Is it a bad thing that Windows, OSX and *nix all use the x86 architecture?

[Leros]

Since Apple has switched to the x86 architecture, that means that the majority of the consumer computers are now using x86 processors. Is it bad that there are virtually no competitors to the x86 architecture in the consumer market?

 

[Gamingphreek]

Not necessarily. x86 is easier to program for that most architectures (For some reason PowerPC sticks in my head as being easy to program). There were other challengers at one point. Intel tried to push IA64 into the consumer market but ultimately it remained in the Enterprise Server Class Market.

I believe Transmeta uses a slightly modified version of the IA64 architecture; in fact I believe Transmeta actually has 128bit and 256bit Processors out.

There was one that was supposed to completely take over x86 that Intel was pushing. For some reason, probably cause it is early, the name eludes, me but it was supposed to be the next big thing. Switching Micro Architectures is a very very hard process. Without an emulator ever piece of software written for x86 would have to be completely recoded.

-Kevin

Edit: EPIC thats it!!! Lol...that is the architecture that IA64 is based around. (Explicitly Parallel Instruction Computing)

 

[CTho9305]

All three current generation consoles use PowerPC.

 

[Leros]

Originally posted by: CTho9305
All three current generation consoles use PowerPC.

Yah, but in the consumer desktop marketplace, everything uses either x86 or x86-64.

 

[cheesehead]

Originally posted by: Leros

Originally posted by: CTho9305
All three current generation consoles use PowerPC.

Yah, but in the consumer desktop marketplace, everything uses either x86 or x86-64.

I would not be surprised if the CELL were to find its way into a specialized audio/video editing workstation, perhaps running a modernized version of BeOs - it's supposed to have above-average SMP support, and all of the video editing apps are already there.

 

[icarus4586]

Bad why? There is still competition between x86 CPU makers, so there's no serious worry about monopoly-related problems. As far as competition driving development of better ISAs, that may or may not be important. I personally am not a huge fan of x86 (have you ever seen x86 assembly code? it's ugly!), but as far as performance goes, the ISA only doesn't impact it much. All recent x86 processors have dedicated hardware to decode x86 instructions into simpler, more RISC like instructions that the execution units can use. This hardware comprises a very small percentage (if I remember right, <5% in P4s) of the transistors on the CPU, and is not a performance bottleneck. What's more, hardly any code is written in assembly language, and today's compilers are very good.

 

[Leros]

If all of the consumer desktops are using x86, where is the motivation for chip designers to come up with a new, improved ISA.

 

[cheesehead]

Originally posted by: Leros
If all of the consumer desktops are using x86, where is the motivation for chip designers to come up with a new, improved ISA.

Game consoles. All the code is written specifically for it.

A quickie listing:

-N64: MIPS 4300i
-PlayStation: MIPS 3000a
-Dreamcast: SH-4 RISC CPU
-PlayStation 2: MIPS 5900
-GameCube: 500mhz PowerPC Gekko
-Xbox: Pentium III - the only X86 CPU in the lot
-Xbox 360: 3.2ghz POWER-based triple-core CPU
-Wii: PowerPC based Broadway at 730mhz
-PlayStation III: IBM Cell (totally different from everything else).

 

[Atheus]

Originally posted by: Leros
Windows, OSX and *nix all use the x86 architecture

Well *nix OSs are not a good example as they run on many architectures - look at FreeBSD for example, if I wanted I could run that on everything from a MAC laptop to a Sun workstation to a Playsation 2.

"Is it a bad thing that 90% of all home computers use the x86 architecture?"

Well I suppose it's not ideal...

"Is it a bad thing that 90% of all home computers use a single well documented architecture?"

No, it's a good thing IMO.

 

[Peter]

I wouldn't mind x86 if only we could finally throw out all the old PC architecture cruft that ensures backwards compatibility all the way back to 1981.

 

[kpb]

Originally posted by: Leros
If all of the consumer desktops are using x86, where is the motivation for chip designers to come up with a new, improved ISA.

Now a days the ISA is mainly just another layer of abstraction. Back in the day each chip would have it's ISA and it would pretty directly represent what the chip was capable of. The downside obviously was pretty limited compatability.

Any modern processor, even risc chips like the PPC, use all sorts of tricks so the chips have much more functionality than the ISA has. Really the only penalty for x86 compatibility is some additional die space for the decode units needed to convert to a simpler set and that really isn't major with the rate things are scaling. Back in the day the original pentiums had external cache due in part to the number of transisters that x86 compatibility took but as processes shrunk it has basically become a non issue.

On the other hand x86 compatability is a big advantage because of the huge software library that is available for it. Having various linux/unix distros, windows and now even mac os all on x86 just increase this value.

Is x86 ideal? no not really.

Is it big enough of a deal that it's holding performance or processor design back? Not that I can tell. You might be able to make a slightly bigger cache or smaller chip (cheaper) with out the x86 compatability but there's no way you would sell anywhere near as much. X86 has managed to implement most any feature you could name and both intel and amd are producing high performance chips that keep pace with or beat any competition in general.

 

[Peter]

Compat-i-bility. Thanks.

