ARM — Heretic In the Church of Intel, Moore's Law 390
ericatcw writes "For 30+ years, the PC industry has been as obsessed with under-the-hood performance: MIPs, MHz, transistors per chip. Blame Moore's Law, which effectively laid down the Gospel of marketing PCs like sports cars. But with mobile PCs and green computing coming to the fore, enter ARM, which is challenging the Gospel according to Moore with chips that are low-powered in both senses of the word. Some of its most popular CPUs have 100,000 transistors, fewer than a 12 MHz Intel 286 CPU from 1982 (download PDF). But they also consume as little as a quarter of a watt, which is why netbook makers are embracing them. It's 'megahertz per milli-watt,' that counts, according to ARM exec Ian Drew, who predicts that 6-10 ARM-based netbooks running Linux and costing just around $200 should arrive this year starting in July."
Too bad the CPU isn't the only thing drawing power (Score:5, Informative)
I love ARMs... (Score:5, Informative)
They are the only chips that you can program and keep your sanity.
The ARM code is just beautyful design, one weeps with joy after struggling through x86 hell.
And computing/electric power ratio is fantastic.
Re:Too bad the CPU isn't the only thing drawing po (Score:3, Informative)
And a 1 GHz Cortex-A8 core is probably in that ballpark.
Perhaps. A 530MHz Dothan was about twice as fast as a 600MHz Cortex A8 in a benchmark I saw. That does not mean the A8 is slower for browsing, as a browser is so complex that a simple CPU bench isn't enough. One has to sit down and use the system.
Re:MIPs? (Score:1, Informative)
Speaking of the architecture, it looks like MIPS already beat ARM to the netbook scene:
http://www.geeks.com/details.asp?invtid=ALPHA-400&cat=NBB/ [geeks.com]
$169.99
* Alpha 400 MIPS 400 MHz Ultralite 7-inch Mini Notebook
* General Features:
* Ultralite notebook
* Netbook form factor
* Linux 2.4 Operating System
* MIPS XBurst 400 MHz 32-bit CPU
* 128 MB RAM
* 1 GB NAND Flash Storage
* 10/100 MB Ethernet interface
* 802.11b wireless
* Supports External Hard Drive up to 160 GB
* Supports SD Card up to 32 GB
* Xiptech application software packages (Xip office, Flash player)
* 7-inch digital panel 800 x 480 true-color
* Keyboard with TouchPad
* Supports File Sizes up to 8 MB
* Built-in SD Card slot
* Battery Charging Time: 4.5 - 5 hours
* Uses:
* Internet surfing
* Instant online communication, chatting
* Music downloading and enjoying
* Flash movies and games
* Picture and image sharing
* Languages learning
* Personal diary
* Office Assistant:
* AbiWord, XipTable and PDF Viewer
* E-mail management
* Daily work plan and management
* E-book reader
* I/O ports:
* Three (3) USB ports
* RJ-45 Ethernet port
* Headphone in
* Microphone in
* Dimensions (closed):
* 1.1 x 8.25 x 5.6-inches
* Regulatory Approvals:
* C-Tick
Package Includes:
* Alpha 400 MIPS 400 MHz Ultralite 7-inch Mini Notebook
* Linux 2.4 Operating System
* Power Adapter (100 - 240V 50/60 Hz)
Additional Information:
* Notes:
* Model: Alpha 400
* Requirements:
* Available power outlet
Misunderstanding Moore's Law. (Score:2, Informative)
Moore's law isn't about what shall be done, or what should be done. It's about what can be done.
The size of a transistor on silicon has been steadily shrinking ever since the introduction of the first silicon chip. That's been driving down prices, and raising efficiencies. For example: the 6502 drew up to 160 mA at 7V. That's a little over one watt of power. The most power hungry Intel CPU on the market draws about 150 times that amount, but can do well over 150 times the work - getting hard facts is difficult, but I'd suggest four or five orders of magnitude, if not more.
So with the shrinking transistor, you can do three things. You can make the CPU more power efficient - able to do the same amount of work with less power. Or you can make the CPU more powerful - able to do more work, for the same amount of power. Or you can do both - able to do a bit more work, for a bit less power.
