Announcement

Collapse
No announcement yet.

where to by a Opteron 6128+board

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

  • #16
    Then show an url where you can see a cpu core good scaling - preferable with the mod you talk about. From that other benchmarks you can do certain guesses like from E6600 to Q6600 there have been 19% speed increase. Buth cpus are very similar just 2 cores more. Would you interpolate then you might guess that 1 core more would give 10% more from that result. But thats not the case when you see Phenom II X3 720 to X4 920 - that's only 5%. A 5% boost for 1 core more is not visable for normal users. So the new estimation is 2 -> 3 cores + 15%, 3 -> 4 cores + 5%. A 3 core optimisation would be very logical when the engine will come out on Xbox 360. That's the same for GTA 4. You can expect that more cores do only minimally improve the speed because only completely independend things can be calculated in parallel. As soon as one result has to wait for another the thread is locked. When you have too many threads then synchronization will require more time than you gain. Luckyly there are things that scale very well - but definitely no game engines. Those require in 99% of all cases first high raw speed and in 2nd maybe 3-4 cores. So with server cpus with 2 ghz instead of > 3 ghz you can only lose any game bench.

    Comment


    • #17
      Originally posted by Kano View Post
      Then show an url where you can see a cpu core good scaling - preferable with the mod you talk about. From that other benchmarks you can do certain guesses like from E6600 to Q6600 there have been 19% speed increase. Buth cpus are very similar just 2 cores more. Would you interpolate then you might guess that 1 core more would give 10% more from that result. But thats not the case when you see Phenom II X3 720 to X4 920 - that's only 5%. A 5% boost for 1 core more is not visable for normal users. So the new estimation is 2 -> 3 cores + 15%, 3 -> 4 cores + 5%. A 3 core optimisation would be very logical when the engine will come out on Xbox 360. That's the same for GTA 4. You can expect that more cores do only minimally improve the speed because only completely independend things can be calculated in parallel. As soon as one result has to wait for another the thread is locked. When you have too many threads then synchronization will require more time than you gain. Luckyly there are things that scale very well - but definitely no game engines. Those require in 99% of all cases first high raw speed and in 2nd maybe 3-4 cores. So with server cpus with 2 ghz instead of > 3 ghz you can only lose any game bench.
      "Then show an url where you can see a cpu core good scaling ":

      http://www.bistudio.com/developers-b...ticore_en.html

      " From that other benchmarks you can do certain guesses like from E6600 to Q6600 there have been 19% speed increase. Buth cpus are very similar just 2 cores more. Would you interpolate then you might guess that 1 core more would give 10% more from that result. "

      thats only true if you test it in the same loserPC-mode scene!

      but if you test it @1500 commander AI's its diverend!


      "You can expect that more cores do only minimally improve the speed because only completely independend things can be calculated in parallel. As soon as one result has to wait for another the thread is locked. When you have too many threads then synchronization will require more time than you gain. Luckyly there are things that scale very well - but definitely no game engines. Those require in 99% of all cases first high raw speed and in 2nd maybe 3-4 cores. So with server cpus with 2 ghz instead of > 3 ghz you can only lose any game bench."

      i do not talk abaut the graphic! i talk abaut the AI! and ArmA2 do not need synchronization the AI because its a multiplayer game every human don't akt like a synchronism swimmer! only the graphical output and the coordinates of the troops need to synchronization!

      every single AI starts a new Thread! means you have over 1500 threads!

      Comment


      • #18
        The target of the AI is to find a path for the troup, so what do you think that means? Of course it has to know the position of all other troups. Logically you need shared data which contains at least the positions of all things on the map. When 1 of it is moved you have to lock the access, that means only 1 can move the same time - otherwise more than 1 could be at the same position which is not logical. You can have have as many threads as you like - on as many core you want, all others have to wait till the write access is over. Maybe you can partition your shared data a bit that this will not happen all the time, but you should really write a sample app that uses threads which are not completely independent from each other. Maybe they use more tricks now, but the AI is certainly not extremely scaling. Compile threads are much more easy, because you can compile lots of source code files which do not depend on each other first and only at the linking stage all have to be there. You example is just stupid for so many cores - it will definitely not get faster. Buy your opteron and you will see how slow it is which that game.

        Comment


        • #19
          Originally posted by Kano View Post
          The target of the AI is to find a path for the troup, so what do you think that means? Of course it has to know the position of all other troups. Logically you need shared data which contains at least the positions of all things on the map. When 1 of it is moved you have to lock the access, that means only 1 can move the same time - otherwise more than 1 could be at the same position which is not logical. You can have have as many threads as you like - on as many core you want, all others have to wait till the write access is over. Maybe you can partition your shared data a bit that this will not happen all the time, but you should really write a sample app that uses threads which are not completely independent from each other. Maybe they use more tricks now, but the AI is certainly not extremely scaling. Compile threads are much more easy, because you can compile lots of source code files which do not depend on each other first and only at the linking stage all have to be there. You example is just stupid for so many cores - it will definitely not get faster. Buy your opteron and you will see how slow it is which that game.
          "The target of the AI is to find a path for the troup,"

          not only! the AI also check for tactical managment and other stuff.

          "Of course it has to know the position of all other troups."

          no tats wrong! the AI do not cheat in arma2 the AI only know troups in view distance and only troups in valid targed podsitions if a enemy is in the gras/green the AI can't see the enemy!

          "Logically you need shared data which contains at least the positions of all things on the map."

          you only have shared data in a low range mybe 5km! but the game is 225mk²!

          ". You can have have as many threads as you like - on as many core you want, all others have to wait till the write access is over."

          i think this is wrong because the AI do not work100% thats because a human do not work 100% they wana have errors because they do not wana GOD's they wana KI played just like humans!

          its ok if a AI have a wrong data and shot in the wrong way thats because humans also shot on wrong positions! because humans are not GOD's and not 100%

          "Buy your opteron and you will see how slow it is which that game."

          you are just wrong i still have 2 Opteron systems!

          and my friend do have a 3,6ghz overclocket E6600

          and arma2 runs faster on my system! with only 4cores@2,6ghz

          in your calculate the speed-up for multi core is only 20% but he do have much more clock speed and still loses!

          his 3,6ghz dualcore system have 100% cpu usage with no AI no troups!

          in the same no troup no ai benchmark with the same hd4890 my system only have 50%-60% cpu usage..

          ok if you put troups in it and AI and you play a game then his system goes down the FPS drops hardly!

          on my system the FPS is just the same the AI do only pull the cpu vom 60% to 80% or more if we put more AI in it!

          in shorts words you are just wrong!

          Comment


          • #20
            Well it scales best from 2 to 3 cores, why do you think it will scale even more?

            Comment


            • #21
              Originally posted by Kano View Post
              Well it scales best from 2 to 3 cores, why do you think it will scale even more?
              because its not an '''xbox360''' game!


              the arma2 devs use 8core workstation systems!
              a 8 core and a 5870 2GBvram is the 'minimum' for arma2.
              the us army use arma2 to and they use 8core workstations to!

              arma2 uses cores like this: 1 core graphic 1 core streaming data (because the running game is uncompresset more than 20gb) 1core is physic 1core is networkstuff an the rest is for AI (you need a lot for highskilled AI)

              surprime comander in a patched version can also use 12 core if you wana play with 50 000 troops many many AI to calculate!


              'ARMA3 will be dx11 means real multicore graphic engine'


              why i sould buy a computer for OFP1 (singlecore) ?

              i buy a computer for arma2/arma3/surpime comander or a raytracing based game!

              i still do not play ps3/xbox360 ports on the pc! this loser games sould DIE!

              Comment


              • #22
                Several European and US vendors list socket G34 mainboards already.

                Comment


                • #23
                  Originally posted by chithanh View Post
                  Several European and US vendors list socket G34 mainboards already.
                  thank you very much.

                  the supermicro singlesocket one cost in europa 250€ in the us 200dollar! LOL!

                  Comment

                  Working...
                  X