Stop the x86 Instruction Set War

HardOCP News

[H] News
Joined
Dec 31, 1969
Messages
0
If you have a few minutes to spare, all of you really should check out this guy’s blog and his call for an open, unified standard for x86 instruction set extensions. Very good reading and definitely a link worth clicking. Thanks to Henry R. for the link.

There is an almost invisible war going on between Intel and AMD. It's the game of who is defining the new additions to the x86 instruction set. This war has been going on behind the scenes for years without being noticed by the majority IT professionals. Most programmers don't care what is going on at the machine code level, so they can't see all the ridiculous consequences that this war has. Those working with virtualization may have noticed that Intel and AMD processors are incompatible when it comes to virtualization software, but this is only one of the more visible consequences of the conflict.
 
If you have a few minutes to spare, all of you really should check out this guy’s blog and his call for an open, unified standard for x86 instruction set extensions. Very good reading and definitely a link worth clicking. Thanks to Henry R. for the link.

For anyone interested in this subject, do some googling on nested KVM. it has been working on AMD procs for quite some time (A year+ I think) and last I read it didn't work on intel chips.

Pretty much for any box I might one day use for virtualization where "general performance" isn't a limitation I generally choose AMD's.
 
Hmm...open standards committee, that'll lead to innovation. :rolleyes:

It actually works quite well, and for those who dont think this matters. take a long look at what NVIDIA is doing with CUDA. for all you know CUDA will be the x86 of the GPU (i know its more high level than x86) but thats really what you are saying.

OpenCL yea that will never work....

I think he is right and there should be a cutoff/ emulation timeline within the series.

Very well written, but $38billion/year cough*INTEL*cough says that wont happen (although I REALLY wish it would)
 
I think he is right and there should be a cutoff/ emulation timeline within the series.

Very well written, but $38billion/year cough*INTEL*cough says that wont happen (although I REALLY wish it would)
Agreed, I think he's right too.
 
I'm not the biggest advocate for so called "standards". Don't get me wrong, standards have their place but at the same time most of these so called standards committees are merely fronts for companies trying to push their capitalistic agenda by an "open" committee.

I would much rather let the market determine "standards". If at some point things need to be cleaned up like with Microsoft and IE for instance, that can be problematic but waiting for a committee to do real work and release products isn't all that either. We would still be waiting for a lot of things with a committee process.
 
I don't know, seems like the SAS and SATA commitees have done alright. Wireless N took a long ass time, but I wonder how much of that was BS.

Innovation in a free market is one thing. But in this case Intel's dominance all but thwarts other ideas, so you're really not seeing too much innovation. Imagine if x86_64 had been stomped out like these other ideas?
 
I'm not the biggest advocate for so called "standards". Don't get me wrong, standards have their place but at the same time most of these so called standards committees are merely fronts for companies trying to push their capitalistic agenda by an "open" committee.

I would much rather let the market determine "standards". If at some point things need to be cleaned up like with Microsoft and IE for instance, that can be problematic but waiting for a committee to do real work and release products isn't all that either. We would still be waiting for a lot of things with a committee process.

In this case I don't think the market has much input, or at least not as much as it should. With x86_64 it did, but the others not as much.
 
If anything I think this article makes a case for saying that the instruction set war is a good thing :p

I mean, they are constantly trying to innovate to beat each other out. Isn't that what we want in a good market, rather than them agreeing on certain standards and screwing the consumer?
 
If anything I think this article makes a case for saying that the instruction set war is a good thing :p

I mean, they are constantly trying to innovate to beat each other out. Isn't that what we want in a good market, rather than them agreeing on certain standards and screwing the consumer?
what i think he's trying to say is standardizing the x86 code to make it more efficient and just stick to making the processors run that standard code faster.

There are plenty of standards that they adhere to on the PC already so why not the base code?
:confused:
 
Well, I agree, having a base of instructions is useful, but according to the article, whichever instruction set won the battle at the time became adapted by the competitor.
 