Intel has chosen to make their CPUs more powerful, at the cost of keeping the power usage high. ARM has chosen to make their CPUs a little more powerful, for a bit less power. Both are equally valid paths.
In the long run? Both choices will carve out their niche in the market place. There'll always be room for computing power at any price. But for the typical Joe Blow off the street, the ARM tradeoff - less performance, at a lower price - is more likely to be useful, assuming the software is there ... and with Linux, it pretty much is there.
Re:No laws overrridden (Score:3, Informative)
Re:They think a bit differently (Score:5, Informative)
Remember, with a netbook, you don't gain much by lowering the CPUs power consumption to less than 5 watts or so. The reason for this is simple, the display, ram, hard drives and everything else consume enough power that it won't really help battery life very much. I can imagine though that a quad core ARM A9 at 1GHz would make for a really nice netbook. Having multiple cores is nice on those for web browsing (playing flash in the background of your tabs, etc), and also for many media tasks. It would also be great if they included a graphics chip (or gpu as part of a SoC system) that could handle h.264 decoding for the netbook.
Phil
Re:Nonsense. (Score:3, Informative)
Looks like you got those numbers from the article rather than the spec sheet.
The power consumption listed is off a bit. That 100k transistor CPU only uses ~5 miliwatts load (less when idle), which is 0.005 watts - or averaged, ~0.002 watts.
However, most companies designing SoCs from it would embed tons of other stuff in the chip, like a GPU, USB controller, networking, etc. etc., so power consumption might increase to almost a watt when they're done, if everything is active.
I know Intel likes to boast about its roughly-one-watt CPUs, but they really have nothing on ARM as far as power consumption... and ARM has nothing on them as far as performance - but luckily Intel's stuff is so insanely fast that even at 1/20th the speed, speedy ARM SoCs are fast enough to run a desktop OS.
Re:Nonsense. (Score:2, Informative)
Go back a couple hundred years and people believed all sorts of weird things. Baths were bad. Bloodletting was good. The moon's made of cheese, earth's flat, earth's the center of everything, We can reach the moon/planets with a giant cannon, etc...
Even nowerdays people believe wierd nonsense and myths... like that one about how people once believed that the earth was flat.
http://en.wikipedia.org/wiki/Flat_Earth_mythology [wikipedia.org]
What else that is commonly believed will turn out the be wrong?
Re:1982?!!??! (Score:2, Informative)
Shame I already used up my mod points...
Re:Too bad the CPU isn't the only thing drawing po (Score:5, Informative)
The trick is that the ARM instruction set is *WAY* more efficient than the x86. The fact that the current ARM's are basically in order units is less important due to the design of the instruction set.
A minor correction here: The ARM instruction set is simpler and is much less feature rich than x86. This, combined with self imposed limitations (mostly around in-order execution) ARM (the company) is able to design CPUs with a much lower transistor count than x86 chip designers are capable of managing.
Not having a huge instruction decoder, having to do instruction reordering, or basically doing any of the things that makes x86 so damn fast, and staying a generation or two behind on manufacturing techniques to avoid leakage issues, enables ARM CPUs to have their amazing power profile.
The instruction set is kind of a mess really. It has been hacked onto a number of times, with the latest additions really showing signs of having wedged into the existing instruction encoding space. With features like the original (crappy) thumb, and now the fixed "Thumb2" (which we are all supposed to be calling Thumb and ignore the old Thumb, or something like that), and the fark-up that is ARM's floating point support (They have 3 implementations, 2 of which are still in use, and those two versions respond dramatically different to some basic key floating point operations), the ARM instruction set isn't nice per say, but ARM did the right thing and by keeping their eye on power consumption always.
(For those who are getting linguistically confused: ARM the name of the company, the name of their CPU line AND the name of their instruction set. Oh it is also the name of their reference manual, the ARM-ARM.)
Re:Why is it... (Score:3, Informative)
If we didn't use the x86 instruction set we'd have to invent something quite similar to it. There are several layers of cruft in there: 2-op instructions and 100s of rarely used complex ops. But it does something that modern 3-op RISC-like ISAs fail to do - it compresses executable code. This means more sits in the instruction cache, and so less decode penalties.
