Thursday, December 8th 2011
AMD Radeon HD 7900 ''Tahiti'' Pictured, 384-bit Memory Bus Confirmed?
A Beyond3D forum member posted a mysterious picture of two graphics cards that could very well be engineering samples of AMD's true next-generation Radeon HD 7900 "Tahiti" graphics cards. The final products most probably won't look like these, with a bare red PCB, but it does look like the reference cooler design is ready. A more important feature in that picture is the spotting of traces for at least 11 memory chips, the 12th one (not highlighted) is apparently near the PCIe slot interface. The presence of 12 memory chips gives rumors of Tahiti featuring a 384-bit wide memory interface a shot in the arm. This will be the first AMD GPU in over 5 years to feature a memory bus wider than 256-bit. The R600 Radeon HD 2900 GPU featured a 512-bit GDDR4-capable memory interface.
Sources:
Beyond3D Forums, VR-Zone
117 Comments on AMD Radeon HD 7900 ''Tahiti'' Pictured, 384-bit Memory Bus Confirmed?
If it is true they must of given tessellation a big boost as metro is over double the performance.
Besides, with the wider bus you do not need the higher clock just to get the bandwidth up.
lighting hits it hard because it uses subsuface scattering think of it this way put a flashlight under your fingers the red you see is light passing through upper layers of skin Metro uses that technique among others
I don't know if DOF is a real D3D11 feature or if it uses compute.
There's many ways in which Metro could see a huge improvement from Tahiti, but I agree that tesselation doesn't seem like one. Even the much much hgher memory bandwidth makes more sense to me.
its effective but as you can tell no current gpus are really ready for full on direct compute tasks in real time while running a game,
DOF via Direct Compute has no improvement in quality vs current methods but has an insane performance impact, essentially tho Direct Compute does allow them to add more to the render pipeline without making it longer, thats the real benefit so eventually this will become a good thing currently gpus arent powerful enough for it .
and since a 6970 stock with all settings turned on gets me around that same mark performance wise then yes at least the 6970 numbers on that chart look correct, and should the GCN be targeted like CUDA cores are towards improved mathmatical performance and super computing type tasks then yes Direct Compute effects would no longer stress the architecture, but again i still call fud on that graph in its entirety
Assuming everything does go well and they come out with decent performance I'll probably grab a 7970.
Last top end card I had was 4870 and well, it was only really midrange XD
I haven't upgraded yet, surprisingly my ddr2 based AM2+ rig is still getting the job done so far.
Granted, I don't play the most demanding of the newer games.
Depending on how my rig runs SW:TOR I may just keep this beast until it dies.
Waiting to see that AMD has up it's sleeve for the 7xxx series, and nVidia for their next GTX series.
Had more fun with a 9800 gt, and then a 3800 series crossfire set up XD