are matrox short of engineers?

[Himself]

If the world were a better place, things like SDE would have caught on ages ago. No need for open source, just use cpu independent binaries with a cpu specific install time compiler. An application not optimized for your cpu, not a problem. File sizes too large for distribution on the net, again, not a problem. Is it capable of being backward compatible? Yes. Is speed an issue? Not likely with everyone using 500MHz+ cpus. Who wouldn't mind waiting another minute or two if the result were a tailored executable for your cpu? People spend hours tweaking for games as it is. All you really need is a compiler back end that spits out the appropriate intermediate format and a translator for your cpu. Hmm, what company supplies the most popular compiler for Windows? What company could add this into the Windows operating system transparently? Could it be Microsoft? Why haven't they done this already? They even help fund SDE, it doesn't make sense. Instead you have them working on stupid things like 3D desktops, which no doubt will require you buy yet another version of the operating system.

[Salmonius]

Well, I would say that AMD and Intel engineers are geniouses. They have done much more then just increase the clock speed of the processors. The underlying architectures of the pII and Athlon are very complex and optimized, much like those of RISC processors, except probably more.

The real problem with x86 ISA is simply the lack of registers. However, AMD will basically double the number of general registers with x86-64 so that will not be an issue so much anymore.

Another problem to the PC platform is the weak FPU. Both AMD and Intel have had phenomenal success on getting the most possible out of the crappy x87 architecture. Assembly programmers will agree that x87 is really bad! Fortunately, AMD will provide a totally new FPU architecture for x86-64, and of course still provide the an x87 interface.

In my opinion, AMD's 64 bit processor is going to be *much* faster than anything we have seen to date. Soon there will only be AMD and Intel processors because they are going to destroy the rest (Ultrasparc, Alpha, etc).

[DGhost]

Platform independant binaries? sounds like you are advocating java. Carmack tried doing something similar in Q3... people still would take the source, compile a platform native DLL and use that because it was faster - by a very minimal margin. People are not interested in interoperability and code compatability.

Another point. Look at Linux and its platform 'independance'. Just because an app will run on the x86 arch doesn't mean it will run on the alpha arch. Even with a recompile. Things change from platform to platform - one line of code may be incompatable, or a whole subroutine. Linux is not a platform independant operating system. it started out as a 'replacement' for Minix, which was an x86 UNIX clone. Linux started out on the x86 arch. the 2.x kernels purpose was to bring multiple architecture support to the kernel, and it has. But x86 level kernel support is not available for the alpha, mips, powerpc, sparc or any of the other architecture. it has taken alot of work for the alpha port to be so stable, let alone the powerpc and 68k ports. What it does have is enough common code to ease porting between platforms.

Also, if x86-64 is just an extension to the x86 architecture like AMD says it will be, clock for clock an Itanium will beat it to the ground. When you consider the fact that the Itanium has the capability to execute up to 3 instructions per cycle in native mode, you are talking a fast processor.

And the alpha is still the fastest microprocessor architecture on the planet. The speed comes at quite a cost tho. Remember they had 500+mhz processors when intel was hitting half that.

[Wombat]

DGhost:
I think you're wrong about your FPGA proposal. Right now a key issue in processor design is getting the data from place to place. They can't even get the data across 2cm of silicon in a clock, there's no way a multi-die solution could hack it.

Salmonius:
The PII and Athlon processor ARE RISC processors.
The real problem with the x86 ISA is not a lack of registers (they've been tagging them on like crazy from generation to generation), but the variable-length instruction is a good first choice, among other things.

AMD and Intel aren't about to destroy anybody. If anything kills Sun, it will be Scott McNealy with his lies and piss poor business practices. If Alpha dies, it's Compaq's fault. IBM? They've got some nice stuff too. HP? Nobody is better than us (not that we don't consider other companies equals, just nobody better ). The AMD guys I've known are pretty smart, but I don't know too many. And Intel....well they aren't a processor design company, they're a fab.

[DGhost]

Originally posted by Wombat:
DGhost:
I think you're wrong about your FPGA proposal. Right now a key issue in processor design is getting the data from place to place. They can't even get the data across 2cm of silicon in a clock, there's no way a multi-die solution could hack it.

Yeah. i didn't say it would be easy... and that their wouldn't be problems... more of a concept than a 'this is what i'm gonna do'. still think that if you could overcome some of the technical problems it would be the fastest thing on the market.

Salmonius:
The PII and Athlon processor ARE RISC processors.
The real problem with the x86 ISA is not a lack of registers (they've been tagging them on like crazy from generation to generation), but the variable-length instruction is a good first choice, among other things.

AMD and Intel aren't about to destroy anybody. If anything kills Sun, it will be Scott McNealy with his lies and piss poor business practices. If Alpha dies, it's Compaq's fault. IBM? They've got some nice stuff too. HP? Nobody is better than us (not that we don't consider other companies equals, just nobody better ). The AMD guys I've known are pretty smart, but I don't know too many. And Intel....well they aren't a processor design company, they're a fab.

Very very very very very true. Alpha went down the tubes when digital was bought by compaq. IBM stuff is very fun =)

AMD has always been plauged by finacial problems. at this stage, if intel beats them, its because they can't get their act together internally, not for a technological reason. i know some people who have worked at AMD - they quit because of the financial problems and what not going on inside.

[Himself]

Platform independant binaries? sounds like you are advocating java. Carmack tried doing something similar in Q3... people still would take the source, compile a platform native DLL and use that because it was faster - by a very minimal margin. People are not interested in interoperability and code compatability.

That's exactly how SDE works, you compile to your target cpu when you install, it's not Java related. Picture all source code distributions, but you can't see the source and the files are halfway compiled for you, just waiting for the final stage to your paticular cpu target. The platform would still be the PC and Windows of course. Consider a transmeta deal, they could get rid of the middle man and convert code at install time instead of on the fly. This conversion would be a lot easier considering the lack of information about program structure you lose with a compiled executable, instead of backtracking through that they would have the original intent of the source. Consider the P4, instead of BS about future support of SSE2, they could have actual support instantly, they'd just have to toss off a SDE to P4 binary generator. What would be on the hard drive would be actual P4 binaries, not SDE code, the code would be in the installer. Another area MS would be ideally suited for.

This is not something new, I first read about it four to five years ago. A varient of it was/is proposed for inclusion with the AmigaOS project.

[alessandro]

Originally posted by Wombat:

The PII and Athlon processor ARE RISC processors.

Infact the Pentium Processor range is a 386 core with some RISC extencions...they are more like a CISC CPU than a pure RISC one.

[Wombat]

The Pentium was that way, but I believe that the transition was made with the Pentium Pro.

[alessandro]

OUPS,

My mistake......
Pentium Pro and Pentium II family of processors translates legacy x86 code into RISC instructions, executes them out-of-order, and then reassembles the result to match the original program flow.So they are more RISC than i thought

[Salmonius]

>The PII and Athlon processor ARE RISC >processors.

That was my point. The x86 ISA is just that, an ISA. AMD and Intel invest a lot of resources and energy into the development of CPU, so there is no doubt that they will end up with better processors than the other ones.

As for SUN, McNeilly has done a very good job, I think. SUN sells the most UNIX platforms by far and that is thanks mostly to the aggressive marketting that it has been doing for the last several years.

There are no rumors that SUN may buy AMD and that makes sense because they could save some cash in CPU development, and AMD is so cheap at this moment. The only problem that I can see is that they would have to use the Alpha bus and I am not sure how SUN would view that. SUN has already said that they are porting Solaris to x86-64.