As much as I love the idea of CPU cores fighting over the honour of its process threads and whatnot, surely we're talking about dual core here?
ARM's next-but-one processor will improve performance by five times, and the company aims to spread the architecture into the server side of the cloud too. The Cortex A15 will consume the same energy as today's ARM chips, but will sport as many as 16 cores running at 2.5GHz, though a dual-core version will come first. ARM …
When is someone going to market an ARM-powered motherboard in mATX or ITX form, with all the usual ports and connectors that one finds on a PC at a price no greater than an Intel Atom-powered equivalent? (SATA, USB, onboard video controller capable of supporting web browsers, office apps, etc).
If someone already does, please tell me. I'm starting to smell an Intel comspiracy.
One thing that they could change compared to PC boards is to dump that 24 pin ATX PSU connector. Give it a 12V input via a back-panel insert and a power brick, and onboard DC-DC conversion for 5V and 3.3V needs.
An ARM-powered Netbook would also be nice.
I have a home system based on an Atom D510 ITX motherboard with a huge convection-cooled heatsink but no fan, in an Aopen S135 mini-ITX case. It's as close to silent as any PC with a hard disk can be. It runs Linux. It's not quite fast enough when watching some video clips, but will do for now.
Looking to the future, I'd like it to (a) use less watts and (b) use a SSD. (b) is just a matter of time, they're too expensive for me at present. An ARM CPU would solve (a).
see the Sharp netwalkers for another arm based laptop.
The Palm Foleo was one as well, although of course it was abandoned
Freescale show a prototype:
the intel & AMD 64 bit cpus don't address 64 bits.
An MMU is fine. 32 bit Intels did fine on NT4.0 Enterprise accessing more RAM than many 64 Bit systems have.
Supported by NT4.0 and later, OS X and Linux.
x64 Physical addressing
The Address space is really 48 bits, not 64,
The upper limit on RAM that can be used in a given x86-64 system depends on a variety of factors and can be far less than that implemented by the processor. For example, as of June 2010, there are no known motherboards for x86-64 processors that support 256 TB of RAM
Ability to use up to 128 GB (Windows XP/Vista), 192 GB (Windows 7), 1 TB (Windows Server 2003), or 2 TB (Windows Server 2008) of random access memory (RAM).
Also with Intel architecture the 4 or 16 cores share the memory, meaning you need 2x to 16x as much memory. If each core has its own memory bus then memory access on a 4 core machine can be x4 faster and physical addressing can be be x4 less.
Address space on a Well designed 32 bit system isn't a problem especially if you have multiple memory interfaces on your SOC.
Beagle board ... UD$ 149 ... no DIMM socket, only 128Mb RAM, no SATA, No Ethernet that I can spot! Disk only via USB2 ... slow, and my experience with USB hard drives is that they always crap out. My Atom board cost about that with 4Gb DIMM, Gb Ethernet, SATA..
Old Acorn - won't be fast enough. ARM is just getting into useful desktop territory today.
It has 512mb of memory built in. It's "disk" is intended to be a flash card. It has an ethernet port. It's purpose in life is to be an open reference board for developers and the curious as an easy way to work with an ARM processor. It's also designed more for a small hand held or smartbook than to compete against a PC. Just look at the size: 3.25" x 3.25" It's clearly not for you. Go play with windoze on your atom, while Linux on the reference board runs circles around it.
IIRC ARM processors powered a perfectly useful desktop sometime around 1986. You can still buy machines that will run RISCOS.
In practice, an ARM based netbook would run most linux distros quickly enough to be useful. Running Linux on an ARM based system is easy - most open-source software will port trivially and several distros maintain ARM ports out of the box.
Winows is harder but technically possible. Given that Windows XP will run quite happily on late '90s vintage P2/P3 hardware, a 1GHZ+ ARM chip certainly has enough juice to support a port of something from the NT5 code base. Cortex based chips are not radically slower than an Intel Atom, and Windows 7 certainly runs on those.
The main obstacle to ARM adoption isn't Windows but the body of X86 specific Windows software. The NT code base already runs on 3 architectures (X86, X64 and Itanium) and used to run on sever others such as Alpha . Wintel is mainly interesting as an architecture because of the software base that runs on it.
