Reviving the PC Industry - Embedded.com

Reviving the PC Industry

OK — I admit it: I'm an embedded head. The PC industry just doesn't get me excited anymore. It's AMD versus Intel, Apple versus Microsoft, Norton against McAfee — six of one, half a dozen of the other. When innovation gave us quantum leaps in horsepower the PC biz was exciting. Seagate's early Winchester disks upped the ante from a floppy to 5 Mb of fast storage. The 386 busted the 640K-memory barrier in a palatable way. Windows 3.1 — love it or not — finally pretty much worked and spelled the beginning of the end of DOS.

But the PC market now looks like the steel industry. It's a big buck, low profit industrial manufacturing enterprise that cranks out pretty much the same old products. Sure, every year processors get a bit faster and memory sizes increase. But today's newest desktop PCs run about as fast, for a real mix of applications, as those available a few years ago.

Intel recently revealed a technology that could put 4 Itanium 2 cores onto a single die along with 16 Mb of shared cache. A billion transistors, 120 million of which go to the CPU cores. No doubt it'll run a 100 gigamegaweeniehertz. It's designed for server applications, of course, not the desktop.

Users will need a refrigeration plant to cool the thing. A single-CPU Itanium 2 dissipates 130 watts and will suck 100 Amps from the system supply, with its current demands slewing at 100 Amps per microsecond. That's almost enough to jump-start a Toyota.

Check out the Itanium 2's data sheet. This is not your father's computer anymore. It has taken Von Neumann to an extreme all out of proportion to the possible benefits. Determinism is out. Branch prediction, speculative and predicated load/stores, and a wealth of other features make reading the data sheet akin to skimming the Daily Racing Form. One chapter is titled “Load Temporal Locality Completers.” Huh?

We'd expect the chip to work, of course. But that's apparently nave. The part is so big, so dense, with such tiny transistors, that it's ground zero for cosmic rays. Intel provides an on-chip system management unit that monitors the part's health, flagging errors when it does something unexpected. Plenty of software support for this is needed, though just what the code is supposed to do when the CPU signals a failure is not at all clear.

The technology is awesome. Intel deserves kudos for being able to build something so complex. But it's like a battery-powered slide rule. Fast, perhaps, cool, for sure, but with few customers. How big is the server market, anyway? Large enough to justify this level of R&D? The part is too complex and expensive for consumer use, and the Itanium's suggest a relatively modest performance boost over a 1 GHz Pentium III.

It seems to me that such a vast increase in complexity over the Pentium, without a corresponding massive performance boost, means Intel and others are spending too much on diminishing returns. Faster CPUs just don't cut it anymore.

I contend Intel is missing the boat. The PC market will stay flat till it gets dramatically faster processing. Not faster processors, mind you, but a complete system that runs like the wind, in a real-world scenario of a half-dozen open apps that saturate any cache trickery. Processors have long been memory bound. We have the ridiculous situation of a CPU issuing 50+ wait states every time it registers a cache-miss.

If the vendors want another spurt in PC sales, akin to the “next killer app,” they should refocus on memory. Don't spend 1 billion transistors on lots of processor horsepower; build a couple of high-speed billion-transistor SRAM chips that couple to the processor with no more than a few wait states. Perhaps this means a processor module, rather than chip, but such modules are not new.

Users will flock to Dell and CompUSA when all of main memory runs at near CPU speeds. Till then, the collective yawn will keep computer makers' stocks depressed.

Jack G. Ganssle is a lecturer and consultant on embedded development issues. He conducts seminars on embedded systems and helps companies with their embedded challenges. He founded two companies specializing in embedded systems. Contact him at . His website is .

Reader Feedback

Amen, Jack.

Witness the media's fascination with Dell spokesmodels, and not Dell's products. 'Nuff said.

If I buy another computer, it'll be one that's powered by ARM (not StrongARM), MIPS (even one from AMD), or SH. That way, I'll get orders of magnitude lower power consumption, and an instruction set and processor model designed by sentient beings.

And did I mention, I'll get more than adequate performance and lower price, too? And it'll run OSen from a handful of vendors, instead of being designed for just one. Bliss…

Those in the know are leaving WinTel and this x86-compatible-clock-speed pissing match behind. I think it won't be long before the rest of the world wakes up and follows.

Bill Gatliff
Independent Consultant


Yes, cpu speed isn't everything. May I also suggest reliable and efficient software – that doesn't leak store like a sieve, and doesn't declare a few megabytes worth of inherited objects it won't use just because it's a cheap hack of someone else's code that uses the same kind of window.

Phil Rowland
Software Engineer
Metrodata


What planet are you writing this from? My recently purchased 3ghz machine (arriving today I hope) is twice as fast as the best from 2 years ago. In 3D rendering, its 4-5X faster.

If you think Intel is “missing the boat” on the need for faster memory, consider the fact that the Itanium 2 core is over 65% memory itself– cache. Intel doesn't make drams; it can't control what the fastest part available is.

And bemoaning the I2's 'minor' performance gains over a P3 misses the point. Itanium is about 64-bit computing– crunching huge numbers, and a 65-bit memory space exploding the 4gig ram limit. If your app fits on a 32-bit processor, well then it belongs there. Apps that require 64-bit integers or tens of gigs of ram run dozens of times faster on an Itanium 2.

In parting, the server industry is quite large enough to support this chip– if it gets bought, that is. Its not designed as a consumer part, and Intel is certainly not planning on Ma and Pa “flocking to CompUSA” to buy it.

Mike Asher


You've chosen the wrong comparison – PCs are like VCRs, not cars. The auto industry is so retarded that the price is way excessive so even inefficient manufacturers can survive. For VCRs, only the most efficient manufacturers survive. It will be interesting to see if Dell survives Wal-Mart.

What it will take to revive the computer industry is a new product category like DVD players, but that took real research and investment on the part of a number of companies.

There is a market for high end computers just like there is a market for “video tape” (digital data storage on tape), but the market for tape is shrinking rapidly in the consumer market. Still VHS will be around for a few decades, just like PCs will be around for a while. The margins for both will be very similar.

Faster memory on a PC isn't going to make me an 99.99999% of the market go out and buy a new PC just like Sony offering a new VHS+ format (that would let you record 9 hours of video) wouldn't cause you to rush out and buy a new Sony VCR.

Just looked at the survey on what it would take to make me buy a new PC – none of offered answers apply. What has tempted me to buy a new PC is Wal-Mart's $199 PC, but I still think the price is too high, so I'll wait until it hits $138.44

mulp
55 year old student, 35 year computer veteren
xe-DEC, axed from Compaq


As usual, you are right. Back in 1998, I picked up a P2, 233. I chose a slow CPU, a hot mother board and video card, and a lot or RAM. I figured that the 233 would out run all the periphials, so, I tried to make everything around the CPU the best, and go with a relatively slow CPU. Still run it today, almost 5 years old now.

Steve

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.