AMD: R9xx Speculation

[Love_In_Rio]

the r300 launched before dx 9 and offered extremely playable framerates on dx 9 games which no first gen dx card has managed to accomplish. It also brought playable fsaa to new games instead of just older games.


So what can cayman do ?

Current cards already play dx 11 games at playable framerates. Perhaps Eyefinity at high fsaa amounts in a single gpu? I don't really know if thats R300 level tho.

The problem with the current dx 11 games is that minimun frame rates are far from the medium fps.

 

[Mize]

Virtually all current games are GPU limited at high resolution and AF/AA so Cayman can allow for higher resolutions with higher AF and AA and that will be great, but it's not going to be the mind-blowing change that the 9700 Pro was where you could finally play games with both at playable framerates for the first time.

 

[Man from Atlantis]

so where is the EDRAM ?

 

[Mize]

Anyone have the rest of the Chiphell leak?

 

[fellix]

I hope at least AMD would keep the Cypress' high-perf GDDR5 controller for Cayman.

 

[3dilettante]

Holy Jesus! That chip is BIG!
Wasnt expecting something so big coming from AMD
I dont believe in Antilles being 2x Cayman, with a big die like that...

What are you going by from the pics?
Not the size of the grid for the package, I hope.

 

[Jawed]

Holy Jesus! That chip is BIG!

Really? You see a chip size indicated?

Overlaying the board schematic upon a picture of HD6870 indicates that the substrate is the same size for both.

 

[mao5]

Anyone have the rest of the Chiphell leak?

chiphell copied these pictures from bbs.expreview.com, then put their logo on these pics, making u guys believe chiphell is the source.

 

[fellix]

The BGA pattern is very similar to the Cypress' contact array. Pin-compatible? Probably not exactly, if the TDP is so sky-high.

 

[mao5]

these pics maybe fake,cause AMD already told us in the 6800 series pdf that 6970 has a off-die buffer.

 

[Man from Atlantis]

these pics maybe fake,cause AMD already told us in the 6800 series pdf that 6970 has a off-die buffer.

sure it is off die but maybe in same package ?

 

[fellix]

these pics maybe fake,cause AMD already told us in the 6800 series pdf that 6970 has a off-die buffer.

Yes, there was a spec like that on a leaked slide few days ago, but it was promptly removed/edited by AMD.

 

[3dilettante]

I think it said off-die buffering, which could just mean writing intermediate results to memory. That doesn't require a special chip.

 

[Mize]

It has 1x 8-pin power and 1x 6-pin power so it's Cayman or Fermi, no?

 

[Malo]

It has 1x 8-pin power and 1x 6-pin power so it's Cayman or Fermi, no?

Fermi doesn't have MDP's though

 

[Harison]

IF AMD promotes Cayman as "new R300", they must be smokin' really good stuff

First of all, its Fudo, take whatever he says with a train of salt

Still I'm pretty sure Cayman will be pretty good chip, and if GTX580 wont be something much better than GF100b, we can expect another year of AMD domination.

 

[Jawed]

I think it said off-die buffering, which could just mean writing intermediate results to memory. That doesn't require a special chip.

Yes the curious thing is that this description of off-die buffering was associated specifically with tessellation. It seems that off-die buffering of tessellation data is an "improvement" on keeping that data in the SIMDs (seemingly in LDS?). Maybe that's because it's easier to share it across the chip?

Can't say I like the sound of this, though - it might be an improvement, but ugh, cached like in Fermi seems preferable. And 2x the geometry performance, if true, sounds like no real improvement.

 

[Alexko]

Yes the curious thing is that this description of off-die buffering was associated specifically with tessellation. It seems that off-die buffering of tessellation data is an "improvement" on keeping that data in the SIMDs (seemingly in LDS?). Maybe that's because it's easier to share it across the chip?

Could it be about capacity?

Can't say I like the sound of this, though - it might be an improvement, but ugh, cached like in Fermi seems preferable. And 2x the geometry performance, if true, sounds like no real improvement.

How is 2× not an improvement? Especially since that's probably 2 × setup rate, on top of improvements found in Barts.

 

[Squilliam]

Interesting these slides say 2GB frame buffer whereas previous ones said 1GB. Any chance we're seeing the first major consumer level implementation of 2Gb GDDR5 modules here?

 

[Jawed]

Could it be about capacity?

Yes, that sounds plausible.

How is 2× not an improvement? Especially since that's probably 2 × setup rate, on top of improvements found in Barts.

Cos that's merely catching up to where NVidia has been for a while, at best.