Why can't Windows run natviely on PPC processors?

[ViRGE]

Originally posted by: slugg

Originally posted by: ShawnD1

Originally posted by: slugg
What's sad is the x86 is a terrible architecture.

Why would you assume this? Wintel PCs would regularly beat Apple computers in benchmarks time after time back when Apple still used PPC. There was even a scandal where Apple was caught posting fake benchmarks on their site that tried to compare the G5 to the Pentium 4 and people noticed that the Quake 3 scores were way lower than what any other hardware site was getting on the P4.

AMD Opteron destroys the G5
Pentium 3 beats the G4

I never said that RISC chips are/were faster. They can be, depending on the application. The reason why x86 benchmarked higher than the PowerPC wasn't that RISC is inferior to x86, but because the specific implementation, PowerPC, was behind (similar to how Intel chips slightly outperform equivalent AMD chips right now). Not only that, but RISC is more dependent on the software's implementation than x86 is.

Specifically, MIPS promotes a high amount of parallelism. Simply, x86 cannot keep up with the multi-pipelining ability of a MIPS ISA. Unfortunately, in order to achieve this peek performance, the MIPS chip must be executing optimized code. Basically, the developer of whatever software is running has to write the code in a way that will allow the compiler to maximize pipelining. So theoretically, a poor written program could starve resources on the CPU and force the entire program to run slowly.

With MIPS, the transistor requirements are much lower, thus lower power requirements, thus lower heat, thus higher reliability. Since multi-threaded programming is required in order to take advantage of the MIPS architecture, multi-core systems are even more practical than x86 since all software targeted for MIPS should be thread and pipeline friendly. With more cores, we increase the throughput-to-power ratio and it eventually becomes linear. Current x86 models are quadratic.

So the main issue regarding x86 versus MIPS is the software, really. The hardware technology is definitely there - it's just not into mass production due to demand.

That being said, MIPS is just one kind of RISC ISA. There are many possibilities

Bear in mind that modern x86 implementations are RISCy too. There's a lot of stuff going on under the hood that the instruction set doesn't directly allude to.

 

[ShawnD1]

Originally posted by: slugg
I never said that RISC chips are/were faster. They can be, depending on the application. The reason why x86 benchmarked higher than the PowerPC wasn't that RISC is inferior to x86, but because the specific implementation, PowerPC, was behind (similar to how Intel chips slightly outperform equivalent AMD chips right now). Not only that, but RISC is more dependent on the software's implementation than x86 is.

My argument was that x86 must be better because we're able to get more performance out of it. How can I say that an architecture like Power is better than x86 when it is slower, generates more heat (G5 mac had 9 fans), and in your words "is more dependent on the software". That doesn't sound like a good architecture.
This whole debate reminds me of how sports teams act. After one team scores a goal, the other will yell "you guys still suck" and "our players are more skilled than yours", but in the end none of that matters. Look at the score board to see which team is better. RISC stuff like PowerPC might look good on paper but they never seem to beat x86

Specifically, MIPS promotes a high amount of parallelism. Simply, x86 cannot keep up with the multi-pipelining ability of a MIPS ISA. Unfortunately, in order to achieve this peek performance, the MIPS chip must be executing optimized code. Basically, the developer of whatever software is running has to write the code in a way that will allow the compiler to maximize pipelining. So theoretically, a poor written program could starve resources on the CPU and force the entire program to run slowly.

The problem here is that MIPS relies on parallelism. Some things like pixel shading can effectively use 800 processors at the same time, but many of our computing tasks are not like that. If you think back to when the Playstation 3 came out, you'll remember that a lot of developers complained that it was quite difficult to split something like a game across 6 cores and do it well. Maybe put AI on this core, video on that one, sound on the next.. and suddenly we find that maybe 3 of those 6 cores are maxed at 100% and the other 3 are doing next to nothing. That doesn't mean the programmer is a retard and it doesn't mean the processor is bad. All it means is that a processor designed for parallel computing is not good for non-parallel computing. Unfortunately most the stuff we do is not easily broken up into smaller tasks.

With MIPS, the transistor requirements are much lower, thus lower power requirements, thus lower heat, thus higher reliability. Since multi-threaded programming is required in order to take advantage of the MIPS architecture, multi-core systems are even more practical than x86 since all software targeted for MIPS should be thread and pipeline friendly. With more cores, we increase the throughput-to-power ratio and it eventually becomes linear. Current x86 models are quadratic.

The G5 was a RISC processor but it still used more power than either Intel or AMD. In Apple's spec sheet they list a single 1.6ghz G5 computer (equivalent to maybe a 2.5ghz P4) as having a max power demand of 420W, and that's with the slowest video card imaginable. The dual 1.8ghz model, again with a low power video card, uses up to 604W. Simply put, the G5 used literally twice as much power as a Pentium 4 and it was still slower. It would be unfair to make a blanket statement about all PowerPC processors, but it seems clear that x86 is not some disgustingly inefficient architecture that falls behind all others.

 

[slugg]

I was defending MIPS, not PowerPC.

PowerPC -> RISC
MIPS -> RISC
PowerPC != MIPS

For example, x86 does not imply x86-64, but x86-64 does imply x86.

Conventional software runs better on conventional hardware. I'll agree with that.

The future is moving toward parallelism and distributed load over multiple cores for all of the reasons I mentioned before (heat, transistors, reliability, etc). If you really want, I'll get some published academic sources to back up my claims... But I still don't think we'll be seeing a MIPS chip in the consumer market any time soon. Think "betamax versus VHS" if you will.

 

[0roo0roo]

Originally posted by: between

Originally posted by: Nothinman

they've backed themselves into a corner and couldn't easily move to another architecture now anyway.

You're not actually reading any of the other posts are you?

you're talking ancient history. could microsoft, today, with relatively little effort, put out a desktop windows for a non x86 architecture? from what I've read, the answer is no.

not sure why you are trolling.

if apple can pull an x86 os out of their keester a much larger company like ms could if they wished support other platforms. they just don't see it as a viable business. end of story.