Wall of text...
I don't have time to respond with essays, and I have no interest in arguing the toss about rumours, so let me just point out a few things:
1. I never stated in any way that Cayman will be 3 billion transistors. Read, comprehend, post... I said:
Nothing more. This is simply a property of transistor density constraints. Stop putting words in my mouth - what I said is right there for everyone to read."If it's true that it's close to 3Bn transistors, then it will be approaching the size of the Fermi die, but we don't yet know the final transistor count".
2. Try to understand the concept of scalability and how it is linked to modularity... This is what drives the revision of GPU design. Of course you could build a more modular GPU with a smaller proportion of control logic, but what the hell would be the point if it can't scale up as well as the previous design?!
3. That Cayman has 3 RPEs which are near-identical to those in Barts is something that I was told by a colleague who has some contact with AMD. It is also hinted at by this AMD slide (the blurred "number of RPEs" looks like it says "3"). A consequence of this is that the ratio of ROPs to SPs is increased, just like it was with Barts. This is logical, since AMDs own testing showed that shifting the balance to ROPs from shaders gave better performance (anandtech). But all this is rumour and we will have to wait and see what the spec of the final design is.
4. By your own speculation, Cayman will have 1536 shaders (less than Cypress), and yet will have a larger number of transisors. Less shaders, more transistors... So, by your own speculation, the ratio of transistors used in control logic to those used for shaders will increase... Which is precisely the point I was making in post 14!
![Confused :confused: :confused:](/styles/default/xenforo/vbSmilies/Normal/confused.gif)
Look, you are clearly very sensitive about all this stuff, but if you want further discussion on this topic then please make short, concise and precise points, instead of meandering back and forth. I don't have time to read a series of 2000 word essays, and I certainly don't have the time to respond in kind. Yesterday I had a half hour of spare downtime which I spent trying to explain the concept of how modularity relates to scalability, but unfortunately that is not the norm for me these days.
Last edited: