They could as well have gone with a really improved older one. Seems to be no problem to port from G5 to Espresso.
But (not existing) SIMD capabilities, weak FPU etc. are a problem.

Keep in mind that Intels Core and to some degree Core 2 architecture basically was an (very) improved P6 architecture chip. Thats hypothetic though because it seems that Nintendo went for a really cheap design. The MCM is what makes the CPU+GPU part expensive.

A PPC750 based CPU with the right enhancements might have done very well.

Alby_da_Wolf said: This. Don't know if anybody already pointed it out, but I'd like to add that POWER is still a very scalable architecture, IBM mainly designs high-end models currently, but its partners in the project also design models down to cheap and power thrifty single core embedded versions or even multicores where each core is a lightweight version, for high-end routers, for example, that need to execute large numbers of simple tasks. Ninty chose an evolution of a quite dated version ( not exactly definable in a single POWER or PowerPC family, it has a more dated PPC base with a few more modern POWER7 features http://en.wikipedia.org/wiki/Espresso_%28microprocessor%29 ), but, while POWER8 hadn't been launched yet and POWER7+ was maybe too recent to already have cheap enough versions (the first models launched were fast high-end server versions) when Wii U was launched, Ninty could have easily chosen a POWER7 chip scaled exactly to its needs about power, power consumption and price.

POWER is only made by IBM. PowerPC is licensed. And POWER doesn't mean that all PowerPC instructions are available.

You can't downscale the POWER chip easily. Its a super complex multi-core beast designed for multi socket. 

Also a low power consumption was not the "primary design goal" of the POWER chips.

Nintendo got the best what they could get cheap. And its fast enough as it has a clever design like asymetric caches...

fatslob-:O said:   The HD twins still have more CPU cache ... mine said: BTW: take a look at the PS4 developer presentations. The small L2 cache is hurting BIG time when CPU AND GPU are accessing the GDDR...   Link ?

I don't think that 2x 2MB L2 is a good design decission for 2 x 4 cores... . Its even more worse as each cache miss means a high latency access to GDDR. Thats why computation intensive games run better on the XOne...

The Wii U has 1 x 2 MB + 2 x 512 KB dedicated to a single core each.

 

Link? I would google for the PS4 guy (czerny or so) and presentation and bandwidth...

Lots of stuff discussed in the thread here, with statements both true and false. Its all reminiscent of the old Wintel-v-PowerPC wars Mac fans waged in the 1990s. And like then, its time to cut through the statements and sort out whats accurate.

- The Wii-U's processor (Espresso) is as a whole a tri-core PowerPC G3, functionally the same as it's slower single-core predecessors Broadway and Gekko. The Xbox One and PS4's processors are AMD Jaguar APUs - chips which combine an incredibly multi-core x64 (64-bit x86) processor with a Radeon R3 graphics processor.

- The processors Intel and AMD build based on the "x86 Architecture" are substantially different from the original 8086 from the 1980s. Although word length (# bits in a single value) and address space (RAM accessible) increased, various different improvements have been made. MMX/SSE have made data structures easier to handle, integrated FPUs have made floating point calculations possible, pipelines allowed multiple instructions to run at once, and the use of L2 caches and multi-cores have speeded up multithreaded applications. Theres way more instructions as well.

- The "RISC" architecture involved several core instructions being optimised to run really quickly (1 each cycle) and all other features implemented by combining the core instructions. Innovations as Cache, high amounts of registers, Multiple cores, and Pipelining were all ideas started as RISC. Many RISC processors (PowerPC, ARM, MIPs) however used more instructions as CISC (x86, Z80) and had to execute more instructions for the same jobs.

- Its said on here that PowerPC 970 (G5) applications can run on PowerPC 750 (G3) chips just fine. The G5 uses more instructions than the 750, has a better FPU, and various other enhancements which make software incompatible with the 970 and 7400 (G4). Notably the the Altivec unit on the G4 and G5 that the G3 lacks. In the last few years of PowerPC Macs, developers had their applications run on G4 processors by default, switching to G5 mode if it detected the processor.

- PowerPC processors are infamous for heat generation and power drain as IBM and Motorola never figured out how to make them run more efficiently. The original roadmap (plan for future chip modification) at the time Apple went to Intel had a focus on making chips run even hotter for more performance. Apple went for x86 as they needed low power processors for laptops, something impossible with future PowerPC processors. Since then lower power and cooler PowerPC processors have been released with embedded systems (game systems, cellphones, tablets etc) have been released, but they can't beat the now dominant ARM.

- PowerPC is the worst processor as far as performance per watt (a term Steve Jobs used to describe the amount of raw performance against the amount of power used - the main argument he had for x86 during Apple's transitional period) is concerned. ARM have the best performance-per-watt, but currently can't match the raw power of the x64 (64-bit x86) processors. Which is why ARM is preferred for Tablets and Cellphones, while x64 is more desirable for Computers and Game Consoles.

