Ouya Dev Consoles Shipping Soon

You are comparing custom RISC processors to x86 CISC processors, that's not a very good comparison.
PowerPC and POWER have changed a lot since the 90's and 00's, especially since the 360 and PS3 were released.

You guys really need to read up on RISC vs CISC and the processor environments.
x86 CISC CPUs are not the end-all-be-all processors in existence, nor are they the best, at anything.
You do realize that x86 CPUs for a long time are really RISC? They've been RISC since Pentium Pro. They're CISC in that they have an interpreter for the RISC core.


A POWER7 CPU will absolutely crush today's top x86 CPU in double-precision FLOPS with little effort, and the multi-processor IPC is in another league of its own.
This is why most supercomputers now are starting to utilize weak x86 processors, not to crunch numbers, but to use their small power (electricity consumption) profile and to feed data to CUDA-based GPUs to do the number crunching.

If you compare x86-only (CISC) CPU supercomputers to RISC CPU supercomputers, the RISC supercomputers absolutely decimate them.

This pretty much defeats the whole purpose of the CPU vs CPU argument.
I agree.
1. Why does In-Order seem to go over peoples heads? The PS3 and 360 are dog slow without Our-of-Order.

2. If Power7 was really that fast then why hasn't Apple gone back to them and preach about Photoshop benchmarks?

3. When you could run Linux on the PS3, people have benchmarked the machine. Even people here in this forum have found it to be no better then a G5. We know how a x86 machine compares to a G5, and we know how a x86 machine compares to an ARM chip like the one found in Ouya. So it's probably safe to say that Ouya isn't to be much worse in performance through the CPU. The GPU though is an entirely different topic.
 
Yes I realize that CISC is really just a RISC "core" with CISC instructions built on top of it.
This is why I think x86 processors are buggy and unstable, and need to die off, but unfortunately the market won't let it, or Intel for that matter, move to another architecture, with the small exception of ARM.

1. I have no idea why it goes over their heads, it does certainly make quite a difference, but it also adds to a large amount of power consumption and die-area for the process to be built; one of the reasons why ARM has been slow to adopt it and why Atom processors don't include it (at least on the older models).

2. Apple will not switch to POWER7 because it was never designed for a desktop/laptop portfolio, not to mention OS X was never compiled nor optimized to work with it. POWER7 chips were designed for a many-processor platform with large amounts of IPCs, interconnects, and nodes, making it more popular in supercomputer and mainframes rather than desktops.
Also, Apple never used the POWER architecture, ever.

3. The G5 processor, aka the 970/970FX/970MP, is actually a highly modified POWER4 processor with reduced IPC functionality, L2 cache, with the addition of Altivec instructions.
Even though it is actually a modified and stripped down POWER4 processor, it is actually using the PowerPC architecture, not the POWER architecture.

As for Linux on the PS3, it was using Yellow Dog Linux, a RHEL distro modified to be optimized for the PowerPC architecture.
The Cell processor, however, is not a straight/complete PowerPC processor, in that it only had one PowerPC core, with the other "cells" being modified processing units.
In other words, the Cell processor itself, unless the code is optimized for it, is not that hot of a processor.


Another thing, you are comparing x86 chips to RISC processors from the mid-2000's, things have changed quite a bit on both fronts.
Seriously, you need to do some research on both processors, as there is more to this than out-of-order execution at work. ;)
 
ppc970.jpg


As I said above, while similar in some areas, they are very different "under the hood".

None of this matters though, as games are optimized for whatever console they are built for, so doing a comparison of ARM vs POWER is completely pointless in this regard.
Honestly, as the others have said, the GPU being used will make the most difference out of the hardware; the rest is up to the coding and software optimization.
 
Majority of PS3 games don't make use of the SPE's. Very few games actually make use of the SPE's.

Much more than "very few" utilize the SPE's. Probably all first party games make major use of them. As such, they are among the very best looking console games this generation. It is true that many multi-platform games hardly utilize them, if it all. But even way back in 2006, an Formula 1 game used them to run weather effects and check precision and color correction on every single pixel. Plenty of high profile third party games use the SPE's. The most recent being Far Cry 3, which is quoted as averaging 65% active SPE usage.

http://engineroom.ubi.com/the-spus-are-hungry-maximizing-spu-efficiency-on-far-cry-3/
Overall, all these systems have allowed us to hit a mean 65% of active SPU usage
 
This is why I think x86 processors are buggy and unstable, and need to die off, but unfortunately the market won't let it, or Intel for that matter, move to another architecture, with the small exception of ARM.

So an architecture that's been in existence for a generation that powers the overwhelming majority of servers and desktops, running millions of programs, over a billion machines, billions of lines of code, is buggy and unstable.

Even if you're right you realize that you have no way practical way to prove your statement without a comparison to another architecture that's as widely deployed and utilized and of course there isn't one at the moment. ARM has a long way to go before it's deployed at scale on desktops and servers.
 