You should be grateful that W3 set the html standard and not motherfucking Microsoft Internet Explorer. [enter six paragraphs of how long IE took to start rendering pages accurately without demanding rewritten scripts and all]
 
You should be grateful that W3 set the html standard and not motherfucking Microsoft Internet Explorer. [enter six paragraphs of how long IE took to start rendering pages accurately without demanding rewritten scripts and all]

I was thinking the same thing. W3C is starting to pay off now and agreed upon specifications are better than a free for all when specifications are used as weapons to screw competitors. Microsoft's inability (or refusal) to follow simple coding rules is criminal when you consider the lost productivity, lost time and incurred cost of it's actions.
 
If anything I think this article makes a case for saying that the instruction set war is a good thing :p

I mean, they are constantly trying to innovate to beat each other out. Isn't that what we want in a good market, rather than them agreeing on certain standards and screwing the consumer?

The problem is that we essentially have duplicated instruction sets, a minefield of opcodes, and a ton of complexity that not only complicates writing software and the compiler, but that slows down the CPU itself due to it needing to not only figure out how to decode variable length opcodes using escapes and such, but also in supporting instructions that absolutely nobody uses.

Imagine if you were making something and you were never allowed to remove something or fix a mistake AND you were working on the same project as someone else AND you two never talked about the changes you were doing - that is the problem.
 
Look, the way I see it we have a base system i386 - P4 is the basic X86 (pentium and later have some additional instruction sets but they are used in AMD as well I believe) we also have X64 an extension to the X86 platform.

the additional instruction sets are useless unless applications are written to make use of them. so from my perspective it does no harm in having this war, but this does make thing more difficult for the programmers out there. I would suggest Intel and AMD work together on creating a simple ID system (pretty much already exists) that would allow the software to identify the needs and execute the program to match the CPU's capabilities. Failing that create a daughter board that fits into a PCIE slot that have Intel/AMD CPU (what ever one you don't have on your system) and install it so that A. you have compatibility, and B. more processing power.
 
hey i dont know anything about x86 architecture and stuff, but i really appreciate the variety and honesty of opinions in this thread and the avoidance of flaming/namecalling like these front pages normally turn into. ive had fun reading all your comments, even if they dont make complete sense to me. keep up the intelligent discussion! :)
 
Failing that create a daughter board that fits into a PCIE slot that have Intel/AMD CPU (what ever one you don't have on your system) and install it so that A. you have compatibility, and B. more processing power.

also, i have no idea if this makes sense or is possible, but it sounds awesome! one intel proc and one amd proc in the same system? sounds uber complex, but also incredibly interesting.
 
Imagine if you were making something and you were never allowed to remove something or fix a mistake AND you were working on the same project as someone else AND you two never talked about the changes you were doing - that is the problem.


Sounds like where I work.
 
Man this guy sounds second-rate...

Oh don't get me wrong - what he says is very technically sound and true. But how does he intend for it to happen? That the government will go ahead and tell Intel and AMD to play nice? Intel will go through with this stuff naturally when it needs to in order to keep Moore's Law alive, not because some guy with a really good understanding of the material told them it was a good idea (welcome to American corporate culture).

But the bigger problem is not hardware support but software support - essentially abolishing the x86 system as we know it would be like the transition Apple made from PowerPC to x86. Good luck, have fun.

The way he presents his argument reminds me of my radical theory to fix 90% of this country's problems almost overnight - to outlaw personal automobiles (emergency and supply are still legal) and instead forcing everyone to take public transportation, which will have been beefed up to hundreds of times its current size in all major cities in the US using funds previously used to maintain our addiction to oil automobiles. It solves our environmental problems, air quality problems, foreign dependence on oil, makes us safer, eliminates drunk driving overnight, saves everyone who will be in an automobile accident and brings back their economic output, makes us healthier since more people will buy bicycles... so why don't we do it? Because we're stuck in our current quagmire of a system - the problem is not what should we do but how we'll get there.
 
The way he presents his argument reminds me of my radical theory to fix 90% of this country's problems almost overnight - to outlaw personal automobiles (emergency and supply are still legal) and instead forcing everyone to take public transportation, which will have been beefed up to hundreds of times its current size in all major cities in the US using funds previously used to maintain our addiction to oil automobiles. It solves our environmental problems, air quality problems, foreign dependence on oil, makes us safer, eliminates drunk driving overnight, saves everyone who will be in an automobile accident and brings back their economic output, makes us healthier since more people will buy bicycles... so why don't we do it? Because we're stuck in our current quagmire of a system - the problem is not what should we do but how we'll get there.

