Announcement

Collapse
No announcement yet.

Radeon HD 6000 Detailed Specs

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

  • #31
    Originally posted by FunkyRider View Post
    Barts XT card leaked!
    256-bit layout, dual 6 Pin, dual slot
    Product name: Radeon 6770?
    BTW, Cooler is a decoy, real one should follow the white outline drawn on the PCB forming a dual slot external exhaust style
    the product name of this card is 6870 and the chip is a 5770 class chip with 256bit interface

    the highend one will be a single gpu 6990

    Comment


    • #32
      in my point of view the 6870's spec is:

      960 Shader-cores on 850 mhz and 4,2 GHz 205Gb/s vram on a 256bit interface
      the TDP is 150watt

      the 6850's spec:
      725 MHz 800 Shader units and 4 GHz vram 256-Bit-Interface
      TDP ~100watt

      the 6990 will have 1920 shader cores

      Comment


      • #33
        the smaler 800-960 shader card is the 6850/6870:

        http://www.computerbase.de/news/hard...-hd-6850-6870/

        Comment


        • #34
          The 68xx series will be a great replacement for the 58xx series.

          And then AMD pulls out the big gun.

          Comment


          • #35
            Originally posted by d2kx View Post
            The 68xx series will be a great replacement for the 58xx series.

            And then AMD pulls out the big gun.
            amd makes clear with the 6870 thats nvidia won in the past i tell you why:

            nvidia uses 4D shaders back in the geforce 8800 no one uses the 5D shader of the hd2900-5870 series in software so all software was optimated for nvidia 4D shaders now amd give up the fight and cut of the 5D shaders to nvidia like 4D shaders.

            nvidia uses 1 triangle per pixel tesselation unit logik in the gtx480 and now amd 6870 uses also 1 triangle per pixel tesselation logik.

            nvidia uses an 2 level cache architecture in the geforce8800 and openCL makes this to an GOD level standart thats why a hd2900 and hd3870 cant run any openCL software the hd4870 brings 1 of 2 nvidia caches and emulate the other in vram to bring a poor openCL suppot thats because openCL is just a copy of nvidia cuda and now the hd5870 brings the second cache in hardware and amd needs the 6870 to brings the full openCL-nvidia speed.


            should i be happy about a 100% nvidia design 6870 card ?

            nvidia hardware design now but no one can use this nvidia technique because physX don't work on amd cards and the nvidia unix video acceleration also don'T work on the amd hardware.


            amd really needs to fight this all dump humans follow nvidia on all cost.

            mybe "AMD Techdemo Mecha Rampage: Bullet Physx over Direct Compute" helps a bit.

            Comment


            • #36
              Originally posted by Qaridarium View Post
              amd makes clear with the 6870 thats nvidia won in the past i tell you why:
              You forgot:
              - HDMI 1.4 in next HD 6000 series, currently available elsewhere
              - Probably more codecs supported in UVD3: symbols in the library mention MPEG-2 VLD and MPEG-4 ASP (DivX) at VLD level too. That's already available on NVIDIA chips

              Anyway, everybody was reporting that HD 6000 series was just a slight upgrade to HD 5000 series. The only thing interesting coming next from AMD is Ontario & Bulldozer processors. Just hope this comes in time with correct support on Linux.

              Comment


              • #37
                Originally posted by gbeauche View Post
                You forgot:
                - HDMI 1.4 in next HD 6000 series, currently available elsewhere
                - Probably more codecs supported in UVD3: symbols in the library mention MPEG-2 VLD and MPEG-4 ASP (DivX) at VLD level too. That's already available on NVIDIA chips

                Anyway, everybody was reporting that HD 6000 series was just a slight upgrade to HD 5000 series. The only thing interesting coming next from AMD is Ontario & Bulldozer processors. Just hope this comes in time with correct support on Linux.
                i don't talk about hdmi because nvidia do not won anything on hdmi because there was no nvidia only standart.

                and the more codex support is a fake no one need's a mpeg2 codex acceleration on an modern system because the cpu is to fast for an fucking mpeg2 vid.

                and nvidia do not won on the video side because there is no nvidia only codex.

                divx yes nice but still no WebM/dirac support

                Comment


                • #38
                  Originally posted by Qaridarium View Post
                  amd makes clear with the 6870 thats nvidia won in the past i tell you why:

                  should i be happy about a 100% nvidia design 6870 card ?
                  There's nothing wrong in adapting one's product to suit the market, if you can't change the market or if there is a better alternative. Persisting on doing things that don't gather much interest would be much worse. Like if AMD CPUs only had 3DNOW and no SSE that has proven to be much more popular. Or if Intel didn't add L3 cache to its cpus just because that was already done by AMD and other cpu makers before.