Few customers will be interested in something that doesn't run their existing software well. Intel learned that the hard way with Itanium and let AMD in the door with the Opteron.
ARM chips cannot run X86 software except in emulation, so implementing a version of Windows that could run X86 software on an ARM is quite difficult. That's not to say it's impossible - Apple and DEC have done something similar in the past by running application code in the emulator and dropping out to native O/S calls.
Microsoft's management are almost certainly quite well aware that Linux will be far more competitive on ARM than on X86 hardware as they lose the inertia of the Wintel/X86 code base. It's not in their interests to support Windows on ARM, and previous RISC ports were not commercially successful.
At this stage, porting Windows to ARM would strengthen the position of ARM in a market which is a potential competitor to their core Wintel/X86 product line. They might do a port if market conditions forced them to but it's not in their interests to do one now.
One could speculate about the possibility of a clandestine R&D project to have an ARM port in the background should they need to make a product based on it, although I haven't seen anything in the press to suggest that it's happening.
"the intel & AMD 64 bit cpus don't address 64 bits."
They do virtually.
The only reason the current physical address space isn't larger is because it doesn't need to be as there's no physical machines that can take the memory, so why waste resources.
PAE is a bandage, like the old near/far pointers in 16bit days. It does work, and in the case of old NT4 systems worked well as back then it was unlikely you'd have a single piece of software accessing more than 4G.
These days desktop apps are increasingly required to deal with larger and larger files and the increased address space offered by a 64bit ISA can be useful.
I wonder why?
I was never really sure why MS avoided porting real windows to ARM. It was either a hangover from its agreements with Intel or more likely a complete inability to get windows to work properly on the arm due to windows labyrinthine code and lack of documentation.
I wouldn't be surprised to see MS being ARM's biggest customer and very little hardware appearing as a result. And then ARM being bough up by MS as an attempt to stop linux, Sakshat et all.
they can still build with the ones they've paid the licenses for but if MS can persuade ARM to do versions for them. If you cant build competitive devices for the 5 years or so it would take for the courts to convene on an anti-competition case it doesn't really matter if you have license for out of date technology.
ARM would be an expensive purchase or an exclusive contract would not be cheap but the point here is MS have a lot more to loose than ARM have to gain. Linux on ARM as a netbook/smartbook/smallcomputerwithapretentiousnamethisweek running at PC speeds and lasting a good day of normal use could be the straw that breaks MS.
Now how much is ARM worth? And how much did MS make last year? So for MS to purchase another 20 years market domination for a years profit is definitely in their interest, and a lot of supporting companies.
But definitely not us - the customers and governments and businesses that are effectively forced to use their wares.
You're in dream world. There's absolutely no way that MS could buy ARM. This was done to death at the time rumours began to circulate that Apple were to mount a takeover bid or take a large chunk of them. From a regulatory point of view, it's impossible. ARM CPUs and chipsets are by far the most numerous on the planet, and for ARM not to be independent would be completely disastrous for competition, also it'd be very bad for their business.
I believe that would be the reason.
They'd get it to work I suspect if they had sufficient motivation. The way I see it, nobody's made a decent arm desktop since the archie.
They did get windows NT on the power pc once but I can't remember the exact reasons why they abandoned it at the time.
NT (and Win7 is NT 6.1.xxxx, first version was NT3.1 in 1993) had at various stages:
386 (won't run on 386 now)
pentium Pro + PAE (Win9x was like a dog on this and killed the Pentium Pro)
DEC 64bit Alpha (NT4.0)
NT4.0 was last version to run loads of non x86
In the 1990 to 1996 era the ARM wasn't very suitable for NT.
However ARM has always been the main cpu for CE (spawning Phone, Auto, Mobile Editions). At one stage ARM became the only supported CPU for Mobile or CE.
"Embedded" Windows is confusing. There are three incompatible families
1) CE based and runs on ARM, There is a version 7. A GUI was optional
2) NT based (NT4.0 was best). There is a version 7. used to Support MIPS and PowerPC, now only x86-x64. A GUI was optional
3) Win 3.x/DOS based. 386 or more (maybe 286 possible) . Killed recently.(Last Autumn?) Used POS and ATM.
