They said they're going to show another Agni's not running on a PC this coming e3.
Scalability has only been one part of this equation. I already talked about the pains of downporting so I'm not really sure if clinging onto "scalability" is a good thing.
Sheesh, I don't want to even go there.
When you say that the amount of Constant and Instruction cache in Latte is extraordinary, do you refer to the big pools of eDram or to other parts of the GPU?
No. I would say that its geared to modern-future engines with 720p on mind. Modern architecture and modest power (although really high efficiency).
It wasn't. But if you support those claims I'll treat you as such.
the biggest thing separating the WiiU from the Durango is the RAM. The Durango still has 5GB of faster RAM (Though IMO RAM speed isn't quite as important after a certain speed at the level we're talking about) versus the Wii U's 1. I'm working with stuff now for our workstations, and cramming as much data as we can into usable RAM now is singlehandedly the biggest task we face.
Because this is not how it works. Modern engines are for modern architectures, while doing something at 1080p requires power alone.
Bailing out, really?
It's not that simple either. The "how much" I was speaking of was in terms of ops per second, but that doesn't have to translate in a "how much" in terms of visual quality because that depends entirely on what anyone expects AND whatever new techniques can be done on the WiiU to do "more with less".
And it will probably run at 1080p60 on PS360 too. What did Origins run at?
The strange thing about that picture, which you did mention in part, is only one of the H/G duplicates seemed to be included in the UVD. It also makes Llano's seem smaller than Brazos' because of the way that picture breaks it up. I still believe the Q blocks match the one in Llano under the block I feel is like W. The block that has it's memory to one side. I think both from a fab size (I understand your other post on this, but I think H/G are way too small. 1 block is 19% of 1 Q's area), SRAM block sizes (I had counted five different sizes in Llano and Latte), and total SRAM amount perspectives it makes more sense. It would just be trying to figure out what it does in Llano.
I gotcha. I think my opinion of no MCs in Latte is why I don't see the same thing.
No, you're doing it wrong. It's BS multipliers.
I was putting this together anyway (attempting to reaffirm the discussion between Fourth and myself), so it should help answer your question at the same time. I'm referring to that large pool of SRAM within the block itself. Here is a picture of the Sequencer (UTDP) in Brazos compared to the block I think is the same in Latte. Also here is picture for the Shader Export. I rotated and flipped both blocks from Brazos to a similar position as the ones in Latte.
UTDP
Shader Export
As I understand it, it was the tessellator itself in ATi/AMD GPUs that sucked. And that it wasn't till the introduction of the 7th generation of tessellators in some of the 6000 series that they were deemed useable.
In a console environment a fixed function unit like a tessellator would ideally have little to no impact on your baseline performance (extra memory use etc... notwithstanding). But my post was about a comparison to the XB1/PS4. If the Wii-U's tessellator is as far removed from their tessellation units as earlier Radeon tessellators were from those in GCN cards, any tessellation implementation on a Wii-U port would have to significantly reduce either tessellation level or frequency - most likely both.
Good stuff, but it's still only part of the picture. GCN saw a significant improvement in tessellation scaling over HD6xxx series cards with the same poly-pushing capability and even outperformed the GTX580 which could push 6 polys/clock more.