                  Comment


                  • #39
                    The daisy-chaining of DisplayPort outputs is somewhat interesting too.

                    Now if only there were cheap DP monitors, and that feature prob needs new cables as well.

                    Comment


                    • #40
                      Originally posted by Qaridarium View Post
                      amd makes clear with the 6870 thats nvidia won in the past i tell you why:

                      nvidia uses 4D shaders back in the geforce 8800 no one uses the 5D shader of the hd2900-5870 series in software so all software was optimated for nvidia 4D shaders now amd give up the fight and cut of the 5D shaders to nvidia like 4D shaders.
                      I was under the impression that the other guys used scalar processors (aka "1D"), not 4-wide or 5-wide VLIW ("4D" / "5D").

                      Comment


                      • #41
                        Full photo gallery of HD68X0-graphics cards: http://plaza.fi/muropaketti/salassap...6870#kommentit

                        From Finnish HW-website AMD will release new HD6850-cards on Friday!

                        ps. Sry bad english

                        Comment


                        • #42
                          Originally posted by bridgman View Post
                          I was under the impression that the other guys used scalar processors (aka "1D"), not 4-wide or 5-wide VLIW ("4D" / "5D").
                          ok thank you. ok i'm wrong on that part

                          good to read your post here.

                          i think the nvidia-focused games are the source of the bad 5D shader usages in games and the basic point of my writing is amd cuts of his 5D shders because of nvidia not because its the better way its just more the nvidia way.

                          Comment


                          • #43
                            Originally posted by devius View Post
                            There's nothing wrong in adapting one's product to suit the market, if you can't change the market or if there is a better alternative. Persisting on doing things that don't gather much interest would be much worse. Like if AMD CPUs only had 3DNOW and no SSE that has proven to be much more popular. Or if Intel didn't add L3 cache to its cpus just because that was already done by AMD and other cpu makers before.
                            right there is nothing wrong.

                            its just a shame that amd allways lose this kind of fight

                            back in 1998 amd ad SIMD 3dnow unit... intel don't get the 3dnow no they do there own SIMD unit called SSE and go on the market 1 year later

                            now amd drop 3dnow and payes money to inte for SSE

                            same on the GPU market the r600 architecture was 5D awesome and nvidia shot the market down with an crappy 1D architecture.

                            best exampel is the tesselation unit ati do have this kind of technik back in time of the 'radeon 8500' and now 10 years later nvdia benefit from this ati techniue

                            Comment


                            • #44
                              Originally posted by Qaridarium View Post
                              back in 1998 amd ad SIMD 3dnow unit... intel don't get the 3dnow no they do there own SIMD unit called SSE and go on the market 1 year later
                              3dnow is an ISA extension, not a new functional unit. SSE was a new ISA and was implemented with a new functional unit. Anyway, same story again with SSE5/AVX. AMD invented the REX prefix, why didn't they think about a more sensible prefix, as AVX's VEX prefix? AVX is good for x86 but a vast copy of VMX or VSX.

                              now amd drop 3dnow and payes money to inte for SSE
                              I wouldn't be surprised if the other way around happens too.

                              same on the GPU market the r600 architecture was 5D awesome and nvidia shot the market down with an crappy 1D architecture.
                              I am not sure the NVIDIA architecture is fully scalar. Some recent patents mention SIMD. Anyway, nobody can really be sure.

                              Comment


                              • #45
                                Originally posted by gbeauche View Post
                                I am not sure the NVIDIA architecture is fully scalar. Some recent patents mention SIMD. Anyway, nobody can really be sure.
                                Yeah, the distinction here is between SIMD/scalar and SIMD/vector or SIMD/VLIW. Most GPUs process multiple pixels or vertices in parallel (SIMD), but some work on a single component of each pixel/vertex per instruction (scalar) while others work on >1 component per instruction (typically 4-vector or VLIW).

                                Comment

                                Working...
                                X