![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
Our good friend Fud (yeah, I know how to spell his name!) over at the Inquirer has reported that, "ATI's R520 has eight Shader units". Ok, thanks for that Fud, now is that pixel shader units or vertex shader units? Do you even know the difference? Whenever I read his articles I always get the strong impression that he doesn't have a clue what he's talking about!!
Given the number of 'shader units' he mentioned, I'm certain he's talking about vertex shader units in R520. 8 vertex processing units sounds about right and follows a nice progression from the previous couple of generations - 9800XT (4) -> X800 (6) -> R520 (8). No big surprise there then.
The questions of clockspeed, pixel pipelines and video memory subsystem are a bit harder to guess though. I think we'll see either 24 (probably more likely) or 32 pixel pipes in R520, think of the pixel power that 32 pipes would deliver! The 90nm low-k TSMC process should be capable of some pretty impressive transistor performance given how good their 130nm low-k process was. I guess a clockspeed between 500-600MHz should be a safe bet even with an assload of transistors (300 million +). I hope TSMC don't hit the same degree of current leakage that Intel did with their 90nm process, I don't want to see a dustbuster-style cooler on R520!!
One interesting area of R520 design is the memory subsystem. I'm sure the current high-end generation of cards is bottlenecked to a certain extent by memory bandwidth, even with 256-bit 1.2GHz GDDR3. Given the raw pixel horsepower of the next generation R520, what are they going to do with the video memory to prevent a serious bottleneck? Are we ready for a 512-bit memory interface in these cards yet? 3DLabs has a video card with a 512-bit memory interface but it's not a card for gamers. I'm sure we'll see a 512MB flavour of R520 at some point but that won't help the bandwidth issue. What else will they do? Maybe some EDRAM or something (which is something R500 is supposed to include)?
Let the spekulayshun begin.
Given the number of 'shader units' he mentioned, I'm certain he's talking about vertex shader units in R520. 8 vertex processing units sounds about right and follows a nice progression from the previous couple of generations - 9800XT (4) -> X800 (6) -> R520 (8). No big surprise there then.
The questions of clockspeed, pixel pipelines and video memory subsystem are a bit harder to guess though. I think we'll see either 24 (probably more likely) or 32 pixel pipes in R520, think of the pixel power that 32 pipes would deliver! The 90nm low-k TSMC process should be capable of some pretty impressive transistor performance given how good their 130nm low-k process was. I guess a clockspeed between 500-600MHz should be a safe bet even with an assload of transistors (300 million +). I hope TSMC don't hit the same degree of current leakage that Intel did with their 90nm process, I don't want to see a dustbuster-style cooler on R520!!
One interesting area of R520 design is the memory subsystem. I'm sure the current high-end generation of cards is bottlenecked to a certain extent by memory bandwidth, even with 256-bit 1.2GHz GDDR3. Given the raw pixel horsepower of the next generation R520, what are they going to do with the video memory to prevent a serious bottleneck? Are we ready for a 512-bit memory interface in these cards yet? 3DLabs has a video card with a 512-bit memory interface but it's not a card for gamers. I'm sure we'll see a 512MB flavour of R520 at some point but that won't help the bandwidth issue. What else will they do? Maybe some EDRAM or something (which is something R500 is supposed to include)?
Let the spekulayshun begin.