3. When you could run Linux on the PS3, people have benchmarked the machine. Even people here in this forum have found it to be no better then a G5.

Um, you didn't appear to read that article, did you. :rolleyes:
The Cell processor is actually a LOT faster than a vanilla 970, and even the 970FX and 970MP CPUs.

"Geekbench also isn’t able to exploit the eight vector processors on the Cell processor. Any program designed and optimized for the Cell processor should be a lot faster than one designed for a generic processor (like, say, Geekbench). So while the Geekbench results might seem disappointing, keep in mind that Geekbench can’t exercise the PlayStation 3 to its full potential."

The PowerPC core of the Cell, on its own, isn't very powerful.
This is why it is hard to compare to other PowerPC processors of that era.

Technically speaking, IBM, only a few months ago, finally retired Cell processor server-nodes.
It's only now that x86 processors are finally able to outperform the Cell in double-precision FLOPS, and even then, the POWER7 processor completely crushes any x86 on the market in this area.

This shows you how fast Cell processors really were compared to their x86 counter parts, especially from 2006.
That benchmark was completely unoptimized and incomplete for the Cell processor, and at best, was an apples-to-oranges comparison, even against a 970 processor, because they are two different types of PowerPC.

Sorry, but that wasn't a very good comparison, at all.
 
Much more than "very few" utilize the SPE's. Probably all first party games make major use of them. As such, they are among the very best looking console games this generation. It is true that many multi-platform games hardly utilize them, if it all. But even way back in 2006, an Formula 1 game used them to run weather effects and check precision and color correction on every single pixel. Plenty of high profile third party games use the SPE's. The most recent being Far Cry 3, which is quoted as averaging 65% active SPE usage.

http://engineroom.ubi.com/the-spus-are-hungry-maximizing-spu-efficiency-on-far-cry-3/

3. When you could run Linux on the PS3, people have benchmarked the machine. Even people here in this forum have found it to be no better then a G5. We know how a x86 machine compares to a G5, and we know how a x86 machine compares to an ARM chip like the one found in Ouya. So it's probably safe to say that Ouya isn't to be much worse in performance through the CPU. The GPU though is an entirely different topic.

Yeah that was a general benchmark that may not have even loaded both threads from the main PPE core. (maybe the benchmark supports multiple threads, I'm just spitting shit right now). Even if it did load both threads, it certainly did not utilize the 6 SPEs available to run non-OS code. Which, again, completely overlooks the entire point of the design. As such, it is not a fair comparison. Cell cannot be comparied by conventional general bnechmarking. You need to look at how fast it meets the same goals on custom code, or how much stuff it does, while still meeting the goal acceptably.

After folding@home hit PS3, it only took I think 3 months for the PS3s out there folding to surpass all the other computers ever, in the amount of completed work in the types of folding available on PS3 (excuse me, I don't know the lingo/terminology associated with folding).
 
So an architecture that's been in existence for a generation that powers the overwhelming majority of servers and desktops, running millions of programs, over a billion machines, billions of lines of code, is buggy and unstable.

Unfortunately, it is a very buggy architecture, and believe me, I wish it weren't.
Even Intel has wanted to move away from it for years, but the entire consumer-market is based on x86-compiled software and OSes; in other words, it would be financially bad for Intel to make that kind of a move.

I'm sorry that I don't currently have the source specifically stating this, but if you find one, please post it.

Even if you're right you realize that you have no way practical way to prove your statement without a comparison to another architecture that's as widely deployed and utilized and of course there isn't one at the moment. ARM has a long way to go before it's deployed at scale on desktops and servers.
ARM was never designed to fill the gap that x86 processors serve, hence it's low-power footprint and lack of out-of-order processing and 64-bit architecture/extensions (though AMD and ARM are changing this).

I certainly don't expect you or anyone else to take my word at face value, and if I can find some sources specifically stating the issue, I'll be sure to post them.
 
So an architecture that's been in existence for a generation that powers the overwhelming majority of servers and desktops, running millions of programs, over a billion machines, billions of lines of code, is buggy and unstable.
Also, I should have said buggy and unstable, relative to other processor architectures. ;)

This is also one of the reasons supercomputers are having x86 processors paired with CUDA GPUs.
The x86 processors can run the OS, memory management, IPC, and other basic functions, while the CUDA GPUs actually do the number crunching.

Take a look at the Titan and Tianhe-1A supercomputers, it's working out quite well for both USA and China.
A RISC CPU-only supercomputer has much higher power and heat envelope, and a much higher cost.

I do understand that x86 processors are here to stay, and while I have wanted ARM processors to finally be the x86-killer, I know that this won't happen for many years, if ever.
I don't think that x86 processors are horrible or anything, but I do believe that this is much to improve upon in the architecture's stability when compared to other RISC processors.
 
