AMD's HD8000 family is called "Oland"

[Wesley]

So while this leak happened a while ago I've been waiting for someone, somewhere to come and pronounce it wrong and untrue or factually incorrect (or even a "No comment" from AMD would be nice), but no-one's done that. It can only be true then! Or at least partly true, but forgive my slight fanboyishness when I see details about new Radeon cards. AMD's next GPU family to expand on the Graphics Core next architecture is codenamed "Oland" and will be manufactured using the 28nm process.

Oland promises to be in every way, shape and form, better than the current Pitcairn family which is undergoing a serious round of price cuts to compete with Nvidia. Hit the jump for my analysis on the purported specs.

Linky

[.exe]

Got anything on the 89xx cards, Wesley? Anyway, the numbers look very encouraging, although how this translates into real-world performance remains to be seen.

I've been looking at getting another Gpu, but with the product lifecycle of current cards already halfway over, and my card still performing admirably in games, there has been little point, however tempting.

But with the 8xxx cards on the horizon, I think it would be a good time sometime next year to jump in and acquire something shiny and new, especially since the next generation of graphics engines will be more demanding, and they can't be all that far off.

PS - is AMD really struggling to keep up with Nvidia? I haven't done the research (sorry) but plenty of people seem to think the Ghz edition of AMD cards are in fact supeior to Nvidia's offerings. (With the right drivers at least.)

[Wesley]

I haven't seen anything yet on the 89xx series but it looks like you could take the same percentage increases on the HD8870 and try apply them to the HD7970 and HD7950. You land up with something like:

HD7970
Transistors: 5.2 billion
Die Size: 463mm squared
Clocks: Might change, might not change, default is 925MHz on the HD7970 first generation, second-gen is 1GHz core
Compute SP: very sketchy, but might be around 6 TFLOPS, which puts it in the range of a GTX690
Compute DP: I still have no idea what default the HD7970 is, but its around 1.2TFLOPS if some benchmarks are to be believed. This might go up to 2TFLOPS? I really have no idea, since its not very well reported, but AMD's always limited compute performance to 1/3 of single-precision, whereas Nvidia limits to 1/8 of potential to not overtake Quadro cards
Texture fillrate: should be around 195 GT/s, up from 118.4
Pixel fillrate: won't change all that much, probably up to 32 GP/s
RAM Bandwidth: probably around 330 GB/s
TDP: Won't be higher than 250W

To put that into perspective for you...

[.exe]

What I find interesting is that the texture fill rate of the 8870 is damn close to that of the 680 GTX, and superior to the 7970. That's impressive.