Announcement

Collapse
No announcement yet.

Considering a new GPU soon. How's the 7700 series on Linux?

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

  • Originally posted by bridgman View Post
    This isn't particularly meaningful on its own, but if you combine the above point with the fact that shader compiler optimization is usually one of the last things to happen in an open source driver, you get the result that shader-intensive workloads on low end hardware with open source drivers are likely to bottleneck on the shader core first, while running the same workload on midrange hardware would be more likely to "bottleneck on everything at the same time" because the balance between hardware resources is different.

    If you look at results on specific benchmarks, you'll see the relative performance on low-end hardware is higher on programs with simpler shaders. It's all about the *first* bottleneck you hit - low end cards tend to hit shader limits first, high end cards tend to hit CPU limits first. Make sense ?
    Good explanation, thanks. I'd like your opinion then on the old OpenArena result, a shaderless test.
    Almost all cards were nearly exactly 50% of the blob, including the low-end card.

    I doubt the Q3-specific optimizations apply to OA, as I recall someone testing OA with the binary name changed to that of Q3 and getting more speed under the blob.

    Since on these generations the fixed function is done via emulation shaders, does that mean that these emulation shaders are crappy? Or that the hw-specific shader compiler would be able to make better use of them, and that they are good at the TGSI level?

    Comment


    • Originally posted by maldorordiscord View Post
      This catalyst-closed-source-tax is just unfair!
      It's like a buffet... if you fill up on ham we don't charge you for the roast beef

      Originally posted by curaga View Post
      Good explanation, thanks. I'd like your opinion then on the old OpenArena result, a shaderless test.
      Almost all cards were nearly exactly 50% of the blob, including the low-end card.

      I doubt the Q3-specific optimizations apply to OA, as I recall someone testing OA with the binary name changed to that of Q3 and getting more speed under the blob.

      Since on these generations the fixed function is done via emulation shaders, does that mean that these emulation shaders are crappy? Or that the hw-specific shader compiler would be able to make better use of them, and that they are good at the TGSI level?
      Good question. My first thought would be that the fixed function apps are likely to be CPU limited but don't know if anyone has looked at them recently. Once the frame rates get over a certain point it's better to spend time on something that runs more slowly...

      IIRC the ff emulation shaders are pretty simple so I wouldn't expect to be shader limited even on the low end cards. Unfortunately all of the performance numbers are ugly 1/(1/A + 1/B +... ) kinds of things so you don't get nice clean breakpoints from shader or CPU or other limiting.
      Last edited by bridgman; 07-13-2012, 11:37 AM.

      Comment


      • Unfortunately all of the performance numbers are ugly 1/(1/A + 1/B +... ) kinds of things so you don't get nice clean breakpoints from shader or CPU or other limiting.
        You'd have those numbers if the performance counters were available

        Comment


        • You'd have *more* numbers but still not enough to make it easy. Performance optimization still follows the scientific method :

          10 look through reams of information, pull theory out of <body part>
          20 implement proof-of-concept, see if it makes target app go faster
          30 excited by performance improvement on target app, test a bunch of other apps
          40 depressed by performance reduction on other apps, GOTO 10

          And yes, the scientific method pre-dates structured programming

          Comment


          • @ bridgman: Thank you for a long and enlightening post. I find it highly reassuring, that you have the time and willingness to answer questions in these fora - even though the questions may be of lesser quality than your answers.

            Originally posted by maldorordiscord View Post
            Anyway if I buy a hd7970 can I get the 100 "catalyst" closed-source-tax back if I promise I will use the open-source driver?
            Can we please get a Catalyst incompatible firmware to make sure we don't need to pay the catalyst-closed-source-tax ?
            Try to think yourself about the ramifications of your suggestions for a moment... We do not need even more division and incompatibility. (Like a "Linux Only" graphics card would be a huge sustainable success - dooooh! )

            Comment


            • Originally posted by bridgman View Post
              It's like a buffet... if you fill up on ham we don't charge you for the roast beef
              Sure but its not a buffet!
              For a rational human the conclusion is: don't buy AMD hardware if you are a open-source driver user because you will subsidize the closed source driver because for every euro the closed source makes a higher percentage because the prices are determined by the closed source driver team and you get less features if you use the open-source driver and there is no compensation to open-source users.

              Market economy based opensource driver users hurt them self by buying AMD hardware.

              I needed long time to understand this lever.

              Now the question is what to buy to max the pressure against this evil antisocial "Lever" ?

              ARM based systems without windows support and without closed source drivers?
              Intel systems with OEM Linux on it to make sure open-source is the case?
              Stop playing games and use 3D features at all?
              Buying server hardware clusters and use software-openGL-cpu rendering?

              This evil antisocial "Lever" must definitely be stopped.

              Comment


              • Originally posted by Veto View Post
                Try to think yourself about the ramifications of your suggestions for a moment... We do not need even more division and incompatibility. (Like a "Linux Only" graphics card would be a huge sustainable success - dooooh! )
                You just don't get the point right now the open-source customers subsidize the catalyst customers!
                And only Catalyst incompatible hardware can fix this!
                Or don't buy AMD hardware !

                Comment


                • You're also saying don't buy AMD if you use the closed source driver because you will subsidize the open source driver, and that NVidia is better than AMD because they don't invest in open source.

                  Are you sure you're on the right side of the argument ?

                  Comment


                  • Originally posted by bridgman View Post
                    You're also saying don't buy AMD if you use the closed source driver because you will subsidize the open source driver, and that NVidia is better than AMD because they don't invest in open source.

                    Are you sure you're on the right side of the argument ?
                    Nvidia is the worst choice thats for sure.

                    Catalyst customers only pay (1/30) per card to the open-source people but open-source people pay (29/30) per card to the catalyst people.
                    This means that the subsidization of open source by the catalyst people do not have a relevance because its only (1/30) per card.

                    This means the market economic-policy-law force you to stop buying AMD graphic solutions and go with Intel for example.

                    I'm sure there can be a solution for this problem why not spend 100 per 7970 open-source customer to the FSF ?
                    the sum should not be too large and would be a nice gesture of fairness.

                    Comment


                    • From the numbers you use I assume you're talking about the Windows driver, not the Catalyst Linux driver, right ?

                      Maybe I'm missing some subtle nuance, but don't all the HW vendors have proprietary Windows drivers ?

                      Originally posted by maldorordiscord View Post
                      I'm sure there can be a solution for this problem why not spend 100€ per 7970 open-source customer to the FSF ?
                      the sum should not be too large and would be a nice gesture of fairness.
                      Instead of supporting open source driver development ?

                      BTW can you check your numbers ? You're suggesting we give you more money than we sell the chips for... is that one of those "only our volume allows us to stay in business" deals ?

                      You know we sell chips, not cards ?
                      Last edited by bridgman; 07-13-2012, 07:55 PM.

                      Comment

                      Working...
                      X