The original issues with ARM in 1990 to 1996:
Though technically a 32bit CPU, in some respects more like a 16 bit RISC. Thumb-2 in 2003 added more 32bit goodness
Originally no 16 bit type. Not good for Porting Windows efficiently.
Originally No MMU
Originally No FPU (not actually a huge issue)
Current higher end ARM could run a ported version of Windows NT. Rumour is that the reason some bits of Win2K, XP, Vista and Win7 are the same is because MS doesn't have the source. Only x86 binary. I don't believe this myself.
Tux knows how to run on ARM, MIPs, x86, x64 and few other things... It's those happy feet.
"Though technically a 32bit CPU, in some respects more like a 16 bit RISC."
As a RISC design (24000 gates in the original ARM) the architecture has been 32 bit (in data, in 1990 it only had a 24 address range. AFAIK this is now full 32 bit and has been for some time) since day 1.
"Thumb-2 in 2003 added more 32bit goodness"
Thumb is a way to increase code density (important with embedded systems where more code space -> more flash -> cost) by introducing 16 bit long instructions, mostly by accessing fewer registers.
"Originally no 16 bit type. Not good for Porting Windows efficiently."
I think you'll find that's an issue for compilers and assemblers.
Originally No MMU
Available as a separate chip since the early 90's, on chip (as part of your IP deal as ARM do not mfg) since the late 90's at most.
Originally No FPU (not actually a huge issue)
Again available as an on chip option for at least the last decade.
IIRC NT had the ideal of a "Hardware Abstraction Layer" designed to insulate most of the OS from the nitty gritty of the processor's registers, stack arrangements etc. It's unclear how far MS have maintained that.
ARM's are quite capable of supporting Windows *if* MS ported it.
"ARM's next processor, the Cortex A9, won't be shipping until the end of this year, with devices expected during 2011."
really? Nvidia's Tegra is ARM Cortex-a9 based and has been shipping for a number of months already. I believe I also read that there will be a number of tablets available this year based on it. And could Nvidia really be the only ones with shipping a9 parts?
and I'll say it again. :)
When native Windows applications are running on a virtual machine sitting on top of .net and the core OS is .net based, that's when you'll see why MS are ARM licencees. Converting the Jazelle technology to execute MSIL byte code instead of Java byte code would give .net a much needed performance boost in all areas - mobile, games, and future desktops.
That said, the lack of 64-bit support is a real killer for ARM at the moment. They need to solve that - and in a real way, not in the lame bullshit way they're attempting to at the moment. Cutting out the silicon needed to support the Thumb instruction set would also make sense for desktops where there are not such tight memory restrictions.
sT0rNG b4R3 duRiD,
I seem to remember that NT was available, for a time, on PPC, Alpha, and MIPS processors. The Xbox 360 OS is a cut down version of the XP kernel which is running on PPC.
"When native Windows applications are running on a virtual machine sitting on top of .net and the core OS is .net based, that's when you'll see why MS are ARM licencees. Converting the Jazelle technology to execute MSIL byte code instead of Java byte code would give .net a much needed performance boost in all areas - mobile, games, and future desktops."
A charming idea marred by 1 problem.
MS has *no* friends in the software industry.
Going into processor supply puts them *directly* into competition with AMD and Intel and the hardware suppliers who have been at their mercy for *decades*.
Being no longer a partner there is no longer *any* reason for them *not* to partner up with any other major OS provider. Even bank rollling compatible software to run down MS's market share.
I suspect many a board and peripheral mfg would quite welcome the chance of a good kicking^H^H^H^H^H^H^H more level playing field.
The bald one is *far* too smart (in market terms) to do something so stupid to his company.
The Arm11, although generally believed to be inferior to the A8, when combined with a good memory architecture (which most implementations don't have) can perform as well as the A8 but at lower processors speeds/power requirements. However, 'customers' hear the A8/9 is faster, so that's what they want.
The fact that AMD & Intel address bus is not 64 bit has little to do with what the industry is harping on about. A 32 bit CPU can only address 4 gigs of ram, additional memory can only be accessed via some slow paging system.
A 64 bit CPU can have a much larger linear address space which is handy for the big apps like Databases and Heavy graphic manipulation. AMD 48bit address space allows for a lot larger (I think 256TB) to be addressed.
I suppose the reason why a 64bit address bus has not been configured on these processors yet is that there isn't really a requirement to address this amount of ram in the first place.
