPU will drop slightly in performance normally around 1 to 2 percent but there have been cases where particular GPU when artefacting has been fixed lost 25%. While something has artefacting issues putting it head to head with other items could be deceptive.

                Originally posted by coder View Post
                No, but your confusion is understandable. Tiger Lake is the 11th Gen laptop/NUC-oriented product line (built on 10 nm SF), while Rocket Lake is the somewhat equivalent (but based on an earlier micro-architecture) for the desktop (built on 14 nm+++++).
                Also, neither of these names count as "branding". They're internal code names. But, they get onto roadmap slides and in certain documents, and then the tech press reports on them. So, that's how we get to know them. You won't find them being used in advertising aimed at normal end users, however.
                https://ark.intel.com/content/www/us...-with-ipu.html --Products formerly Tiger Lake--
                I will admit I got it wrong. " Products formerly Rocket Lake"

                Reality is intel is ending usage of code names and moving over to generations only.

                https://www.wepc.com/news/intel-tige...cpu-marketing/
                coder tiger lake is a fun one where they did use the code name in the marketing aimed at normal end users. I am not wrong on the getting rid of branding. Intel marketing department for quite a few years used the code names in marketing because they had attention grabbing names.

                Comment


                • #38
                  Originally posted by oiaohm View Post
                  Coder I understand Michael not doing it. As noted on the 2021 video that even today the big Xe IGPU has higher cases of artefacting. This artefacting also true under Linux.
                  I don't see where he said that, however. He published two benchmark articles of its GPU, one analyzing different software versions and the other looking at generational improvements. In neither case did I find any comments about artefacts, corruption, or non-compliance.
                  Just looking at the software differences, I can see why it might've been early to do GPU benchmarks at launch, but the situation has hopefully stabilized to the point where he could do it now.

                  Originally posted by oiaohm View Post
                  I will admit I got it wrong. " Products formerly Rocket Lake"

                  Reality is intel is ending usage of code names and moving over to generations only.
                  Thanks for following up on that. I'm glad we could clear it up. It's indeed shady of them to use virtually identical numbering schemes for very different chips. Oh well.

                  Comment


                  • #39
                    Originally posted by coder View Post
                    I don't see where he said that, however. He published two benchmark articles of its GPU, one analyzing different software versions and the other looking at generational improvements. In neither case did I find any comments about artefacts, corruption, or non-compliance.

                    Just looking at the software differences, I can see why it might've been early to do GPU benchmarks at launch, but the situation has hopefully stabilized to the point where he could do it now.
                    The reality here is the intel igpu have had artefacts problem for quite some time. So intel to intel is not problem because it is one gpu with artefacts compared to another. It comes a problem when you compare intel to amd that you are no longer comparing apples to apples on what the user will experience or be expected to put up with.

                    Yes a year latter and Intel still has not fixed all the problem. Getting closer. Maybe 2022 when a fair compare between intel igpu vs amd APU can be done. Remember if you did a compare between intel igpu and amd apu at the moment you still have to say that the intel gpu has artefacts and might lower in performance in future.

                    Please note intel artefacts is not non-compliance and its not technically corruption. Artefacts with intel is more rendering faster than it should resulting in a bad output. So the intel gpu is able to do perfectly good output question is how many frames do you need to give up.

                    I would call what Intel done benchmark cheating. It bench well but results are not as good as they should be. Question is how much you have to loss to fix it. Could be 1 to 2 percent could be 50%. While this the case you really cannot do a apples vs apples of intel igpu vs amd apu. Thinking intel igpu cheat basically same way it is far todo intel igpu vs intel igpu.

                    Originally posted by coder View Post
                    Thanks for following up on that. I'm glad we could clear it up. It's indeed shady of them to use virtually identical numbering schemes for very different chips. Oh well.
                    Really not more shady than renaming there nm without in fact improving the tech.

                    Comment

                    Working...
                    X