Gonna share a post I read in the comments from another site that I found interesting.
"Okay, let me expand on what Mr. Martin is saying,defining all the terms you might not know.
The cores on the Wii U are much more powerful than it may appear. It's an out-of-order design - meaning the processor can rearrange the order it runs things in order to get much better performance, as long as it still gets the same results. Every PC CPU since the Pentium Pro and the K5 have done this, except the ultra-weak Atoms. The Xenon in the 360 and the Cell in the PS3 are also in-order designs, so the cores spend a lot of time idling because they can't shuffle instructions around so they can utilize their processing elements effectively.
The Xenon/Cell (they're actually a very closely related design - the Xenon is a triple-core variant the main CPU of the Cell) try to compensate for it by using SMT, or Simultaneous Multithreading. Essentially, while a single program thread cannot take advantage of lulls in activity for one processing element of the core, another thread can. Intel chips call this "Hyperthreading". You rarely get twice the actual performance of a non-SMT core - you usually only get a 20%-40% improvement or so, unless you're a very odd design like the recent UltraSPARCs.
The clock speed on the Wii U is 1.24GHz. That's rather slow, relatively speaking, but you can't directly compare clock speed unless it's between two processors of the same architecture. So you can say a 4.0GHz Core 2 Quad is twice as fast as a 2.0GHz Core 2 Quad, but you can't say a 4.0GHz Fusion is twice as fast as a 2.0GHz K10. That's because, as Mr. Martin says, there's a lot that can affect IPC (Instructions Per Clock - depending on dozens of other factors, your processor may be able to run multiple instructions on each clock cycle, and may be forced to spend many cycles idling, doing nothing). And that's assuming the instructions are equal - some processor architectures like to make instructions that do a whole lot at once (for example, x86), while others like to keep them small and simple, hopefully allowing it to run them faster (ARM, for instance).
The comparison he makes is P4 versus Core. The Pentium 4 was a mixed bag - it had very high clocks, but several architectural decisions made to reach those clocks caused it to be very inefficient. It was joked, at the time, that the Pentium 4 was the fastest CPU for programs that fit entirely in cache, because every time the CPU had to go out to system memory to grab some data, the processor basically froze. And branch misprediction caused terrible stalls. To make things worse, heat and power consumption scale quadratically with clock speed - roughly speaking, if you double the clock speed, you *quadruple* the heat and power draw, not double it. It gets so bad you really can't get a processor too far beyond 4GHz without heat issues.
Intel knew this, sort of. They had kept the earlier Pentium III design around in the Pentium M, using it mainly in laptops. But eventually they figured out that it's better to make an efficient processor, rather than an inefficient one that you can throw more power at. The Core, Core 2, Core i3/i5/i7 chips are all derived from the Pentium M. They took an efficient design, and brought it fully up to modern standards. And they have kept improving it - the cores on a Sandy Bridge/Ivy Bridge processor are *extremely* powerful and extremely efficient. You can't find better cores in an x86 chip - the only ones that can beat it are in massive server-only chips like the POWER or SPARC64.
The analogy he's making is comparing the Xenon and Cell to the Pentium 4, and the Wii U processor to the first-gen Core. It makes sense - Microsoft/Sony made many compromises in the design to get impressive statistics, like high clock speed or core count. Like, for instance, making them strictly in-order processors. If Nintendo went with a more efficient design, I would be surprised if they *couldn't* outperform the competition clock-for-clock. Keeping the clock speed low was just a way to keep the small form factor and the low noise/low power design.
There is, however, another thing that makes it a "weaker" processor: the relatively weak SIMD units (Single Instruction Multiple Data). It's pretty much on par with the Wii in this. The Xenon/Cell, however, have massive SIMD capabilities. Particularly the Cell - I often call it "one core with eight SIMD units attached to it", to describe how it really isn't an 8-core design. But even the Xenon has some extremely powerful SIMD abilities. The PC also uses various SIMD elements. It started with MMX, then 3DNow!, SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, SSE4a, AVX, AVX2, and so on.
SIMD is good for doing the same thing to a lot of data. Most graphics cards use some sort of SIMD or MIMD (Multiple Instruction Multiple Data). There's a lot of use for it in some games - it's good for big open-world stuff, processing for dozens or hundreds of NPCs. It's great for big, showy physics, particularly simple particle effects with huge numbers of particles. It's good for procedurally generating textures.
But when you think of Nintendo games, you don't really think of anything like that. The only one of their games that would really benefit from SIMD would be Pikmin, and even then, not so much. So that may be why Nintendo didn't put powerful SIMD capabilities on the Wii U - they didn't need it themselves. But there's perhaps another reason.
As I mentioned earlier, most graphics processors are made of hundreds or even thousands of SIMD cores. And with GPGPU software, you can run any code you want on the graphic cores, not just shaders or other graphics-only stuff. On the PC, Nvidia uses it to run PhysX, and it's also the power behind most modern supercomputers. It could very well be that Nintendo expects most developers to put the load that would have been on the SIMD units on the GPU instead, as they have a much, much more powerful GPU than the PS3 or 360 (go look at a die photo of the Wii U processor - the graphics section looks to be about four times the size of the CPU section, which makes it roughly four times as powerful if it's on the same process node).
I suspect that's part of the reason why the ports from other consoles seem bad, but the native games like Zombi U seem much better. GPGPU was in its infancy when the last console generation launched, so you can't really do it on the 360 or PS3. Since these are very early, probably very rushed ports, they didn't go and translate much of the SIMD code into GPGPU code.
One last thing he mentions is an ARM "secure processor". On the Wii, there's an additional, ARM-based processor core (like the one probably in your phone), not on the CPU, but on the motherboard chipset. It was used mainly for downloading data and synchronizing stuff while the system was nominally "off". It's actually a rather clever idea, so I'm not surprised Nintendo did it again. One report I saw claims the ARM core is used to run the entire OS and system software, leaving the CPU just for the game to use, but that doesn't seem too likely in my opinion, especially since certain other claims in the article have been completely debunked.
So that's essentially what Mr. Martin has said, with a bit of my own commentary and analysis. If you'll permit me a bit more, I'll describe the possible long-term effects.
If all this information is true, the Wii U will be better suited than the PS3/360 to do certain types of games. This sort of processor would be much better at "intelligent" AI - more strategic, more thoughtful, less zombie-like. It would be worse-suited to things that involve massive numbers of simple enemies, and it would be worse-suited for destructible environments.
It's likely that we will see the quality of ports improve, particularly if engines get ported to it better (IIRC, Epic has not itself ported UnrealEngine to the Wii U - any UE3 games on the Wii U were ports by the game developer, which is a frightening prospect).
Given the GPU on it, we're also likely to see better-looking games on the Wii U than on the 360/PS3. At least in terms of more polygons, higher texture resolution, more particles, and so on. It depends on the developer as to whether they make it actually look *good*.
Games that absolutely depend on heavy CPU power might go unported. But once the new Microsoft/Sony consoles hit, all bets are off."
Thanks tomakerofthegameson Kotaku.
A mouthful I know lol but informative.
Log in to comment