About 3 years. By the time Windows 9 rolls around, Microsoft would migrate to ARM. If you want to run your old x86 Win32 programs, use an emulator. The FOSS world will move to ARM or MIPS because there is no binary compatibility problem that plagues closed source software. The legacy chip lost its benefits. Now that 64-bit code is popular, x86 doesn't even have the code density advantage anymore. It won't survive long with bigger code, less registers, irregular encodings, more baggage, useless decoder complexity, obsolete 16-bit modes, higher engineering costs, and higher chip prices. Even Intel has an ARM license these days.
>>4,5
Keep dreaming. Smartphones will remain the biggest ARM consumers for the near future. x86-64 was a bit of a setback but there's plenty of 32-bit applications around.
>>8
That's the first thing, and with Microsoft now having an ARM alternative that is bound to grow, x86 will be pushed off to niche markets. Not good enough for serious computing, not low power enough for casual computing.
>>4
Increasing the pointer size to 64 bits doesn't magically make the encoding for the register-register instructions less dense. And speaking of registers, you know they added more of them, right?
If you are willing to abandon binary compatibility, something like the X32 ABI could make good headway on tablet size devices. There's really no need to maintain PC compatibility for devices like that, either - just boot straight to protected mode and all those awful problems (mostly) go away.
Phones are definitely a lost cause for Intel at this point though. The market has already decided on ARM and Intel is nowhere near good enough to justify developers recompiling all their applications.
>>13 X32
What a confusing name. They should've chosen something like 64P32 or some other name that makes it clear this is actually running in 64-bit mode.
just boot straight to protected mode
The 80376 did that. It was not very successful.
There's really no need to maintain PC compatibility
You underestimate the strength of being able to market something as "IBM PC-compatible". An 8086 has 29K transistors, which is basically nothing on a die with several hundred million or even several billion these days; including backwards compatibility is so cheap that they would be stupid NOT to.
Phones are definitely a lost cause for Intel at this point though.
It doesn't matter because of Java. The Chinese have MIPS on their side. This is one of the areas where x86 isn't as good of a match, but you never know...
>>14
Those 29K transistors form the core instruction set for the other billion transistors. Prefix bytes and 2-register instructions are directly because of that ``compatibility''. It's better to use a good instruction set and emulate that obsolete set of hacks if they really need it.
Name:
Anonymous2012-10-10 1:56
for how long do you expect Jews to run the world economy?
for how long do you expect Christfags to worship circumcised Jew?
for how long do you expect Muslims to worship Yahweh?
>>14
Never knew about the 80376 - it's interesting that that was tried so early. When that chip was released, DOS still ruled the PC. I'm not so sure that's true anymore. A legacy free system running coreboot stays in real mode for maybe the first 10 instructions.
The compatibility bits may be cheap in terms of transistors, but you pay for them in other places. All the added complexity is very expensive to maintain and test. Ever wonder why the PC has a while cottage industry of software vendors writing chipset initialization code that all the other platforms seem to do without? Real mode code is expensive.
>>15 jewops basically are the core instruction set now.
>>14
Also - there's a large enough installed base of native applications on phones to ensure that x86 will always be at a disadvantage. Anything on Android that uses the Native API is tied to ARM until and unless Google decides to do something about it. For the time being x86 is just barely competitive there is no reason to force a change.
All the added complexity is very expensive to maintain and test.
It's nothing compared to everything else on the die. The 16-bit subset has been around for so long and is well-characterised to the point that they probably don't have to test that much.
>>18
Curious fact: ARM as implemented on the various iDevices does not have a hardware divide instruction. Even MIPS has one.
Name:
Anonymous2012-10-10 6:31
Anything on Android that uses the Native API is tied to ARM until and unless Google decides to do something about it.
Not true. The Android NDK allows compiling for ARM, MIPS, and x86. You can even put all three into one package file.
>>14
You might have noticed that the Chinese have also started churning out ARM processors for mobile devices. MIPS will be relegated to supercomputers until they decide to change their "national computer architecture" as well.
>>17
There's a whole industry writing firmware and bootloaders for other architectures as well, so that's a silly argument.
>>19
Even performance-oriented architectures like Alpha and IA64 don't have hardware division, it's not that important. And if you really want to, you can switch to Thumb-2 mode.
It's funny seeing people arguing about the benefits of one ISA over another. Do you honestly think modern x86 chips decode instructions each time, and not cache them decoded to uops?
x86, ARM and MIPS are all equally shit for modern aggressive superscalar reordering implementations. Instruction decoding is trivial compared to tracking dependencies and all the other OOE costs.
Name:
Anonymous2012-10-10 9:15
>>10
Too bad there's no open source toolchain capable of compiling HDL to bitstream.
Name:
Anonymous2012-10-10 9:40
>>23
No, quite the opposite. MIPS doesn't exist at the low end. But China's national strategy of computing independence is based on domestically designed (and manufactured, as soon as they have the ability) MIPS CPUs.
