On 01/26/2011 07:25 AM, Luke Kenneth Casson Leighton wrote: > On Wed, Jan 26, 2011 at 5:14 AM, Rob Landley <[email protected]> wrote: >> On 01/25/2011 09:42 PM, Luke Kenneth Casson Leighton wrote: >>> at 28nm it's going to be... irrelevant that the main RISC CPU is >>> 74,000 transistors (MIPS 64-bit) because it'll be running at 2ghz, be >>> running in a quad-core or even 16-core arrangement and... who gives a >>> damn if an x86 gets even 100% more performance at those kinds of >>> speeds! especially when x86 does so by having to still be a thousand >>> times more transistors and so uses vastly more power. >>> >>> ... or am i preaching to the converted, here? :) >> >> No, you are not preaching to the converted. Underestimating the ability >> to adapt a specific instruction set to take advantage of manufacturing >> improvements is something you'd think people would grow out of after 30 >> years. >> >> I'm saying that network effects mean the system with the most users is >> the one everybody wants to write code for, and the system with the most >> software is the one everybody wants to use. Costs are almost entirely a >> question of unit volume, it's all start-up amortized over a production >> run. When you say "RISC will make chips cheap", you're making arguments >> that people made 30 years ago and it simply did not happen. > > rob - with respect, you are making a lot of assumptions about what > i've said, and are placing words into my mouth that i haven't said, > and then making points supporting your position based on that.
I admit I'm mostly responding to a single phrase from your first message, and then your defense of that phrase. You said: > 45nm and the upcoming 28nm geometry is going to allow SoC RISC CPUs to > become massively more price as well as power efficient than CISC x86 > architecture could ever be. And saying "oh the competition can't possibly catch up with this huge advance we've made, our success is guaranteed" is something I find personally hilarious. (Side question: why can't CISC be manufactured on a 28m process, again? Hasn't Intel's main strength been manufacturing rather than chip design ever since the company was founded? Yeah, Atom ain't taking off in smart phones but that's not because of manufacturing process, it's because iPhone and Android were first to market by several years and have billions of seats and software "marketplace" distribution channels to ship to those seats. That doesn't help MIPS. Trying to inject MIPS into the smartphone market today is like trying to displace 8086 with m68k in the late 80's, and then PPC (apple) or Sparc (Sun) in the 90's.) > * take a look at the specifications for the upcoming version of > android (2.3), and the recent report from a korean CTO stating that > the minimum requirements to run it are a 1ghz *dual* core CPU... with > hardware-accelerated OpenGL ES 2.0. You know, it's really too bad Qualcomm's taken so long to release Hexagon support for Linux. That's the multimedia codec in the Nexus One and HTC Inedible, and it handles all the multimedia stuff via software rendering. They booted Linux natively on it over a year ago. That was kind of exciting for a while, but what have they actualy _released_ yet? https://www.codeaurora.org/xwiki/bin/Hexagon/ So somebody already shipped hardware capable of displaying full screen video mp4 and mp3 audio on a mobile device entirely via software rendering, it's been out for a couple years now, and you probably haven't even heard about it. That's how much of a splash it made. Hardware, by itself, does not determine what the market will do. > * two Set-Top-Box companies are already crossing over from > "set-top-box" into "general purpose" SoC CPUs. > > one of them has a 500mhz (MIPS) CPU with 1080p60 Video decode. the > CPU costs **** 6 **** - i repeat - *SIX* - i repeat, that's $USD 6.00 > - i repeat - that's six United States Dollars - in mass-volume. they > are now doing a 1.5ghz version. So what? Having 1080p decode these days is like having a floating point coprocessor, and $6 isn't the cheapest processor out there. You can get cortex M0 processors for 65 cents each if you're willing to buy in lots of 10k: http://ics.nxp.com/support/documents/microcontrollers/pdf/iq.magazine.nxp.launches.lpc1100.pdf ("If you build a better chip, they will come" simply isn't true. The rest of the system and software cannot simply be left as an exercise to the reader. Just ask the DEC Alpha design team.) > unfortunately this same company is a massive GPL violator, who are so > paranoid that they will not even let manufacturers of products > utilising their CPUs even design the PCB. A recipe for success, right there. > android is (unfortunately) what smashes the linux kernel into > mainstream prominence - exactly as you stated is required, rob, to > begin the cycle of acceptance of these much more cost-effective SoC > CPU solutions. No, the iPhone and iPad were doing it quite well on their own before Google shipped their _response_ to the iPhone, and other clones of those would be using the same basic Arm chip regardless. Arm has something like 90% market share in smart phones just like it had 90% market share in previous phones. And rewriting all elements of the Linux user interface from scratch in a non-open-source way is pretty much required to get end users. Open source sucks at UI design, it's a structural problem with the development model and I've blogged about that repeatedly: http://landley.net/notes-2010.html#13-08-2010 http://landley.net/notes-2010.html#10-03-2010 http://landley.net/notes-2009.html#09-06-2009 And so on, and so forth. (You wondered why Ubuntu was throwing out Gnome and starting over in-house? Why the small teams behind Firefox and Chrome displaced the large team behind Mozilla? Whether or not unity is actually any god, the attempt is a positive sign if you ask me, but far too late to matter. Windows 7 is a "good enough" 64 bit PC OS the way Windows 95 was a "good enough" 32 bit one: its users are no longer actively looking to replace it, and that battle shifts to the new hardware platform that will obsolete it. I think Android throwing out the whole of userspace and going with Java was a bit of an overreaction, but it's certainly one way of avoiding the "too many cooks" aesthetic and the "shut up and show me the code" look and feel we've had for so long with desktop Linux distros.) > no mention of x86 or microsoft was made at any time in > that sentence. Not in that sentence, no. This was the sentence in the email I was replying to that mentioned x86 and microsoft: > p.s. yes i know my history on microsoft's decisions to drop all but > x86 for NT 4.0, rob, due to pressure from the hated win95 team... But I have no actual objection to reverse engineering the Voodoo card so people starting using real FreeBSD on the their Macintoshes instead of Apple's rewrite of it. Er... s/FreeBSD/Linux/; s/Macintoshes/Androids/; s/Apple/Google/; s/Voodoo/PowerVR/ Rob _______________________________________________ Celinux-dev mailing list [email protected] http://tree.celinuxforum.org/mailman/listinfo/celinux-dev