Also, I should have said buggy and unstable, relative to other processor architectures. ;)

And this is my point, how do you prove this? There is no architecture that's as old and widely deployed as x86 with witch to compare.

This is also one of the reasons supercomputers are having x86 processors paired with CUDA GPUs.
The x86 processors can run the OS, memory management, IPC, and other basic functions, while the CUDA GPUs actually do the number crunching.

But GPUs are designed for number crunching and they aren't particularly good at linear processing such as x86 CPUs. Pointing out the strength of one design doesn't prove weakness in another when the comparison isn't equivalent.
 
And this is my point, how do you prove this?
OS X 10.5 on IBM 970xx vs Intel C2D.

10.5 on the PPC processor was rock-solid and for years, never had any issues.
10.5 on the Intel constantly had issues, lockups, and all kinds of problems when the CPU was under load.

Before you say that Apple had just moved to Intel and that 10.5 wasn't ready, Apple had been co-compiling and developing OS X since 10.0 for x86, so the OS was more than ready and optimized for the architecture.


There is no architecture that's as old and widely deployed as x86 with witch to compare.
I'm sorry, but this just shows you've been working with Windows for far too long.
Other OSes are compiled and optimized for more than just x86; I know Windows RT is compiled for ARM, but wooptydoo, one whole other processor architecture.

Linux and UNIX OSes are optimized for multiple processor architectures, and this is the test-base I personally use.

GNU/Linux, especially distros from the Debian branch, are compiled for not only x86, but also a slew of non-x86 RISC processor architectures as well.
Of the distros that I and others I know have worked with and operated, have all been much more stable on RISC processor architectures than on the x86 architecture; the one exception being PowerPC on Ubuntu 10.xx and 11.xx.

I know that I'm using personal experience for examples, and mind you this is over years of usage, not just a few tests.
This is why I'm saying, you don't have to take my word at face value, I'm asking you to help me find some sources stating this, other than my own.

I have read them over the years, some old, some recent, and am still looking for them.
 
Really? So what do the overwhelming majority of Linux desktops and servers run on?

x86 of course.
That doesn't mean it is solely optimized for x86.

Boy, you don't like it when I say something bad about x86 processors or Windows, do you? :p
 
x86 of course.
That doesn't mean it is solely optimized for x86.

Boy, you don't like it when I say something bad about x86 processors or Windows, do you? :p

You can say all the bad things you want, I was simply pointing out that you made a ridiculously broad statement that can't possibly be proven. If x86 is so unstable and buggy then I'm sure you would have made a better chip by now.
 
You can say all the bad things you want, I was simply pointing out that you made a ridiculously broad statement that can't possibly be proven.

It was just a joke, hence why I put a " :p " next to it. :rolleyes:
As I said before, I'll see if I can find any of the articles stating such.

If x86 is so unstable and buggy then I'm sure you would have made a better chip by now.
Oh wow, you got me there. :rolleyes: :rolleyes: :rolleyes:
 
It's called sarcasm. /s
So in other words, it wasn't sarcasm.

You know, you should pull that stick out of your ass, it's not making you look any cooler than you think you already are.
Just a suggestion.
 
Posted on another forum:
Yep, that's right. The 64 bit extensions to the x86 architecture were invented and first implemented by AMD. Intel were laggards to the 64 bit party because they were hoping to transition away from x86 architecture and move us towards IA64. IA64 already existed in the form of the Itanium and Intel was hoping to transition towards that. Despite being a far better and more efficient architecture, lacking the inherent design flaws of x86, it suffered through lack of compatibility and it never took off except in the commercial space. AMD effectively forced Intel's hand by introducing the Sledgehammer, which was effectively a 64bit chip based on the x86 architecture. Hence the name AMD64.

I know this really isn't a good source, but at the moment, it's as close as I can get.
 
Posted on another forum:

I know this really isn't a good source, but at the moment, it's as close as I can get.
There was never a strategy to transition Intel's desktop and mobile chips to IA64. IA64 was designed as a workstation and mainframe chip, and wasn't designed as scalable hardware.

To this day it's generally believed because some tards on thereg claimed it based on clueless analyst comments. There was a great debunking of it on RWT a while ago. However, IA64 was supposed to be Intel's platform for the server (Itanium is firmly a big iron product for years now), and it's continuously fallen short on growth estimates. It could certainly be argued that the Opteron shifted the server market solidly to x86, not only later when higher clock speeds were hit and had a significant performance lead, but also in lowering the cost of servers and server chips. Itanium is a large chip and would have never been competitive in the mainstream server segments (1P/2P) after x86 solidly took hold.