Oh.. thanks I need the laugh this morning (and this coming from someone that uses public transit every weekday).

That is the funniest thing I've read in a long time. It takes me longer to go six miles using a heavily developed transit system in DC then it did to drive 30 miles to work when I lived in Phoenix. It is costing $2B+ and 20 years to build an 18 mile extension to metro.. yah this will work great.

On topic, Intel owns x86 and I don't see them turning it over to a committee. It would be nice to see some more cooperation, but it would not be in their best interests to just turn to full thing over to an open initiative.
 
Hmm...open standards committee, that'll lead to innovation. :rolleyes:

It does actually.

USB is an open standard
TCP/IP is an open standard
...

This kind of BS will eventually fragment the market or cause the competition to vanish and I am sure intel knows this
 
interesting read. i've always found it weird that modern cpus are backwards compatible with the stone age of computer technology (A20 gate!?), but he has a point with "marketing".

i'm not an expert at all in this field, but how much silicon real eastate could a cpu manufacturer save if it gets rid of all the legacy stuff the chips got for "backwards compatibility"? i mean there must be some logic somewhere to execute these opcodes. would it take out significant complexity and make room for something else or is this a nonissue from a chip design point of view?
 
blog said:
Today, there are far more than a thousand different instruction codes, and many of them use complicated combinations of escape codes, prefix bytes, and postfix bytes to distinguish the different instructions.

That doesn't sound healthy. :rolleyes:
 
i'm not an expert at all in this field, but how much silicon real eastate could a cpu manufacturer save if it gets rid of all the legacy stuff the chips got for "backwards compatibility"?



The die space is not a big deal these days. The real hit, as the article mentioned, is that no one can write a decent universal compiler.

Refining the instruction set could make the processor run faster not due to extra transistors freed up but rather by simplifying the workload. For instance an instruction without extra escape codes will be smaller and free up cache capacity/bandwidth. Simpler instructionscould reduce multi-cycle decodes into single cycle decodes (especially beneficial for a stalled processor). There are plenty of good reasons to standardize.

Perhaps if EFI ever gets here we could have a truly legacy-free system that could use a BIOS-emulating VM by default for older OS's. New OS's supporting a cleaned up IA could just make a standardized call to the EFI to not run in legacy mode.
 
Today, there are far more than a thousand different instruction codes, and many of them use complicated combinations of escape codes, prefix bytes, and postfix bytes to distinguish the different instructions.

Not only does it not sound healthy, it sounds a lot like the IRS
 
The die space is not a big deal these days. The real hit, as the article mentioned, is that no one can write a decent universal compiler.

oh right, the compiler... i'm dumb, yea he mentioned that. thanks for the hint. ;)
 
None of this seems 'sound' to me. What it 'sounds' like is a lazy programmer. Let's not forget that every computer used to be specialized until the x86 ended up winning by sheer marketing. Right now there are still completely different processors and instruction sets out there and plenty of companies willing to write different software code for each one. And on top of different instruction sets, there are different OS's, different types of processors using the same instruction sets or slightly altered versions, etc.

The hardware companies are competing for dominance and through that competition we get progress. If there was a standards process involved, that evolution would be hamstrung. As for virtualization, it seems to be progressing nicely and is still a fairly new development. Trying to say the hardware is hampering it, is just a cop out for not wanting to think about how to make it work with the hardware you have. This seems to be a guy who has never had to program for Sparc or SGI or motorola RISC chips and has only been programming for x86 and yet he whines about 'one' set of mostly 'standardized' instruction sets? And I say that because remember, for better or worse AMD and Intel are in the x86 bag together.
 
Oh.. thanks I need the laugh this morning (and this coming from someone that uses public transit every weekday).

That is the funniest thing I've read in a long time. It takes me longer to go six miles using a heavily developed transit system in DC then it did to drive 30 miles to work when I lived in Phoenix. It is costing $2B+ and 20 years to build an 18 mile extension to metro.. yah this will work great.

On topic, Intel owns x86 and I don't see them turning it over to a committee. It would be nice to see some more cooperation, but it would not be in their best interests to just turn to full thing over to an open initiative.