16 cores running at 2.5GHz using the same same energy as today's ARM chips!
I bet that is secretly getting Intel's attention even if they won't admit it!
Its about time some serious competition came along to the bloated x86 processor design. Intel for too long has layered more and more complexity over decades of legacy processor designs, resulting in too much energy wasted and too much chip space wasted in their ever growing processor design. That was bad enough years ago, but now with the move to ever more cores, Intel's design has to duplicate a lot of bloat for each core. That means they struggle to put many more cores in to their designs and when they do, their designs end up eating a lot of power. That's opening up a big opportunity to take them in both numbers of cores and overall power usage. So I bet Intel are watching ARM very closely.
I can't wait! :)
(I wonder how much power globally we are wasting on x86 designs?). Surely sooner or later Intel have to break from X86 designs?
One problem, Intel have to be legally blocked from ever buying out ARM, because if they do they will sideline ARM so they can keep on selling their bloated designs for years to come.
If ARM play this right, over the next decade they could even become the new Intel, with the majority of Intel's profits, because Intel's designs are in trouble. I know Intel control the PC, but ARM sell way more processors already and mobile devices of this kind of processing power are a serious threat to Intel's future position.
I really can't wait! :D
As a software publisher originally for the BBC and later the Acorn Risc Machine (ARM) my company was in at grass roots level the start of the ARM story. There were so many things that could not be done, according to conventional wisdom. Thank goodness for the band of innovative and often self taught programmers who made the Archimedes (generation 1) and later RiscPC platforms, do things they were not intended to do. It was due to the largely self taught nature of many of these guys that produced the CAN DO attitude the UK is famous for.
My guys produced a simulation of the ARM chip running BBC Basic on a BBC Micro. This was before there was any silicon and was based solely on the specification from Acorn's brilliant team. They wrote applications that worked first time on the pre production prototypes!
I would love somebody to look into the early background of the ARM chip with emphasis on the early usage of ARM computers and the Acorn RISC OS operating system. There were many rumours of systems being used in high security applications because of their speed but also because nobody bothered to write viruses for RISC OS. The Hong Kong stock exchange was one name that kept cropping up.
There is a fascinating story to be uncovered and I wonder how many of the old programmers are still around that could squeeze more out of these processors than convention would anticipate. These guys often wrote in machine code but to get programs to market quickly they would often use BBC Basic coupled with machine code where the speed was needed. It was not until the C compiler arrived that we started getting the BLOATWARE so common in the PC market.
Current software such as the XARA series on PC were firmly rooted in the old Acorn market
Anyone want to take on a project?
You know, I'd love to do some kinda docu about ARM, Arcorn and Psion to show how close the UK came to being a massive force in the tech industry.
It's such a shame that poor management has always brought down all potential UK tech giants save for ARM. Imagine where we'd be if Psion had capitalised on their position in the PDA market and Acorn had kept the RISCPC afloat.
If you fancy getting in touch, drop me an email by using my username twice, sandwiching an @ inbetween and sticking a .com on the end ;)
Yep, I was there in the late eighties inlining ARM asm in BBC Basic. ARM was the first non-basic language I taught myself. It was a few years later that I switched to C during the first weeks of my first real job.
I used to have so much fun programming ARM, and probably still would do today. It's one of the nicest architectures I've ever had the pleasure of using.
"My guys produced a simulation of the ARM chip running BBC Basic on a BBC Micro. This was before there was any silicon and was based solely on the specification from Acorn's brilliant team. They wrote applications that worked first time on the pre production prototypes!"
Dick Pountain mentioned this on the original Byte article on the ARM. I had thought they had written it in house prior to first silicon.
Every time Intel try to find another money-maker outside their x86 comfort zone, they shoot themselves in the foot, don't they?
So yes they need to, they have even tried to, but so far all we have is a string of disasters, from iAPX432 through I2O through integrated graphics through WiMAX to IA64.
I wouldn't bet on whatever's next either, especially as (courtesy of the US Supreme Court) we now know officially what some of us have suspected for many years, ie that the Dell/Intel relationship was as much about bribery, blackmail, and false accounting as it was about having Intel do all Dell's R+D for them for free.
Biting the hand that feeds IT © 1998–2019