Shin'en's Manfred Linzner Discuss Wii U Shaders, eRam Usage

Wii U has a lot more potential than people give it, Shin'en have done a top job optimising it and they're third party. Others really need to take note!

jb said:
Wii U has a lot more potential than people give it, Shin'en have done a top job optimising it and they're third party. Others really need to take note!

Absolutely. In as far as next-gen system design goes, a distinction both the X1 & Wii U share, the new Nintendo system epitomises those engineering qualities.
Fast interconnected machines encompassed by multi-level cache regions and globalized by a huge primary cache, having deep, wide page files and minimal external RAM. Essentially putting as little on the external bus as possibke is all very next-gen engineering semantic.

Without meaning to anger Sony fans, the Wii U is actually the polar opposite in memory hierachy to the PS4 a haute x86 version of the PS3 design, with greater overall performance. Large external bus RAM that by definition is distinctly last-gen; where next-gen in these closed systems (and indeed all devices) are using modern memory hierachy.

If groups like Shi'nen identify that in hindsight, further optimization of the pipe nets them tighter code by 40% at minimum, then it suggests quite clearly that getting close to the metal is deep indeed.

Now if only they could get third parties to make some games it.

Developers that took the time to invest in the hardware and games: "Wii U is strong and awesome."
Developers that are too lazy for that: "Wii U is weak and terrible."

""are not really optimised" because they were "fast enough" and that they could be made 30-40% faster "by better pipeline usage or better hints for the shader compiler.""

Honest developers FTW! 

Seriously, I get fed up with developers who claim they cant do better on a system - even when other developers clearly have already.
If the restriction is time or budget thats absolutely fine! If its learning the hardware, not a problem.

I sometimes think a lot of them are scared to be honest, and have to pretend everything they do is "pushing it to the max!".
I cant remember the last time a developer said they could optimise their work better if they needed too

Large external bus RAM that by definition is distinctly last-gen; where next-gen in these closed systems (and indeed all devices) are using modern memory hierachy.

Exactly how is an eDRAM buffer "next gen" or "cutting edge" when the 360 has had one for 8 years?

Anyhow, choosing to declare one as "next-gen" and one as "not" is a matter of semantics and frankly irrelevant. Implying the unified memory architecture of the PS4 with 8GBs of DDR5 RAM isn't cutting edge engineering is kind of funny.

Also trying to say the PS4 is an x86 version of the PS3 is simply inaccurate. They're worlds apart in so many ways, from concept to implementation; the distinction actually starts with the RAM. The PS3 used a split RAM architecture: 256 MB for video memory, 256 MB for the Cell processor. Part of the reasoning to go unified with the PS4 was due to feedback from developers that they wanted a simplified memory system. Already this point right here shows a different mindset when developing the PS4 verses Sony's approach with the PS3.

In fact, slighting the PS4's memory setup shows general ignorance of why the Wii U and XB1 use eDRAM caches in the first place: DDR3 RAM is significantly bandwidth limited. In order to get around the fact that DDR3 isn't fast enough they have to implement complex caching schemes. What did Sony do instead of adding yet another buffer level? 

They used a vast 8GB pool of bandwidth rich GDDR5 RAM. RAM that is fast enough to the point you don't need pools of eDRAM floating around. It's a high-tech solution that developers are in love with. Developers are coming out of the woodworks to talk about what Sony has done, including folks like John Carmack who really know what they're talking about. You don't see that same level of enthusiasm for the complicated caching scheme used by XB1 and Wii U in comparison.