OT: Gotta love WMATA (Metro)......they are the sinking ship....
 
None of this seems 'sound' to me. What it 'sounds' like is a lazy programmer. Let's not forget that every computer used to be specialized until the x86 ended up winning by sheer marketing. Right now there are still completely different processors and instruction sets out there and plenty of companies willing to write different software code for each one. And on top of different instruction sets, there are different OS's, different types of processors using the same instruction sets or slightly altered versions, etc.

The hardware companies are competing for dominance and through that competition we get progress. If there was a standards process involved, that evolution would be hamstrung. As for virtualization, it seems to be progressing nicely and is still a fairly new development. Trying to say the hardware is hampering it, is just a cop out for not wanting to think about how to make it work with the hardware you have. This seems to be a guy who has never had to program for Sparc or SGI or motorola RISC chips and has only been programming for x86 and yet he whines about 'one' set of mostly 'standardized' instruction sets? And I say that because remember, for better or worse AMD and Intel are in the x86 bag together.

It's more about having efficient systems in place for x86. I think the following summed it up pretty well:

Refining the instruction set could make the processor run faster not due to extra transistors freed up but rather by simplifying the workload. For instance an instruction without extra escape codes will be smaller and free up cache capacity/bandwidth. Simpler instructionscould reduce multi-cycle decodes into single cycle decodes (especially beneficial for a stalled processor). There are plenty of good reasons to standardize.

So in other words, we could increase performance, and if I understand this correctly, probably reduce power consumption if x86 was more efficient/standardized. Sounds like a good thing to me.
 
It's more about having efficient systems in place for x86. I think the following summed it up pretty well:



So in other words, we could increase performance, and if I understand this correctly, probably reduce power consumption if x86 was more efficient/standardized. Sounds like a good thing to me.

And that is why RISC > CISC (esp CISC that is designed by commitee).
CISC is great *IF* you hardly ever have to use an intermediate step. the moment you start having to back-peddle CISC commands you start wasting processing time/resources...
 
RISC didn't catch on 15 years ago when it still had a chance, there is no way it would unseat x86 now. Making things more effecient has nothing on maintaining backwards comaptibility for most people.

Look at all the screaming and hatred for Vista when some hardware and software wouldn't work? If you completely changed everything and said sorry, too bad it would never get any adoption.

Apple was able to switch architecures because they had a much smaller (and overly fanaticaly) base and even that took them years before they killed PowerPC compatible software completely.
 
you the funny thing is intel tried to do what he wants it was call the I64 and no one wanted it because they have software that cost billions of dollars and years to write, that they did not want to pay to re-write.

The military stilled used 286 in 2000 or so for a couple pieces of equipment because they did not even know how to fix the code any more and it worked. Why pay to have someone re-write the code and maybe have a few bugs that could kill people? I'm not going to say which piece of equipment in case they are still using it.
 
"A designer knows he has achieved perfection not when there is nothing left to add, but when there is nothing left to take away."

The A20 gate... Yes, the prodigal son.

Personally I think the SSE/AMDnow/Via "extra" instructions should have undergone an audit a looooong time ago. There are so many unused instructions its insane. But likewise some programmers have found some use for them.

Digging through every single instruction to find one that accelerates your code 10% is fine - but you do have to ask yourself if its really worth it in the end. If that one instruction did not exist at all - and the transistors were put towards say a larger cache for faster overall cache - would you take the cache size increase instead. If every programmer did that - things might not be spinning out of control.

But then again - doing it that way is kind of communist too. Innovation comes in the strangest of ways and code is no different. There have been real and ridiculously faster executed code in some of the less documented instruction sets.

The main problem is of course, that once the chipmakers burn it into their designs its permanent. And once a handful of programmers use the instructions, the chip designer is obligated to keep support for it - even if it is an instruction that sucks donkey balls.
 
OpenCL yea that will never work....
Correct me if my history is wrong, but didn't ATI embrace OpenCL after it abandoned its own Close to Metal PI? How is this analogous to an open standards committee? Seems more analogous to AMD sending up the white flag and doing things Intel's way.

eeyrjmr said:
USB is an open standard
TCP/IP is an open standard

Two examples of adopted standards don't prove your point. That is anecdotal, inductive. TCP/IP is 30-years-old and wasn't established by some committee. Plus, it was a first entry into a then non-existent market (two nodes between UCLA and Stanford didn't exactly call out for competition). I think I could find some experts who believe there are some inherent security problems with it. But it is kind of hard to change the rules of the road 30 years later, once all the freeways are built.

