AMD: R9xx Speculation

Have you seen this:



My guess is that Barts will slot in where the old 5770 did, maybe a little higher. If it can beat 460/1G, then it will probably be priced at $189/199 to drive NV into the gutter. If it can't beat the 460, then it will be ~$169/179.

In any case, it is going to be a <200mm^2 chip that stomps a 240mm^2 GF106 silly, and makes life very unpleasant for a 367mm^2 GF104. AMD can price it anywhere they make money and make NV's life VERY unpleasant.

-Charlie

http://semiaccurate.com/forums/showthread.php?p=71680#post71680
:LOL:


Perhaps Charlie is doing the same- adding some..... :oops:
 
Interesting.
Gone are the days of 3-way or Quad-CrossfireX in the HDx8xx line of products? Or is the single-chip high-end really moved up to the HDx9xx range?


BTW:
Tessellation Unit is still singular. So either left at it's current state or they significantly beefed it up, but from the wording it seems highly unlikely that they've taken the distributed approach Nvidia is so proud of.
 
Tessellation Unit is still singular. So either left at it's current state or they significantly beefed it up, but from the wording it seems highly unlikely that they've taken the distributed approach Nvidia is so proud of.
Well wasn't it the common understanding that it was triangle setup/rasterization that was holding the tesselator back anyway?
 
I'm quite curious about the mentioned "Image quality enhancement technology"

UniversalTruth: I think members are trying to separate fake and real leaks. It's not very helpfull to start speculations based on obvious fakes...
 
Well wasn't it the common understanding that it was triangle setup/rasterization that was holding the tesselator back anyway?

No, the limitation is in the front-end. Juniper and Cypress are able of one tri/clock. But the front-end is to slow to produce this triangle with tessellation.
 
Last edited by a moderator:
No, the limitation is in the front-end. Juniper and Cypress is able of one tri/clock. But the front-end is to slow to produce this triangle with tessellation.
I think tessellation was made slow because setup/rasterisation is the bottleneck. It's literally pointless producing small triangles faster, because they'll just fill a queue, which then grinds to a halt.

In other words, without fine-grained setup and rasterisation, increased-tessellation throughput offers no gain.
 
Even if egos were hurt, there's no way it could have affected Northern Islands. It taped out in April, just a few days after Fermi was released, according to Charlie (considering the October/November launch, it really couldn't have taped out much later). When AMD saw Fermi's tesselation performance, there was no time to do anything about it.

Well, GF100's tessellation approach is quite brute force...

While Evergreen is slow, it seems to be connected to something else than the tessellator itself or even the rasterizer/setup throughput.
 
Well, GF100's tessellation approach is quite brute force...
It's quite elegant considering slide below:
2rnzt34.png
 
Last edited by a moderator:
Barts XT should not be in the 6800 class, that would be epic failure on AMD part....i know its just business but.... the lowest goal i can accept is to call it 6790 and go for 239 bucks. Talk about GPU sloooowdown...

BTW will the HD6 series output audio and video streams separately from the connectors? It is a bit silly to always on ma AVR just to get to ma desktop with HD audio.
 
Back
Top