If we remember, Intel was stating that 64-bit x86 wasn't needed on the desktop, while it was adding that functionality to Prescott. It never added or had designs for desktop/mobile IA64, which you'd imagine Intel would have if that was the strategy. It's not like we don't hear about aborted Intel processors. :p

BTW, Intel did investigate extending x86 to 64-bit* back in the Pentium 1/2/3 era, but decided against it. Even if Intel resurrected that version, the instruction encodings would have been different, and registers (AMD added only 8 new ones) and other things would have made them incompatible with AMD's instructions, even if both did the same function. That was the basis of the warning MS made back around 2000 that it would only support one 64-bit instruction set for desktop processors, and choose AMD because the specs were public and it would be for sale sooner.

*Which is what AMD did. x86-64 ("AMD64") is not a brand new set of instructions, it's just x86 which can operate on up to 64-bit integer operands and with extra registers (which are new opcodes), plus some new processor management instructions. It's very useful for servers and certain desktop applications, but it was a relatively small change from 32-bit x86 ("IA32") compared to say when Intel extended the 16-bit 80286 to 32-bits with the 80386 (tons of changes to the basic operating modes, memory addressing and pretty much all of the non-SIMD application instruction set used even today).

I'm not against AMD64. I think it was a very good thing. Intel was trying to segment the market with 64-bit, and it failed due to AMD64. What could you expect from the company at the time when it was designing Netburst? Lots of bad decisions were happening. ;)
 
Um, guys, why are you saying the Wii U's CPU is PPC, when it is a custom POWER7 CPU.
PowerPC != POWER
You're misunderstanding the concept of ISA (instruction set architecture) vs implementation.

The Wii U has a custom designed PowerPC processor, which are the reduced cost versions of the POWER ISA, intended for embedded applications where power and cost are a factor.

The only thing the PPC750 variant in the Wii U (which it identifies itself as) has with POWER7 is the compatible instruction set. The size of the Wii U CPU die also suggests the strong PPC750 lineage vs anything approaching a POWER7 execution core. It's just not large enough to have 3 POWER7 cores (excluding the cache) in the size of the Wii U's CPU die.

There is no problem acknowledging that the Wii U processor uses the POWER ISA, however it's not the same as the POWER7 implementation. For example, the Atom processor is as much a member of the x86-64 ISA as an 8 core Xeon made for MP systems using thousands of processors and custom connection fabric. In this sense, the Wii U got an "Atom": a vastly simplified core, a la embedded PowerPC, with very much reduced performance. Understand?
 
Also, I should have said buggy and unstable, relative to other processor architectures. ;)

This is also one of the reasons supercomputers are having x86 processors paired with CUDA GPUs.
The x86 processors can run the OS, memory management, IPC, and other basic functions, while the CUDA GPUs actually do the number crunching.

The crazy speedups you see via GPGPU aren't necessarily down to a shitty x86 chip in comparison - we're talking strictly dumb FP heavy data. The 10-500x speedups you've seen are mostly due to bad implementation and optimization on the CPU in the first place. If you take roughly equally optimized code for either, the speedup is in the area of 2x when going GPGPU. That's a lot less than some of the crazy figures you've seen posted but it's still twice as fast :p

x86 isn't "unstable and buggy," it's just not as efficient in die area and slightly higher in power consumption; given an equal process, something like ARM will have a 2-3% smaller die. That isn't much and can obviously be made up for in finer hand tuning and even a smaller node and better/smaller fabrication.

The compiler for x86 is also a bit messier than something like ARM. CUDA is favored by many in the HPC space simply because of it's closer-to-metal approach.

I imagine that anything programmed on the Ouya box will be Java purely because it allows for the game devs to stretch across everything and still run. The Tegra 3 is a pretty poor gaming SoC, though... Tegra 4 (Wayne) is being introduced in Q1 2013 at CES which should offer a significant upgrade over the aging Tegra 3. I kind of wished they'd waited to decide on the hardware
 
You're misunderstanding the concept of ISA (instruction set architecture) vs implementation.

The Wii U has a custom designed PowerPC processor, which are the reduced cost versions of the POWER ISA, intended for embedded applications where power and cost are a factor.

The only thing the PPC750 variant in the Wii U (which it identifies itself as) has with POWER7 is the compatible instruction set. The size of the Wii U CPU die also suggests the strong PPC750 lineage vs anything approaching a POWER7 execution core. It's just not large enough to have 3 POWER7 cores (excluding the cache) in the size of the Wii U's CPU die.

There is no problem acknowledging that the Wii U processor uses the POWER ISA, however it's not the same as the POWER7 implementation. For example, the Atom processor is as much a member of the x86-64 ISA as an 8 core Xeon made for MP systems using thousands of processors and custom connection fabric. In this sense, the Wii U got an "Atom": a vastly simplified core, a la embedded PowerPC, with very much reduced performance. Understand?

Thanks for the info, I stand corrected.
I guess some of my knowledge was a bit on the legacy side.
 
Back
Top