Announcement

Collapse
No announcement yet.

35-Way Linux GPU Graphics Comparison, Initial NVIDIA RTX 40 SUPER Linux Benchmarks

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

  • #11
    Originally posted by avis View Post

    Highly praised and beloved here AMD for some reasons didn't want to undercut NVIDIA, thus you can "enjoy" the insane pricing that we have.

    Had AMD released 7900 XTX and 7900 XT at $200 lower NVIDIA would have never retained the same pricing that they've had.

    Too bad AMD decided to play along and enjoy fat margins.

    But you will still love AMD and hate NVIDIA even though both companies care a ton more about their shareholders than customers.
    Exactly, both companies are in the business of making money, which is why businesses exist in the first place.

    For some off reason AMD gets this ridiculous love from its advocates, they act as if Robin Hood is the CEO.

    Comment


    • #12
      With Vulkan ray-tracing workloads even for the relatively basic Quake II RTX, the performance continues to side with the NVIDIA RTX GPUs. The RADV ray-tracing support has made great strides over the past two years and is now at least playable for more ray-traced titles on Linux, but the best ray-tracing game performance on Linux continues to be with NVIDIA GPUs.

      Want to know why it runs better on Nvidia?

      Because they created it:

      The eagerly-anticipated ray-traced remaster of the classic Quake II is coming soon. Learn about the latest improvements and enhancements, check out a new trailer, and see how RTX ON compares to the original 1997 release.



      Furthermore, the source code will be posted to GitHub, enabling others to expand on our work with further advancements and enhancements, or to use the code to ray trace Quake II mods and total conversions.‚Äč

      For me the take away is that a game released in 1997, redone using raytracing reuires such beefy hardware to be playable at decent frame rates.

      I wonder if they will release a Quake III RTX or an UT RTX?

      Even better, Return to Castle Wolfenstein RTX or Doom III RTX, how awesome would that be?

      Comment


      • #13
        Can someone educate me about the art of benchmarking? Why not include the top AMD card across the benchmarks for reference?

        Comment


        • #14
          Wow, that's a lot of tests Michael. Thank you. However I would have liked to see more AMD tests, and the usual performance/power summary at the end.
          Last edited by muncrief; 31 January 2024, 02:48 PM.

          Comment


          • #15
            Originally posted by TheJackiNonster View Post
            Isn't PyTorch already compatible with ROCm on AMD GPUs via HIP? Why aren't these included in the benchmarks? Maybe it's not the same as CUDA but it's also kind of dishonest to keep it separate. Both require the usage of proprietary drivers. I think it would be interesting to see how much difference it actually still makes to buy into Nvidia for such tasks.
            it is plain and simple wrong to claim that PyTorch need Proprietary drivers i did run PyTorch with ROCm/HIP with the fedora distro packages all on the open-source driver and no driver downloaded from AMD.com

            this CUDA propaganda is well harmfull to the people and the market.
            Phantom circuit Sequence Reducer Dyslexia

            Comment


            • #16
              Originally posted by avis View Post
              Highly praised and beloved here AMD for some reasons didn't want to undercut NVIDIA, thus you can "enjoy" the insane pricing that we have.
              Had AMD released 7900 XTX and 7900 XT at $200 lower NVIDIA would have never retained the same pricing that they've had.
              Too bad AMD decided to play along and enjoy fat margins.
              But you will still love AMD and hate NVIDIA even though both companies care a ton more about their shareholders than customers.
              there is no competition in the market if Nvidia find a niche to make more money and profit and also there is no competition if AMD find a niche to make more money and profit.
              and both have niche... Nvidia is on the AI train with CUDA and AMD is on the Niche of open-source drivers and Open Standards and open technology like FSR3
              they will only start to compete in price again if their niche is dry and they can not grow there anymore means no more profit and turnover in the niche.

              and just remember the crypto currency boom ? high prices for many years only because AMD/Nvidia did make more money with selling GPUs to cryptominders in comparison to any serious business and gamers.
              Phantom circuit Sequence Reducer Dyslexia

              Comment


              • #17
                Originally posted by qarium View Post

                it is plain and simple wrong to claim that PyTorch need Proprietary drivers i did run PyTorch with ROCm/HIP with the fedora distro packages all on the open-source driver and no driver downloaded from AMD.com

                this CUDA propaganda is well harmfull to the people and the market.
                Was this before or after Israeli intelligence hacked your computer using a nonexistent virus delivered via a nonexistent exploit?

                Comment


                • #18
                  Originally posted by qarium View Post
                  Nvidia is on the AI train with CUDA and AMD is on the Niche of open-source drivers and Open Standards and open technology like FSR3
                  they will only start to compete in price again if their niche is dry and they can not grow there anymore means no more profit and turnover in the niche.
                  Are you saying that their niche needs to be wet?

                  Comment


                  • #19
                    Originally posted by TheJackiNonster View Post
                    Isn't PyTorch already compatible with ROCm on AMD GPUs via HIP? Why aren't these included in the benchmarks? Maybe it's not the same as CUDA but it's also kind of dishonest to keep it separate. Both require the usage of proprietary drivers. I think it would be interesting to see how much difference it actually still makes to buy into Nvidia for such tasks.
                    Sadly, HIP is in a state where it may work today and be completely broken tomorrow for no apparent reason, for example in RDNA2 some days Blender works great with HIP and some other days completely crash my system by just opening blender.

                    Even tho Pytorch support HIP in some versions, you basically depends on how the weather is on Neptune and the amount of neutrinos hitting the earth in the exact opposite side of the planet of where you are lol(it may work great or work but randomly slow down or may just crash you system to hell, so pray to the goods, close your eyes and gamble on)

                    CUDA has some quirks here and there from time to time but mostly just works, that is why

                    Comment


                    • #20
                      4080 and 4080 Super are very similar, and Windows benchmarks are showing barely 1% uplift on FE edition overall per TPU. Why such a large uplift on Linux?

                      Comment

                      Working...
                      X