The newer ISAs (Core-2 and i7) do an amazing job of extracting parallelism to boost instruction throughput. During the Mhz race years instruction latencies spiked on the P4 ISA, the more recent ISAs have really lowered those latencies to RISC-like levels.
I don't know if you've ever spent time doing low-level profiling on an Intel platform, but they do make good hardware. Here's a rule that they don't bandy around, but if your working set fits inside L3 then all of your memory accesses are free. Yes - free. I can see why they wouldn't want to make that guarantee part of the ISA, but it does mean that register pressure (the biggest problem in x86) is a phantom issue. Doing low-level paging between registers and the heap (which is what a modern compiler does with load/spills) provides a huge working set.
The article mentions efficiency - but there are plenty of situation where power consumption is irrelevant and I really do want maximum performance per thread. Any home desktop falls into this category. As any server or laptop doesn't it will be interesting to see what happens as Intel shifts from maximising performance to maximising performance per watt.
Re:Too bad the CPU isn't the only thing drawing po (Score:3, Informative)
One motivation for putting an ARM in a netbook is to make a better product (overall coat-performance-battery life trade-off), not just a cheaper one, so why would a manufacturer not put a decent/large amount of RAM in one? Have you checked RAM prices recently - it's practically being given away.
As far as web compatibility, note that the iPhone is ARM based and has a decent browser (youtube compatible since youtube switched to H.264 video), and incidently Adobe is trying to get Flash working on the iPhone...
FYI Google's new JavaScript VM is here in the Google Chrome browser... The JVM is called V8 and does indeed compile to native code when it wants to (JIT), and runs rings around other JavaScript implementations (such as that in Firefox) in terms of speed. Considering how simple ARM machine code is (it's a totally orthogonal instruction set)it's hard to imagine porting the JIT compiler to ARM would be a big deal. As far as security, JIT makes no difference.
Re:Why is it... (Score:3, Informative)
Well, in a word, no. The reason the Pentium IV sucked more than any other processor ever (in terms of price:performance ratio, power consumption too) is that it had an enormously long pipeline. Several pipeline stages are "drive" stages to wait for signal propagation across a chip (probably more literally to wait for inputs to stabilize.) The problem with a deep pipeline is that the cost for branch misprediction increases; not only do you have more wasted pipeline stages before your branch, but you have more potential instructions which now have to be retired from the pipeline because all other operations occurring in the same thread are now invalid. Intel went back to Pentium M which was derived from Pentium III to create the Core series of cores which we are now using.
Re:Nonsense. (Score:4, Informative)
Yeah, but even your average phone is more powerful than your average PC was in 1982...
Actually, when the ARM appeared in 1987 it wasn't touted as a low-power chip, but was developed by Acorn as part of a desktop workstation [chriswhy.co.uk] chipset that could easily show the x86s of the day a clean pair of heels. One of the first products was actually an accelerator card for the PC [chriswhy.co.uk].
(Sorry about the PDF links: the parent site is http://acorn.chriswhy.co.uk/ [chriswhy.co.uk])
The Archimedes/Risc-PC "workstations" stayed in production to the late 90s (and there have been Amiga-style holdout products until very recently) and were always decently fast - but they couldn't compete with the Wintel dupooly and started to lose out when FPUs and, later, accelerated graphics cards became the norm on PCs. By this time, ARM had been spun off and had (wisely) started to concentrate on embedded systems which (at the time) didn't need such things.
Re:They think a bit differently (Score:3, Informative)
You got modded informative because you had nVidia [nvidia.com] in a link.
TI always pairs their Cortex CPUs with beefy DSPs capable of very complex decoding. The OMAP 3530 [ti.com](in use in devices right now) is able to decode 720p h.264 by offloading it to the DSP. A DSP is similar to a GPU, but this one lacks floating point capabilities. It's just really fast for integer stuff.
They'll probably pair an even faster one for the Cortex A9's, enabling 1080p h.264.