 

[CTho9305]

Really the only penalty for x86 compatability is some additional die space for the decode units needed to convert to a simpler set and that really isn't major with the rate things are scaling. Back in the day the original pentiums had external cache due in part to the number of transisters that x86 compatability took but as processes shrunk it has basically become a non issue.

I think there is overhead beyond the decoder - for example, the integer multiplier has to support all sorts of sizes of operands and result sizes. The AGUs have to handle the various LEA instructions which send results back to registers and other execution units rather than just the cache / TLB / load-store unit, or your ALUs need to support LEA-type operations. The shifting instructions have annoying behaviors with respect to flags when you shift by 0. You have to support segment checking and add the segment base into address calculations (even though almost no OSes support segments). I'm sure there are other examples, but those are a few I've heard people talking about or have dealt with myself.

 

[kpb]

Ok I oversimplified it a bit but with transistor counts scaling the way they are x86 compatibility is worth the decreasing cost.

 

[Thyme]

It's taken a long time to get compilers to be as efficient as they are today. You'd need a substantially better architecture to outweigh that except in the very long term.

 

[Genx87]

Originally posted by: Leros
Since Apple has switched to the x86 architecture, that means that the majority of the consumer computers are now using x86 processors. Is it bad that there are virtually no competitors to the x86 architecture in the consumer market?

There hasnt been a real competitor in nearly 20 years. So this is nothing new, just the 1.5% of computer users who use Apple will now fall under the x86 fold.

 

[keeleysam]

Originally posted by: Genx87

Originally posted by: Leros
Since Apple has switched to the x86 architecture, that means that the majority of the consumer computers are now using x86 processors. Is it bad that there are virtually no competitors to the x86 architecture in the consumer market?

There hasnt been a real competitor in nearly 20 years. So this is nothing new, just the 1.5% of computer users who use Apple will now fall under the x86 fold.

It's more like 5% and rising, but I digress...

I think the most important part of the consumer parts all being x86 is development can be more focused. Look at Intel's Merom chips. High performance, low thermal output, and can run every consumer operating system.

 

[Fox5]

Originally posted by: Leros
Since Apple has switched to the x86 architecture, that means that the majority of the consumer computers are now using x86 processors. Is it bad that there are virtually no competitors to the x86 architecture in the consumer market?

There is competition in the x86 market place though, Intel, AMD, and VIA.

This hardware comprises a very small percentage (if I remember right, <5% in P4s) of the transistors on the CPU

<1% actually. Much less than 1% at this point I think.

Anyhow, being stuck with the x86 architecture somewhat limits alternative computer paradigms, particularly ones that depend on a different memory hierarchy, but the preexisting software and coding practices also limit that.

-N64: MIPS 4300i
-PlayStation: MIPS 3000a
-Dreamcast: SH-4 RISC CPU
-PlayStation 2: MIPS 5900
-GameCube: 500mhz PowerPC Gekko
-Xbox: Pentium III - the only X86 CPU in the lot
-Xbox 360: 3.2ghz POWER-based triple-core CPU
-Wii: PowerPC based Broadway at 730mhz
-PlayStation III: IBM Cell (totally different from everything else).

1. Those were all preexisting architectures targetted for other markets prior to consoles.
2. Prior to this gen, the xbox's cpu was the most powerful console cpu ever seen; and it was slow for a pc cpu at the time of the xbox's launch. (and for some code, the cell and 360 are only supposed to be running between the same speed as the xbox cpu to twice the speed)
3. Cell is still Power based, with coprocessors.

You might be able to make a slightly bigger cache or smaller chip (cheaper) with out the x86 compatability but there's no way you would sell anywhere near as much.

x86 instructions are actually less cache hungry than many other ISAs. (at the side effect that they're a real pain to design hardware around or program for, at the low level anyhow)

 

[CTho9305]

Originally posted by: Fox5

This hardware comprises a very small percentage (if I remember right, <5% in P4s) of the transistors on the CPU

<1% actually. Much less than 1% at this point I think.

Do you have a source for that? If you look at this image (assuming he labeled everything correctly), the "4x microcode flash memory", "micro code sequencer", and "complex instruction decoders" blocks could all probably disappear (or at least shrink) if the instruction set were simpler... the "3x seg. limit checking" block is an x86-only thing (which isn't even used when you're running most OSes).

 

[Fox5]

Originally posted by: CTho9305

Originally posted by: Fox5

This hardware comprises a very small percentage (if I remember right, <5% in P4s) of the transistors on the CPU

<1% actually. Much less than 1% at this point I think.

Do you have a source for that? If you look at this image (assuming he labeled everything correctly), the "4x microcode flash memory", "micro code sequencer", and "complex instruction decoders" blocks could all probably disappear (or at least shrink) if the instruction set were simpler... the "3x seg. limit checking" block is an x86-only thing (which isn't even used when you're running most OSes).

No link, sorry, it's just something I read a while ago (I believe on arstechnica). However, it was talking specifically about the hardware included to convert the instructions in the original pentium pro, I wouldn't be surprised if both AMD and Intel added more hardware to handle the instructions since then for performance reasons. I believe on the Pentium Pro the hardware took up about 20% of the transistors. (I don't know about density though, so I can't say how much die space it took up, which is the more important figure)