We build what we love (or "how to save us from an x86-only world") | Hacker News Reader

We build what we love (or "how to save us from an x86-only world")

dieblinkenlights.com

26 pointsrbanffy17 comments

How many applications that run only (or, at least, better) on non-x86 boxes were developed in the last couple years?

The answer to this question is that a vast number of applications that only run on the iPhone's ARM have been developed in the last few years. Also a large number of web applications that run fine on on both ARM and x86 have been developed.

The dying architecture is actually x86 because Windows is becoming irrelevant for consumers. Not overnight, but it's happening. OS X runs on x86, but I wouldn't be too surprised if Apple some day moves their laptops to ARM, considering they have to do something interesting with their new, homegrown chips.

Agreed: ARM is in no danger of going the way of the dodo. It's the dominant platform on mobile devices.

Unfortunately, the ARM's dominance can have a negative effect in the diversity of the segment, thus slowing down evolution on mobile platforms. Still, until app availability weights the market down, mobile is more resistant to fragmentation and friendlier to hardware diversity.

I would be delighted to see an ARM-based full-featured notebook, running an open and portable userland, but I fear Apple is not a company that will favour diversity. Given time, we may end up with the same problems we face today.

We need more than one alternative to x86. We need a fragmented market.

I cannot see what benefit a fragmented processor instruction set market would bring to anyone. It was more fragmented in the past and as far as I can tell not much was different then except that even fewer programs could run cross platform or virtualized.

If we move away from a single binary architecture without compromising a software compatibility layer (that doesn't need to be at the binary executable boundary), we become free to experiment with other technologies that have not been explored for a couple decades now because they would not be compatible with our software.

It's a little embarrassing the coolest processors we have today are more or less binary compatible with processors of the 70's and 80's and that the operating systems they run follow fashion that was conceived at that time.

I have said it before: where are massively-parallel, hardware-assisted garbage-collecting connectionist machines of today? "Nowhere", we say, mostly because they can't run Windows.

What happened to Alan Kay's "if you are serious about software you should build your own hardware" idea?

And just imagine how happy would Intel and AMD's engineers if they could make all the silicon around the execution units of their processors do useful stuff instead of emulating past 8086's 80286's, 80386's, 80486's, and Pentium I, Pro, II, III, 4, MMX, SSE instructions (and quirks) for binary compatibility's sake.

It could be good under the assumption that other technology - compilers, virtual machines, virtualisation, web apps - has moved on to the extent that a wider variety of instruction sets could bring benefits rather than problems.

One could add that many of the compiler advancements happened to counterbalance binary architecture limitations.

Contrast aggressive register optimization for function calling on most RISC boxes with SPARC's moving register window or the hoops you have to jump through on register-starved instruction sets like x86s.

The MIPS architecture is also not going anywhere, but its use in desktops and laptops computers is still minimal. Since ARM seems to be gaining popularity in the personal computer market where it used to have no market share I wouldn't be surprised if MIPS started showing up as well.

That would also be interesting. A couple years back, Movidis has shown a MIPS-based multi-core server that had much better than x86 (at the time) power-per-throughput numbers. It was very Niagara-like.

I don't get the point of this article. After all, the OS is an abstraction layer other the hardware and anyway, most cool developments happen in software, not hardware nowadays.

Most cool developments happen in software (and have for the past decade) precisely because we have standardized our hardware.

Look how long it took for multiprocessors to become mainstream. It's no coincidence it only happened after Windows NT displaced Windows 9x as the most common desktop OS. In fact, it only happened when Windows 9x's share became neglectable.

Yes, but hardware still matters. Look at virtualization - yes, VMWare and co mostly make it work, but it's slow, buggy and full of vulnerabilities. In short, it's a kludge.

I wouldn't describe VMWare ESX as slow or buggy.

I was under the impression that I/O performance, in particular, was still rather bad. The VMWare forums contain some links to http://vmfaq.com/?View=entry&EntryID=33, for instance, which suggests about a 50% performance degradation for most virtualization products.

Does this not apply (anymore) to ESX? I imagine it's improved somewhat, but AFAIK it can't match "real iron" speeds. I suppose that only gets worse as you add SANs, SSDs and other fun stuff.

As to "buggy" - well, I consider anything with repeated severe security issues to be buggy. Just search for "VMWare vulnerability".

Every x86 processors, however, is decidedly a kludge...

Yes, but you can hardly blame VMWare for that.

No. Some kludginess is, in fact, excusable on grounds of being a counterkludge to unkludge the physical processor. ;-)

What about stuff like power efficiency, which is 'cool' because your portable device runs cooler, has a smaller battery, and gets a longer runtime between charges?

Reply on news.ycombinator.com