>Everyone talks about how ground breaking the new Crusoe chip is. And, I >admit it has some impressive innovations and is a really strong >challenger to the mobile AMD/Intel processors, but how much of an >improvement is this technology over say the StrongARM or Super H chips >that have comparable performance/power ratios? A decade or more ago it made sense to consider alternatives to the x86. With the continuing maturation of a wide range of x86 software, of which MS is the dominant but by no means only vendor, the attractiveness of the alternatives has been on a steady downslide for more than a decade. So much so that we made the decision a year ago in our lab to focus on the x86 platform to the exclusion of all others. We *constantly* get the question, why are you going with this electron-guzzler of a platform when you could be using the StrongArm? The answer is very simple: the x86's software base is mature, and the enormous inertia of modern software makes this factor more significant than mere power considerations, even when lightweight power is important. For that one reason, the intertia of software, the StrongArm is a lost cause even without the Crusoe, which merely seals its fate all the faster. We also often hear that the StrongArm architecture lends itself to power optimization in a way that the x86 does not. While I didn't believe this, I admit I couldn't prove it. Thank you Dave Ditzel for settling the question. Just about any reasonable platform can be tuned for good performance/power by a sufficiently good computer architect, and there is no intrinsic reason why the x86 should have been an exception. The StrongArm got there first only because it was tuned for low power first, not because the x86's architecture is an insuperable obstacle to tuning for power. Jumping from strategy to tactics, can the StrongARM run at the equivalent speed of a 500 MHz Pentium and then automatically slow its clock down and reduce its voltage appropriate to the prevailing load, with no additional power management software? This is a very neat trick for keeping down power requirements. And the additive constant (power under zero load) is neglible. Vaughan Pratt Stanford Wearable Computing Laboratory
From Comp.sys.wearables Newsgroup Archive (CSW)
Maintained by R. Paul McCarty
Archive created with babymail