"Wii U GPU is several generations ahead of current gen" Shin'en

[Pemalite]

Netyaroze said:

Consoles have advantages through their system architecture, it wasn't done like that before. Decades ? Shared memory =/= unified memory. APUs are limited because of DDR3 bandwith, and can't use their strengths on a PC yet. Yes PC is and always will be stronger. But efficency will play a bigger role for PC, because of rising manufacturing costs. I don't need a history lesson. I played my first PC game on a x286.

Also IGPs have to adress the ram seperatly and CPUs can not use the data, it must be copied. One for CPU one for GPU. They can not share identical data even if the data is the same. Kaveri will do that but its not out yet, and aslong as the ram limits the performance its not that interesting.

PCI is a bottleneck and even with 4.0 alot of the bandwith will go to redundant tasks, also GPU CPU cycles will. No unified system memory will do that. Right now its no big deal but it will be in the future once the code changes that GPUs run. CPU GPU communication will be more important. PC has to throw money and ressources at everything and excells at everything. But with costlier shrinks in the future PC has to change or costs will rise. 

Consoles got more PC like and PC will get more console like. Its weird to assume Sony/MS Engineers didn't try to improve the PC Architecture in the only way they can. Afterall in terms of raw performance they will always be behind and in terms of Chip architecture they can be at best just a couple of months ahead, like 360 once. 

Unified Memory Access or UMA, HAS been around for decades, the only thing you are getting into is the semantics of how the memory is allocated and how the devices talk to each other.
Even on old Intel graphics that shared system memory like the i830 and i915, they could talk to the CPU, make some writes in the CPU's cache and vice versa, the Intel drivers actually had specific commands recognised by the drivers kernel in order for the CPU and GPU to talk to each other directly via system memory.

Heck, on some old Intel IGP's the CPU would actually perform some graphics functions such as TnL, do the processing for that then send it to System memory for the graphics processor to do it's thing.

As for APU's, they aren't that flexible, the idea of AMD's eventual goal with fusion was to have the CPU and GPU handle every-day tasks that each are best suited for, for example, the GPU could handle the floating point math for the CPU as it's well adapt at doing so due to it's parallel nature, unfortunatly we still aren't there yet, Trinity/Kabini was another step forward to that eventual goal.

Also, PCI is going to be a bottleneck, it's 133MB/s. PCI-E however doesn't have such restrictions, it's not holding anything back except in some extreme pure compute scenarios.

As for PC's getting more PC like, AMD started developing APU's for the PC, not for consoles first, so it's purely consoles are getting more PC-like at this stage.
AMD outlined in several whitepapers on what it's Fusion initiative was going to head towards when they bought out ATI, long before these consoles were even in the design phase, that's for sure.

walsufnir said:

I have to disagree a bit because you couldn't use the memory for gpu or cpu as you wish with igp. And decades is also exaggerating. It was shared memory but it wasn't unified memory until APUs were made.

The only difference between an APU and IGP is that an APU has the graphics processor on-die with the processor, that's pretty much it, the APU still sets aside some dedicated pages in Ram to use for itself.

Intel IGP's for example would dynamically take memory from system memory to use for itself depending on load, even when it was integrated into the motherboard chipset, some motherboards would even allow you to adjust it manually.
The only difference between the two platforms is the developer makes the choice not the driver or the BIOS or the user to how much memory is allocated to each processor.