Announcement

Collapse
No announcement yet.

Blender 2.82's NVIDIA OptiX Support Is Performing Very Well

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

  • Blender 2.82's NVIDIA OptiX Support Is Performing Very Well

    Phoronix: Blender 2.82's NVIDIA OptiX Support Is Performing Very Well

    Continuing on with our Blender 2.82 benchmarking for this open-source 3D modeling software update that debuted last month with numerous improvements, here are some fresh benchmarks of the CUDA and OptiX back-ends for NVIDIA GPU acceleration...

    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
    Why didn't you test the barbershop scene?

    Ahh, right...

    Comment


    • #3
      #BlenderFavorsNVIDIA

      Comment


      • #4
        Originally posted by tildearrow View Post
        #BlenderFavorsNVIDIA
        #BlenderJustTriesToPerformTheBestOnEachPlatformUsi ngAvailableTechnologiesAndFrameworksAndDoesItVeryW ell

        Comment


        • #5
          Originally posted by Dunexus View Post

          #BlenderJustTriesToPerformTheBestOnEachPlatformUsi ngAvailableTechnologiesAndFrameworksAndDoesItVeryW ell
          No, it doesn't. The initial 2.81 build slowed AMD GPUs by a whopping 50% and was partially fixed because people complained. Developers said they wouldn't waste their time fixing that any further.

          It is honestly discouraging seeing how the Blender development team doesn't have the patience with OpenCL that we have with their constant fuck ups.

          Comment


          • #6
            I'm new to Blender, but I found that on my fairly modest system (9700K + RTX2070) it was still faster to use CUDA with CPU+GPU instead of Optix.

            Is that an expected result? Does that mean Optix rendering is really only for folks with multiple high-end GPUs in their system? (which is probably not uncommon in the rendering business, I guess)

            Comment


            • #7
              Originally posted by Imout0 View Post

              No, it doesn't. The initial 2.81 build slowed AMD GPUs by a whopping 50% and was partially fixed because people complained. Developers said they wouldn't waste their time fixing that any further.

              It is honestly discouraging seeing how the Blender development team doesn't have the patience with OpenCL that we have with their constant fuck ups.
              At least nvidia's tools/libs work. No one is stopping AMD from making Blender run better with their GPUs. At this point it's fairly obvious that AMD needs to invest a
              lot more in the software side of things to make their hardware more appealing. Intel and ARM understand this (with their timely contributions to toolchain infrastructure, ...),
              MIPS didn't (or couldn't) and they're more or less dead now.

              Comment


              • #8
                Originally posted by mlau View Post

                At least nvidia's tools/libs work. No one is stopping AMD from making Blender run better with their GPUs. At this point it's fairly obvious that AMD needs to invest a
                lot more in the software side of things to make their hardware more appealing. Intel and ARM understand this (with their timely contributions to toolchain infrastructure, ...),
                MIPS didn't (or couldn't) and they're more or less dead now.
                Same could be said about Blender. Every other piece of 3D software works better than Blender and here we are blaming a piece of free software for being difficult. Until very recently Blender was an utter nightmare to work with and after getting better it is still nowhere near other 3D software ease of use.

                I know people want things working but the Blender Foundation should understand that they are at this point because users stuck with them even at bad times. Throwing AMD users under the bus isn't the right way to do things and the Blender Foundation should be all about it.

                Comment


                • #9
                  Originally posted by Imout0 View Post

                  No, it doesn't. The initial 2.81 build slowed AMD GPUs by a whopping 50% and was partially fixed because people complained. Developers said they wouldn't waste their time fixing that any further.

                  It is honestly discouraging seeing how the Blender development team doesn't have the patience with OpenCL that we have with their constant fuck ups.
                  Your lack of knowledge is disturbing.

                  In 2.80, Blender devs added to Cycles the 4D nodes (noise, vornoi, etc) from the 2019 GSOC. It was always a known issue that adding these nodes would take a hit on AMD OpenCL's performance. (You can't really blame neither AMD nor blender devs for that. Blame the OpenCL spec for it). For now the problem is workarounded, but is stated that BF is working with AMD on a better solution, if any.

                  OpenCL is nice to use for parallelizable operations, but has quite a lot of limitations that CUDA doesn't have. You can blame Apple for these. That's why it took years (and some AMD employees) to have OpenCL support in Cycles, and that's why other render engines that support openCL have problems too (vray as example) and some others just refuse to support openCL (octane)

                  Comment


                  • #10
                    AMD has its own rendering engine that is already optimized for Blender and available through all major OSes, this might be the reason why AMD sucks on Blender, and this is surely a huge blame!

                    Comment

                    Working...
                    X