It's like multi-quote was made just for me.
Azure? Why be concerned about researching too much? That's part of the purpose of this thread.
My guess for WiiU are still 27MB 1T-SRAM that would allow Wii BC and a framebuffer for 720p w/ 4xAA or 1080p w/o AA and 2 tablets (even 4, but this probably won't happen because of other reasons)
My guess is something around these specs but I am hoping for a higher SPU count (considering relatively low clocks). I think I read somewhere (Beyond3d?) that the GPU part will be around 1GFLOPS, but maybe this number is also for the whole system (which would still be quite above the upcoming AMD Trinity APUs (~700 MFLOPS)
With stew saying 32MB, there's really no need to guess unless they change it.
And I hope it has way more power than that. That's not even a Gamecube.
That rumor came from a Japanese website and that the kit had an AMD GPU that was "beyond 1TFLOP", but wasn't a 4890. The 4870 is the only one beyond 1TFLOP and could definitely explain early overheating. If it were a 4870 clocked down to 500Mhz, then it would be 800GFLOPS and have a fillrate of 8GP/s and 20GT/s. Compare that to other cards:
4770(RV740) - 896/12/24
4830(RV770LE) - 736/9.2/18.2
4730(RV770CE:750Mhz) - 896/6/24
4670(RV730XT) - 480/6/24
On paper it would be outperformed by some of these. It could be used to explain why Digital Foundry came to the conclusion they did.
The explanation to that is that the amount of GPU-local edram one can expect on the WiiU cannot meet the needs of deferred shading algorithms, not without some form of tiling.
What we can hope for is edram covering a 720p x 4AA or 1080p x 2AA RGBA8 + zs32 backbuffers, which anyway you look at it is 30-34MB. Hence my last 'optimistic' guess in this thread was 40MB (albeit, presented in a 6bit signed int ; )
It should. Though history knows cases of early devkits playing bad jokes on devs *cough* dual 970MP filling in for a Xenon *cough*
Ok. Yeah I remember the tiling part since that's why I wanted to learn about the calculation. I guess the (32bpp Z) is what I still haven't grasped totally. But like luigiv
And be fair, MS meant well on that. They just, well, messed up pretty badly trying to rush the launch.
3 cores dev kit, or final design?
How would he know what Nintendo plans to do in their final design?
I assume console makers give out targets, but dont give developers final designs till
late in the game due to leaks. Especially leaks to the competition.
Im sticking with 4 cores, due to it being a Power7, the adding of the extra controller, and the perception of looking like last gen Xbox 360.
Final design. Those devs/pubs are a console maker's business partners since they are using the console to sell their games. You don't do that to business partners. They tell them what the plan is for the console. Things can always change, but even when they do change you let them know. And one thing about Nintendo when looking back at how they've worked, that kind of perception won't matter because they will focus more on how it looks on screen.
What? You guys are going back to old posts now?
this thread is madness!
New leaks please
32MB of eDRAM.
You need depth for a lot of things, the least of which is being able to render the scene correctly (order of opaque objects, since alpha blended items aren't part of this, alpha test can take advantage of the z). Then there's shadowmapping, some post-processing (depth of field for example).
Which is why when I first did my own calculations I think you and blu were talking about the amount needing to be much larger to do it "properly"?
lherre's laughing at us, knowing there will be breakdowns abound once the final specs are leaked.
So, are the parts still off-the-shelf?
He said they had the ugly version of the retail case so maybe they've gotten to prototype parts. Just a guess.
As I've said previously, the only reason to use something like RV770/5830 today is because the final GPU will have the same feature set - meaning DX10.1 plus some kind of a tesselator. If this GPU would have support for DX11 then why use such old h/w as RV7x0 GPUs? Perfomance wise there are GPUs in current AMD's DX11 line up which are on par with RV770/5830 - why not use them?
My guess was that they were going to use a VLIW4 GPU, but since there are none available with a lower ALU count that they used a GPU that would be close to it.