- It would not be possible to have a G5-based Wii-U as the PowerPC 970 architecture is no longer in development, it will have compatibility issues with Wii software, and it will have to be the size of the Xbox One to hold all the cooling apparatus (the fastest G5 PowerMacs used liquid cooling). It would also have been far far more expensive to build and thus to sell.

- Although the POWER8 as a server chip is better on raw performance than any x64 part, its considerable size, heat generation, power consumption and price would make it impossible to build into a desktop system.

- A cache is used to get around slow memory access times, allowing processors to finish an instruction without waiting for the memory to respond (its saved in the cache until it can be written out). Adding more cache can make a system faster by allowing it to store more data in the cache, but is usually only really decent for batch jobs (like rendering a shape or decrypting a chunk of data), not timing-critical code (like drawing a screen or playing a sound).

- A lot of the small input/output jobs processors used to do - floating point calculations for graphics, driving serial ports etc - are now done by extra support chips in the system.

- Developing for any architecture nowadays does not require you to dabble in instruction code as theres C/C++ compilers as standard, each usually optimised to get the best performance out of a processor. However different APIs and Drivers can cause issues, as well as the quality of the compilers and other dev tools themselves (the Wii-U has being claimed to have a terribly unreliable SDK).

So considering all this, why did Nintendo go for PowerPC instead of ARM (similar performance at lower price) or x64 (to put it on level with Xbone & PS4)? Because I believe to them it was the easiest and cheapest option - they were focused on the gamepad as a whole and went for a low cost upgrade to the Wii for the supporting hardware. They didn't spend much time on the OS either - at 1GB used constantly and horrible loading times, the Wii-U is infamous for having an OS even more performance-killing than Windows Vista, with only a fraction of the functionality or desirability.

mine said: Alby_da_Wolf said: This. Don't know if anybody already pointed it out, but I'd like to add that POWER is still a very scalable architecture, IBM mainly designs high-end models currently, but its partners in the project also design models down to cheap and power thrifty single core embedded versions or even multicores where each core is a lightweight version, for high-end routers, for example, that need to execute large numbers of simple tasks. Ninty chose an evolution of a quite dated version ( not exactly definable in a single POWER or PowerPC family, it has a more dated PPC base with a few more modern POWER7 features http://en.wikipedia.org/wiki/Espresso_%28microprocessor%29 ), but, while POWER8 hadn't been launched yet and POWER7+ was maybe too recent to already have cheap enough versions (the first models launched were fast high-end server versions) when Wii U was launched, Ninty could have easily chosen a POWER7 chip scaled exactly to its needs about power, power consumption and price. POWER is only made by IBM. PowerPC is licensed. And POWER doesn't mean that all PowerPC instructions are available. You can't downscale the POWER chip easily. Its a super complex multi-core beast designed for multi socket.  Also a low power consumption was not the "primary design goal" of the POWER chips. Nintendo got the best what they could get cheap. And its fast enough as it has a clever design like asymetric caches...

http://en.wikipedia.org/wiki/Power_Architecture

Many years ago POWER was IBM-only, while PowerPC was the first alliance IBM made with Motorola and Apple to make simpler, more mainstream RISC CPUs initially utilising a hybrid IBM POWER and Motorola 88000 chip architecture and a subsed of the most used instructions in the POWER instruction set (while for POWER back compatibility the least used ones, not used for native PPC SW, where emulated).
Later PowerPC was merged back into the POWER architecture.
Since POWER8, IBM started licensing the whole designs of its high-end chips to partners http://en.wikipedia.org/wiki/OpenPOWER_Foundation , but even before the POWER architectur had already become for years a multi-company alliance and IBM, despite making its chips under its brand, has been licensing its tech to other companies for years, as evident in the latest CPU used by Ninty, that despite not being a full POWER7 design, uses some POWER7 tech in an older overall PPC design.
The latest chips made by other companies are POWER, not just PowerPC, see for example Freescale: http://www.freescale.com/webapp/sps/site/homepage.jsp?code=PCPPCP
What's true is that IBM in the past was willing to co-develop and produce also a full range of PowerPC CPUs  for desktop and notebook PCs, but starting with the notebook version it lost interest in it, forcing Apple to switch to x86, but even in recent times it's been willing to co-develop versions for consoles, the latest being Wii U's Espresso.

There is a major part wrong of that.

ARM went that strong because of their licensing model, and yes, they went for ultra low power designs. But those low power designs are terribly weak in some points, eveen right now they're getting better.
Power and PowerPC architecture have been very, very efficient performance per watt wise for years. Which is part of why Power based super computers where the most efficient for years.

Yes, Some Power CPU's where pretty hot, but at the same time had a very good performance per watt.
The die shrinked versions of Cell an Xenon are still somewhat efficient, even if dated. Espresso seems to eat up almost no power. And even the gigantic Power8 has a good performance per watt.

