Announcement

Collapse
No announcement yet.

A Lot More AMD RDNA4/GFX12 Graphics Driver Code Has Landed In Mesa 24.2

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

  • A Lot More AMD RDNA4/GFX12 Graphics Driver Code Has Landed In Mesa 24.2

    Phoronix: A Lot More AMD RDNA4/GFX12 Graphics Driver Code Has Landed In Mesa 24.2

    A lot of AMD GFX12 IP enablement landed in Mesa 24.2-devel over the past week for bringing up the RadeonSI Gallium3D driver and RADV Vulkan driver for the upcoming RDNA4 graphics...

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

  • #2
    AMD!!!

    GPU Computing for all RDNA4 family, derivated and sucessors. Please.

    Comment


    • #3
      And... so what - do you think AMD Linux parts will be 100% polished & ready at day-one, at launch day? (this question is rhetorical)

      Comment


      • #4
        Originally posted by lejeczek View Post
        And... so what - do you think AMD Linux parts will be 100% polished & ready at day-one, at launch day? (this question is rhetorical)
        Imagine...

        Comment


        • #5
          Originally posted by timofonic View Post
          AMD!!!
          GPU Computing for all RDNA4 family, derivated and sucessors. Please.
          you are right its a miracle why AMD does not understand that people expect to tun their LLM/AI on their GPU...
          also vram LLM need much vRAM this means the AMD radeon 8800XT should have 32gb VRAM at minimum.
          my Vega64 is more or less obsolete because it can only run 7billion parameter AI... because it only has 8GB vram...
          so even 16gb vram is not enough the Mistral 8x20B opensource modell expect at minimum 20-24GB vram.

          and its not only AI/LLM people of course also expect Blender and other software to run well.

          Blender also loves a lot of Vram.
          Phantom circuit Sequence Reducer Dyslexia

          Comment


          • #6
            Originally posted by timofonic View Post
            AMD!!!

            GPU Computing for all RDNA4 family, derivated and sucessors. Please.
            LMAO. That's pretty funny.

            Comment


            • #7
              Originally posted by qarium View Post

              you are right its a miracle why AMD does not understand that people expect to tun their LLM/AI on their GPU...
              also vram LLM need much vRAM this means the AMD radeon 8800XT should have 32gb VRAM at minimum.
              my Vega64 is more or less obsolete because it can only run 7billion parameter AI... because it only has 8GB vram...
              so even 16gb vram is not enough the Mistral 8x20B opensource modell expect at minimum 20-24GB vram.

              and its not only AI/LLM people of course also expect Blender and other software to run well.

              Blender also loves a lot of Vram.
              LOL except Blender doesn't love AMD and AMD doesn't love Blender - so mutual. Even if AMD put 48gb or more on their gpu, it will still suck eggs in Blender.

              Comment


              • #8
                Originally posted by Panix View Post
                LOL except Blender doesn't love AMD and AMD doesn't love Blender - so mutual. Even if AMD put 48gb or more on their gpu, it will still suck eggs in Blender.
                benchmarks already show that vram is king in blender even on nvidia cards. even with similar compute power cards with 12gb vram do better than cards with 10gb vram. and cards with 24gb vram do better than cards with 8gb vram.

                this has exactly zero to do with AMD... by the way i have a AMD PRO w7900 with 48gB vram you see vram size sell cards if you want it or not.
                Phantom circuit Sequence Reducer Dyslexia

                Comment


                • #9
                  Originally posted by qarium View Post

                  benchmarks already show that vram is king in blender even on nvidia cards. even with similar compute power cards with 12gb vram do better than cards with 10gb vram. and cards with 24gb vram do better than cards with 8gb vram.

                  this has exactly zero to do with AMD... by the way i have a AMD PRO w7900 with 48gB vram you see vram size sell cards if you want it or not.
                  So? It still sucks in Blender despite more vram and it being expensive...for what?

                  Comment


                  • #10
                    Originally posted by Panix View Post
                    So? It still sucks in Blender despite more vram and it being expensive...for what?
                    well just compare 48GB vram cards with other 48GB vram cards then you will discover that the AMD card costs 3300€ and the Nvidia cards 8000€...

                    if you translate "Sucks in Blender" into: Sucks per Dollar then AMD wins big.

                    but sure you can freely and openly spend the 8000€ and claim you are superior.
                    Phantom circuit Sequence Reducer Dyslexia

                    Comment

                    Working...
                    X