3 ms·

ARM's profit last year was 170 million pound last year, that $68 million per quarter vs intel's $3.2 billion. There are of course the big chip manufacturers ,b

• 6ren 15y ago
  I didn't realize ARM was so much less profitable than intel. It's partly because ARM's model of development is shared with its licensees (e.g. Apple's A5 is based on ARM, but customized inhouse, and I believe uses third-party components); so a fairer comparison of the ARM "business" would include all these designers, assemblers and distributors, as well as the chip manufacturers you mention. The ARM model is "dis-integrated", both in stages (design, manufacturing, marketing) and in components within a particular SoC. Intel is "integrated" on all those fronts (even to consumer advertising - those funky cleansuits, "intel-inside"). The other factor is that Intel is stunningly profitable, due to the wintel franchise. There's huge prediction issue here, of what will happen to the smart-phone market, that reasonable people could disagree about. I believe it will disrupt "PCs", just as PCs disrupted workstations/mini-computers, which disrupted mainframes (the latter of each may still exist, but only in the most demanding applications in the highest tiers). The reason is that as it improves in performance, it will become good enough for more and more applications (while the PC's improvement will make it more powerful than most applications need); and also offer other benefits: size, consumption, mobility (and multi-touch, GPS, camera). If you can plug your smartphone into your HDMI monitor, mouse and keyboard, and have all your information there, would you still use your desktop? As for the server, it's widely reported that power consumption is the dominant issue as they grow in size. SSD are of interest for this reason... and ARM cpus will be also. The crucial dimension is processing power/power consumption. Intel is not ahead on this score, only in raw performance per power-hungry cpu. But the really amazing difference is that because the ARM architecture is open to customization, it is possible to optimize silicon performance for highly specialized tasks - instead of writing time-critical code in C, or assembler, you write it in silicon. Or, more generally, you write time-critical modules in silicon, like crucial parts of what key-value stores are doing. I think we may see the rise of startups integrated along silicon-software lines - like Apple is. Oracle is trying to do this (with Sun's SPARC). Imagine the performance advantage you could get with this! You would annihilate the competition. I totally agree with your last line about the windows franchise. Once webapps and the cloud get their act together (I'm looking at you reddit/amazon), it seems that windows could indeed be obsolete (as long-forecasted). OTOH, incumbents traditionally win their markets, even through dramatic technology changes, because they know their customers (they have trouble when their customers change). But this is usually when the market demands more of the same thing... In the server market, power consumption has become more of an issue, over raw processing power, and Intel has already been adapting to it. But they don't have ARM's experience in low-power consumption, nor the flexibility of architecture to optimize for it. The only difficulty I have is that my analysis predicts that ARM has already won the server market... There's also an aging factor, in that the firebrand startup founders that made it all happen - andy grove, gordon moore, etc - have retired. Look at HP without H and P.
  • ippisl 15y ago
    Another way to compare prices for intel vs. arm: Intel's tablet hardware cost $75 for z670 cpu + $20 for chipset. this compare to $20 tegra 2. There doesn't seem much of an incentive for intel to sell $20 systems on a chip. And regarding the critical code sections becoming silicon: i don't think it's becoming silicon , because of the huge volumes that are require to manufacture silicon. but they might become FPGA code. FPGA is basically a programmable hardware. it has been used in the past as computer co-processor , in fields like biology, oil and gas and database accelerators. but the main barrier is that programming them is complex and it's hard to find FPGA programmers. But i see improvements in these areas. convey computer claims that it has made programming an FPGA as easy is c programming. it woud be interesting to have convey computer offer via the cloud. But let's return to the intel vs ARM servers: according to arstechnica , arm would find it very difficult to offer high per-thread performance at low power. but given a cheap enough arm server processor , and the help of GPU's(and FPGA or other coprocessors) intel could be seriously threatened.