Announcement

Collapse
No announcement yet.

AMD Is Still Looking At The Feasibility Of Mantle On Linux

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

  • #31
    Originally posted by Kraut View Post
    18 times less CPU time for draw calls it not 18 times less CPU time for the overall driver work.

    And how do you get the idea that OpenGL could be faster then a thinner layered API like mantle? (That many still suspect to be designed to much in favouritism of the GCN architecture and may not be viable for other GPUs)
    It seems at least Nvidia can be as fast as Mantle using D3D11 instead: http://www.digitalstormonline.com/un...oost-idnum209/

    Comment


    • #32
      yada yada,

      in theory everything could be 1000 times faster, the point is, today its not as fast, and they would not invest into more improvements if they would be faster than mantle today.

      And "from nvidia" I give a shit about this retards give clear proofs, and even than, whats the difference you say mantle sucks because its amd only, if this improvements are in the nvidia driver only, and amd maybe include that in 2 years into their driver, you again have a vendor-exclusive api. ( the speedy version )

      I dont give much about some powerpoint stuff from nvidia, (yes I posted one ^^) but the point is, TODAY there is no proof except a worthless statement from a shitty company. What is then working maybe in 2 years in opengl we can talk then if that happens. But maybe mantle is than another 30% faster than today or so.

      And the its 8x faster blabla is some very basic theoretical benchmarks and there will never be a game where you get 8 times more fps or you can play with a 1 core instead of a 8core cpu because of the api.

      Comment


      • #33
        One thing is certain mantle does still need sdl or something similar for replacing dinput and dsound in linux. Cryengine and UE4 are both working for getting opengl 4+ renderer and linux support. And cryengine will have mantle too, so we might see direct comparison upcoming months.

        Comment


        • #34
          So if mantle claim 9x more DRAW CALLS you believe them without any doubt despite no game shows that improvements (BF4 has 40% in the best cases and only in CPU bound scenarios) so 9x not in overall driver work, then if Nvidia talks about Directx 11 on StarSwarm and Thief surpassing Mantle, is just bullshit.

          Anyway Nvidia Dx11 improvements that seems to surpass Mantle and actually are bounded to Nvidia didn't have too much impact over Linux gaming scenarios nor mantle does actually.

          In return Opengl has seen a nice conference with code examples and where all three vendors "AMD?s Graham Sellers", "Intel?s Tim Foley", and Nvidia's "Everitt and John McDonald" has talked about how to achieve this improvements with all the vendors, this seems to be bullshit too.

          So, let my resume this:

          1. Mantle is thinner layered API, Ok so we are going to use the same method as blackiwid and we are not going to give much in some vendors powerpoints Mantle is not necessarily thinner than other api's we don't have source code so we can't compare it, and according to in game performance is not as fast as AMD claims before it was released.

          2. Other more mature apis seems to surpass Mantle, one of them one of them being supported but all vendors and being OpenSource.

          3. People don't give a shit cause for some reason they expect that mantle surpass Opengl in Cryengine that is well known that it's not even a pure Opengl engine cause it uses a translation layer, is just like compare Glbind methods vs GL Texture array multidraw noDsp methods once again.


          So please just take a look at the slides here: http://blogs.nvidia.com/blog/2014/03/20/opengl-gdc2014/

          And remember that despite being hosted on Nvidia's site some of the slides are clearly made by AMD and Intel (you can see their logos on it).

          Comment


          • #35
            9x more drawcalls or 15x less driver overhead doesn't mean 900 or 1500% more FPS. That would only be the case if the overhead was 100%.

            Comment


            • #36
              Originally posted by blackout23 View Post
              9x more drawcalls or 15x less driver overhead doesn't mean 900 or 1500% more FPS. That would only be the case if the overhead was 100%.
              This.

              Comment


              • #37
                Originally posted by Vim_User View Post
                It seems at least Nvidia can be as fast as Mantle using D3D11 instead: http://www.digitalstormonline.com/un...oost-idnum209/
                Exciting news if true. I expect AMD to strike back with a faster Mantle driver. Let the API overhead wars begin! We users can only benefit here.

                ... just hope OpenGL will get Mantle like extensions asap.

                Comment


                • #38
                  Originally posted by log0 View Post
                  Exciting news if true. I expect AMD to strike back with a faster Mantle driver. Let the API overhead wars begin! We users can only benefit here.

                  ... just hope OpenGL will get Mantle like extensions asap.
                  What kind of extensions are you looking for?

                  Comment


                  • #39
                    Originally posted by log0 View Post
                    Exciting news if true. I expect AMD to strike back with a faster Mantle driver. Let the API overhead wars begin! We users can only benefit here.
                    We users would benefit more if AMD/Nvidia come together and improve one API isntead of heaving an API war, which only will hurt game/engine developers.

                    Comment


                    • #40
                      Originally posted by johnc View Post
                      What kind of extensions are you looking for?
                      I don't have access to Mantle. So I'll have to guess. Command buffers/lists, DMA and Compute queues would be a good start. Being able to setup rendering/compute and upload/update data fully asynchronously would be great in general. It is already possible to some degree, but the more control here the better.

                      Comment

                      Working...
                      X