>>27
Many routers still use MIPS, but several of the router chip makers (eg. Cavium) have bought ARM architecture licenses as well. Sony gave up MIPS after the PSP (PS3 is PPC, Vita is ARM).
The goal of the Chinese was to design a CPU from scratch. The fact that it implemented the MIPS ISA was more or less coincidental. It was only later that they worked out a deal with MIPS and got an official license. By designing their own chips they don't have to worry about American export restrictions, they know the only backdoors are the ones they added themselves and so on. Currently the chips are manufactured by ST Microelectronics, but you can bet they're working on that as well.
Name:
Anonymous2012-10-10 22:40
so what's the conclusion?
Name:
Anonymous2012-10-10 23:01
>>29
you have no real waifu and your life expectancy is short.
Sadly, a small universe worth of SystemVerilog test benches means exactly jack shit when it comes time to write the BIOS. Getting every weird little dinglet on the board to talk in 1MB of address space is where x86 comes apart at the seams.
>>22 There's a whole industry writing firmware and bootloaders for other architectures as well, so that's a silly argument.
Name three companies that specialize in developing firmware for non-x86 platforms. Anyone on this board can probably name two PC BIOS vendors.
Only real mode PC compatible firmware managed to evolve into an enough of an asspain to be capable of sustaining multiple independent corporations.
>>24
Cache density and fetch bandwidth still matter.
>>28
The Chinese probably learned about MIPS from Western CS textbooks. It's cheap to manufacture and relatively generic-looking compared to ARM, so they can more easily sneak out of buying licenses. A lot of their mid-range and high-end media players use unlicensed MIPS SoCs made by various short-lived companies with obscure names. (The low end is primarily 8051, Z80, and 6502.)
Name three companies that specialize in developing firmware for non-x86 platforms
There are many, many more than BIOS vendors.
The reason why BIOS vendors have not proliferated is because there isn't that much demand for anything other than already provided.
Getting every weird little dinglet on the board to talk in 1MB of address space is where x86 comes apart at the seams.
BIOSes have their init code running in flat real mode ever since the 386. Coreboot wasn't the first to do it.
>>33
FRM can't be used by add in devices. Since you absolutely need at least a disk controller and a video adapter to boot, that is a fairly significant limitation.
Name:
Anonymous2012-10-11 2:26
>>33 sneak out of buying licenses.
Soon U.S. will invade China to establish a "democracy" there.
Name:
Anonymous2012-10-11 6:39
>>33
It at least used to be the case that you can't copyright an instruction set (dating back to lawsuits of the late 70s). Nevertheless ARM is known for litigating against reimplementers, whereas there was precedence that as long as you left out certain patented bits (unaligned loads and stores) you could implement MIPS without legal issues.
>>35 U.S. will invade China
And fight them with what, sticks and stones? The biggest irony is that while USA's retarded politicians, intellectual property artificial rarity industry lobbyists and helicopter parents are arguing about restricting essential digital freedoms, China's government is paying for the design and production of open source firmware architectures, which is, freedom-wise, further than any company in the ,,freedom-loving'' West ever made it. Don't be surprised if China someday gives the WIPO the middle finger and simply stops recognizing shit quality patents (as well as software patents).
>>37 Don't be surprised if China someday gives the WIPO the middle finger and simply stops recognizing shit quality patents (as well as software patents).
Chinks wanna get paid too, and unlike everyone in the West they still know how to play the long game.
Name:
Anonymous2012-10-11 13:00
>>39
The way they did it was, at least at the time.
>>42
Ignore me, I hadn't refreshed the page in a while
Name:
Anonymous2012-10-11 15:41
>>37 stops recognizing shit quality patents
Wait wait wait... China recognizes patents? With the amount of copy products coming out of that country you'd think they'd have violated at least one patent at some point.
>>44
At least they're currently saying that they do, just to shut up the West.
Name:
Anonymous2012-10-15 7:46
I always find these discussions intriguing, as ARM for example isn't responsible for any CPUs as far as I'm aware. PIC and TI sure seem to make a lot of microcontrollers and SoCs with their technology, but afaik there has yet to be a proper ARM ``desktop'' processor. Are there such devices in production?
Also, it would be cool if someone reverse engineered a ``large'' (over 100,000 LUTs) FPGA and (finally) produced a FOSS toolchain so that paranoid people like me can just make their own ICs with no fear of hardware backdoors as unlikely and easy to detect as they may be. I would totally write a Lisp machine.
Name:
Anonymous2012-10-15 12:31
>>46
They started out making workstation CPUs, but I don't think anyone is dumb enough to try competing with Intel on the desktop. There's a few companies trying to make server CPUs, but they're after low power, not performance.
>>49
The company in Intel's cross-licensing cartel? Do you really think they're competition and not just a company Intel throws a bone to every once and a while to keep the DoJ off their back?