News - AMD presentation reveals more details on Radeon HD 6900 architecture

Quote:

---

Fundamental changes to the Evergreen architecture hint that Cayman might be a different beast.

---

Read more.

*sigh* confused now. Are we really going to have 3 different architectures in one series? We know the 68x0 is a tweaked 5800 architecture, so assuming the 69x0s are this new architecture with 4 'smart' shaders per block (and a 16:1 shading:texturing ratio) and the 67x0 are going to be rebadged 57x0s, it would appear we are. Come on ATI, I know you've merged with AMD but there's no need to go that far to the "green" side... ;)

Quote:

---

Originally Posted by scaryjim

*sigh* confused now. Are we really going to have 3 different architectures in one series? We know the 68x0 is a tweaked 5800 architecture, so assuming the 69x0s are this new architecture with 4 'smart' shaders per block (and a 16:1 shading:texturing ratio) and the 67x0 are going to be rebadged 57x0s, it would appear we are. Come on ATI, I know you've merged with AMD but there's no need to go that far to the "green" side... ;)

---

I'd rather have different architectures under the same series than have different series for the same architecture ;)

But in this case they could probably argue that the core architecture isn't that different, it's just wrapped slightly differently.

The only people saying the 5770 will be rebadged, well, started on Fud I believe.

Think of it this way, take Barts core, which offers more performance per mm2, or take Cayman which offers more performance per mm2.

Now with two updated more efficient architectures meaning they can get the same performance out of a smaller core, as the wafers cost the same no matter what you make, the smaller the core the cheaper it is and the more you can make.

So the options presented, are continuing to make a 177mm2(iirc, somewhere around 170mm2 anyway) 5770's and call then 6770's, sell them for a reduced cost as their performance vs top of the line is reduced, less profit.

Use the barts architecture, with less shaders, get the same performance for around 35% die size space saving as seen on Barts, sell a card, thats 35% smaller for the same cost, and make 35% more profit?

Or use a cut down caymans, we aren't sure how much extra performance per mm2 it offers, and probably get more than 35% reduction in size for the same performance, and therefore make 40% more profit.

But Fud/Nvidia biased sites would have you believe AMD will choose to continue making a larger less profitable chip than necessary just for the hell of it? Sounds retarded, yup.


As far as I know the only reason someone decided it was to be rebraded was some slides show 6970/50, 6870/50, and an arrow that shows the 5770 being sold throughout Q4. It doesn't have 6770 listed anywhere ahead of that arrow, all the slide means is the 5770 is the midrange card through Q4 this year, meaning a replacement is due in Q1, AMD have not once said this will be a rebrand and I can't think of a logical reason for them to want to do so.

I'd say of the article, its a little off to say the 6970 is like 2 cut down 6870's, they share very little in common architecturally seeing as the large portion of the core is shaders, and the shaders are completely different.

http://www.hexus.net/content/item.php?item=20289&page=3

The only thing thats different in reality, is the graphics engine up the top, the shader blocks were always in two separate groups, the global data share was always there, basically, its easier to load balance 800 shaders worth of throughput in one engine and have two engines, than load balance 1600 shaders in one engine. As shader number increases this was almost always certain to happen.

When you compare the 6870 and the 6970, its not even remotely like theres two 6870's in there.

http://www.hexus.net/content/item.php?item=27053&page=2

Infact, there's your 6870, theres even less change between the 6870, the two shader blocks, the 2 separate thread dispatchers, the only fundamental(from terms of how data moves through the gpu) difference is the graphics engine has been split in two, and again with the 6870 having less shaders than a 5870, theres no real need to have two engines, with the 6970 and an increasing number of shaders, its pretty much expected. Considering it even has the same number of rops and general bits and bobs its even odder to call it "2 6870's".

Then when you actually look at each individual piece of logic, there will be many differences, shader's have changed drastically, rops/tmu's/thread dispatcher are likely all slightly different. The seeming off chip buffering we heard of being the two extra rows of shaders below the main blocks.

Quote:

---

Originally Posted by drunkenmaster

... As far as I know the only reason someone decided it was to be rebraded was some slides show 6970/50, 6870/50, and an arrow that shows the 5770 being sold throughout Q4. ...

---

Well, there's always the rapidly pulled Catalyst PC Vendor ID List which listed the lower end 6000s as Evergreen. Sure, it could just be a slip of the copy & paste, but coming from AMDs own website it's rather suggestive.

Add to that the possibility that the profit margins they're already seeing on the lower end 5000s are sufficiently high that they can easily absorb a minor price snip, that they're also already performance competitive with their nearest nvidia rival (the GTS450), and they already sit at a price point that no other AMD card encroaches on, and it's possible that the cost of taping out and testing new silicon for the lower range isn't worth the potential increase in profit that a smaller die might bring. Why spend money creating a new design when your existing one fits the market requirements perfectly?