Visit Your Local PBS Station PBS Home PBS Home Programs A-Z TV Schedules Watch Video Donate Shop PBS Search PBS
I, Cringely - The Survival of the Nerdiest with Robert X. Cringely
Search I,Cringely:

The Pulpit
The Pulpit

<<   |  If at first you don't succeed, drive, drive again  |   [ We have met the enemy and it is OOP ] >>

Weekly Column

If at first you don't succeed, drive, drive again: Why Intel is reinventing MMX

Status: [CLOSED]
By Robert X. Cringely
bob@cringely.com

If you stayed awake through much of my Triumph of the Nerds miniseries, maybe you noticed my car, a 1966 Ford Thunderbird. It's very similar to the car used in the movie Thelma & Louise, though in this case I play the role of Thelma. Actually, my car is in most ways better than Thelma's, since mine is newer (a '66 versus Thelma's '65), has the rare 428 cubic inch engine (Thel's was only a 390), and has never been involved in a multistate police womanhunt. My red convertible is among the last convertible T-birds ever built and has every available option except power vent windows and the 8-track tape player. Why, then, do I hate this car?

I hate the car because it looks good, sounds good, but it is totally unreliable. Partly this is the peril inherent in owning any car more than 30 years old, but mostly I think my car is just a dog. And while it has the occasional spectacular mechanical problems, like when the two rear tires decided one day to spontaneously (and simultaneously!) explode, or when the carburetor float sticks (making it impossible to climb more than a 10 percent grade), most of the problems are electrical.

With the exception of those little vent windows, this is an all-electric car. The most impressive electrical subsystem of all is the convertible top, which uses eight electromechanical relays to control its raising the lowering. You see, the 1961-66 Thunderbirds store the convertible top in the trunk, rather than behind the seat as in most ragtops. This innovation not only makes it pretty much impossible to carry more than an overnight bag in what ought to be an enormous trunk, it also holds out the possibility of the entire raising/lowering adventure grinding to a halt with the top halfway down and the trunk lid (it hinges at the back) standing vertically open, presenting to the wind a six foot aerodynamic spoiler. It happens all the time.

The small trunk problem isn't really much of a problem at all, given that: a) these cars were originally bought primarily by married women of questionable virtue for the singular purpose of driving to the Bide-a-Wee Motel for nooners, and; 2) I don't trust the car enough to take it out of town, ever.

The electrical problems themselves are attributable somewhat to a system design that taxed the technology of the day (today we'd replace all those relays with a $5 computer module and the multistep top launching process would work without a hitch), but the main problem is just bad solder. The electrical connections tend to corrode and become intermittent. Funny, that could also describe my sex life.

Which brings us, strangely but surely, to Intel. Don't ask me to explain this segue, just run with it.

Here's a rundown on the next several microprocessors we can expect from Intel following the Pentium II, which shipped last month. The code names are all rivers.

"Deschutes" will ship in late 1997. It will be a low-voltage version of the Pentium II aimed at notebook computers. The 0.25-micron CMOS chip will run at up to 333 MHz.

"Katmai" is scheduled for mid- 1998 and includes an expanded MMX instruction set, support for 3-D geometry and dual floating-point operations.

"Willamette" will appear in late 1998 and is supposed to be the last processor of the X86 architecture. Primarily a faster version of the Pentium II, it will run faster than 300 MHz and support up to a 1000 MHz bus clock.

"Merced" comes in 1999 and will be the first processor in the new family developed together by Intel and Hewlett Packard. This is a 64-bit chip and is intended to extend Intel's domination well into the next decade.

What we care about this week is Katmai, the chip after next. And the reason why we care about it is because of its expanded MMX instruction set, generally called MMX-2.

Why MMX-2? With MMX-1 barely out the door, why does Intel have to do another version? The official story is that MMX-2 is aimed at improving 3-D graphics, while MMX-1 is just for 2-D. The unofficial story is that MMX-1 sucks and MMX-2 is intended to correct the damage.

It's true that nobody in the PC world seems to be turning cartwheels over MMX-1. The promised 800-1000 percent improvement in 2-D rendering doesn't seem to have actually taken place. Intel attributes this to a lack of MMX-optimized software, but even the few applications that have been optimized for MMX don't show much improvement. And what improvement there is can be traced mainly to the chips larger onboard cache memory.

I first reported this weeks ago and have spent a lot of time since trying to understand what's really going on. Several discussions with top software developers later, I think I understand.

First, pay no attention to those colorful MMXniks dancing in Intel TV ads. They represent nothing except that Intel needs us to buy MMX chips whether they are any better or not. Not. The question that needs answering, though, is where did MMX go wrong? My developer friends say the MMX-1 instruction set and architecture were derived primarily without developer input. Intel decided it was going to optimize a chip for games and video, but then never asked the game or video programmers what they wanted. This is like making my Mom music director for a gangsta rap concert series. Intel didn't know what it was doing and was too arrogant to even know there was a problem until it was too late.

The result is we watch these catchy TV ads then try to convince ourselves that our new PCs are really doing faster graphics when they aren't.

So what about next year's Katmai processor and MMX-2? Intel may be arrogant, but it sure isn't stupid. Katmai fixes the problems with MMX-1 and then some. My developer friends say this time around Intel did extensive market research and adjusted the MMX instructions and architecture -- including the 2-D technology that supposedly didn't need adjusting -- with the result that MMX-2 truly will bring a 1000 percent increase in rendering speed. Let the little TV men dance!

This is bad news for vendors of graphics cards, so-called Windows accelerators, and the emerging category of so-called media processors, like Chromatic Research's Mpact. Katmai motherboards will make most of that additional hardware unnecessary.

I take a similar approach to my T-bird, having declared the convertible top to be unnecessary technology, too. Now that it's June and the rain is over, I'll just leave the top down until October.

Comments from the Tribe

Status: [CLOSED] read all comments (0)