USB, is that really the best possible standard, or just the standard we have all adopted because or Intel's 800-lbs gorilla power? Firewire is arguably as good or better, but it was pretty much killed by USB. So USB might be an open standard, but it may have killed better technology. And how many other technologies haven't even been developed because of the hopeless barriers to entry?
 
When the goal is clearly defined, I think the standards committees tend to come up with very well designed and tightly specified standards, there's a looong list of them that you use every day. The process may take longer than a private company developing some tech and releasing a product quickly, but these technologies tend to have a lot more warts and over time usually end up getting replaced with an open alternative that is better designed. Short term the proprietary technology might not have a chance as it's only marginally better and has all of the downsides associated with being tied to an individual company. It doesn't make business sense, and that is the reason these hypothetical superiour technologies won't be successful; were they to provide their technology openly for all to implement, and if it were a compelling product, it's got a chance. But a marginal improvement over an existing standard isn't an improvement at all when you've got one company running the show.

I think you're looking at this the wrong way around. There are a lot of good technologies that might never have seen the light of day if they didn't have a consortium and standards body behind them to set in stone how they were going to work. Do you really think the Internet would have sprung up if not for a well-defined standard on how computers can communicate among each other? The Web with HTTP and HTML? E-mail? I don't think any of these would have reached any sort of widespread use if not for the standards backing them.

Firewire is arguably as good or better, but it was pretty much killed by USB.
I don't think you could have come up with a worse example to prove your point. First of all Firewire is IEEE1394, designed by committee and an open standard, arguably more inclusive than USB that was designed by a more private committee. Furthermore, Firewire is too expensive to implement for cheap devices like mice and keyboards, costs more on the host side as well, and since its capabilities are fairly similar to USB, using the cheaper interface as the ubiquitous one makes more sense. I don't think it had much to do with Intel's involvement, but more that it just makes more sense.
 
So an open standards committee-based product failing to get the best technology to market doesn't prove my point that open standards committee-based products are not the best way to get the best technology to market?

The point is, these committees are not a panacea. And the Firewire is too expensive to license is an urban legend. And had it been mass-adopted, the expenses would drop.

I don't think you could have come up with a worse example to prove your point. First of all Firewire is IEEE1394, designed by committee and an open standard, arguably more inclusive than USB that was designed by a more private committee. Furthermore, Firewire is too expensive to implement for cheap devices like mice and keyboards, costs more on the host side as well, and since its capabilities are fairly similar to USB, using the cheaper interface as the ubiquitous one makes more sense. I don't think it had much to do with Intel's involvement, but more that it just makes more sense.
 
So an open standards committee-based product failing to get the best technology to market doesn't prove my point that open standards committee-based products are not the best way to get the best technology to market?
No, you're comparing two open committee-based products that are competing, but trying to make the argument that there's something bad about committee based products. I'm not sure how comparing two committee based products, one very successful, is furthering your argument that there's something wrong with committees.

The point is, these committees are not a panacea. And the Firewire is too expensive to license is an urban legend. And had it been mass-adopted, the expenses would drop.
Not a panacea for what? Anyone can implement either of these standards, and both are widely used and successful. USB has emerged as the dominant player for general use, and Firewire is used where it makes sense. Yes, they overlap, but they are quite different in some important ways as well. In either case you're not tied into a particular vendor or software package because of your choice, which is what happens when you have proprietary standards or extensions like this article is discussing.

It's not just the licensing, which is a factor for cheap devices. Firewire is much more complicated and requires a lot more logic in both the device and the host. If it were mass adopted of course the cost is going to drop, but USB being simpler and cheaper to implement gets this advantage as well.
 
. so why don't we do it? Because we're stuck in our current quagmire of a system - the problem is not what should we do but how we'll get there.

And you just cant consider its a terrible idea? How about we have some discussion before you go solving all the worlds problems YOUR way on our dime.
 
Back
Top