The G5 was another story. Apple wanted custom tailerod CPU's for their systems, being developed by IBM. And it seemd that IBM did not want to do so for an actually small customer, chipdesings getting more and more expensive. The G5 design itself was far more efficient than Netburst.
Clocking that chip up like done in the later PowerMacs generated much more heat than originally intended. But that goes for any CPU.
So basically overclocking a CPU (that is what Apple did by default) is ruining your performance per watt.

With the Core architecture Intel just came to be the better, more practical supplier for Apple, because Intel is making all those CPU's anyway.

 

Edit:

 

oh, that thing with 360 code running more or less without any changes on Espresso came from some devs.

generic-user-1 said: powerpc is outdated in many ways because ibm isnt investing enough money.   intel invests way more in x86 and thats why they are more modern. but that doesnt mean x86 is better for every use. the wii u is showing how a small ppc can do the work of a bad 8 core x86(amd is just bad)

AMD is not bad. MS and Sony wanted a shitty 8core in their systems, so that's what they paid for. AMD doesn't make horrible CPUs.

Snesboy said: generic-user-1 said: powerpc is outdated in many ways because ibm isnt investing enough money.   intel invests way more in x86 and thats why they are more modern. but that doesnt mean x86 is better for every use. the wii u is showing how a small ppc can do the work of a bad 8 core x86(amd is just bad) AMD is not bad. MS and Sony wanted a shitty 8core in their systems, so that's what they paid for. AMD doesn't make horrible CPUs.

The Jaguar  core they used is just meant for low power usage applications, like laptops. Not AMD's fault they had to keep the power envelope in check.

captain carot said: There is a major part wrong of that. ARM went that strong because of their licensing model, and yes, they went for ultra low power designs. But those low power designs are terribly weak in some points, eveen right now they're getting better. Power and PowerPC architecture have been very, very efficient performance per watt wise for years. Which is part of why Power based super computers where the most efficient for years. Yes, Some Power CPU's where pretty hot, but at the same time had a very good performance per watt. The die shrinked versions of Cell an Xenon are still somewhat efficient, even if dated. Espresso seems to eat up almost no power. And even the gigantic Power8 has a good performance per watt.

You need to consider that Espresso itself is a 3-core version of a G3 processor, itself having many optimisations over the 10 years since Apple stopped using it. In addition, the Xenon and Cell were custom designs that could be improved over time (by die shrinks and other methods) and not limited to IBM's PowerPC roadmap like Macs - they could be mass produced and gradually improved over several years while Apple needed faster models every 12-18 months.

captain carot said: The G5 was another story. Apple wanted custom tailerod CPU's for their systems, being developed by IBM. And it seemd that IBM did not want to do so for an actually small customer, chipdesings getting more and more expensive. The G5 design itself was far more efficient than Netburst. Clocking that chip up like done in the later PowerMacs generated much more heat than originally intended. But that goes for any CPU. So basically overclocking a CPU (that is what Apple did by default) is ruining your performance per watt.

The "custom tailored cpus" you mention were basically processors promised when the first G5 processors were made - a 3ghz processor (for PowerMac G5) and a lower-power processor (for a PowerBook G5 which never happened). Even at a low speed like 1.6ghz, the G5 was a significant power hog. Both the PowerMac G5 and iMac G5 needed a heavy amount of thermal hardware to keep the processors cold. Same goes for the Xbox 360 and Playstation 3, the former notable for the RROD problem due to (mostly) cooling system failures.

CPU clock does contribute to power gain, correct. Thats why almost every system built this century uses Power Management features, which clock up and down the CPU (even turning it off at intervals) to prevent running at full heat and power usage. Its why battery power on devices depends on use - my laptop battery for example gives me 4 hours internet browsing or 90 mins gaming.

You're right though that IBM refused to honour the promise, claiming Apple was a minor customer. Ironic to think the move away from PowerPC doomed the processor architecture as far as popularity and acknowledgement is concerned, due to Apple having been the most well-known supporter of it.

captain carot said: With the Core architecture Intel just came to be the better, more practical supplier for Apple, because Intel is making all those CPU's anyway.

Originally Apple investigated both Intel and AMD for x86/x64 processors. They went with Intel because, at the time, AMD didn't have many low-power designs available (the original Core Microarchitecture was based on the Pentium M, itself heavily optimised for mobile devices). If AMD had made an Athlon 64 as power efficient as a Core Duo, we would be using Macs with AMD APUs inside.

captain carot said: They could as well have gone with a really improved older one. Seems to be no problem to port from G5 to Espresso. But (not existing) SIMD capabilities, weak FPU etc. are a problem. Keep in mind that Intels Core and to some degree Core 2 architecture basically was an (very) improved P6 architecture chip. Thats hypothetic though because it seems that Nintendo went for a really cheap design. The MCM is what makes the CPU+GPU part expensive. A PPC750 based CPU with the right enhancements might have done very well.

Marcan said Espresso's SIMD was weak, not non-existent.

"The Espresso is an out of order design with a much shorter pipeline. It should win big on IPC on most code, but it has weak SIMD"

http://www.eurogamer.net/articles/digitalfoundry-wii-hacker-reveals-wii-u-cpu-secrets