A response to BILOW :
A not so private communication:
Hi Jim, Regarding Bilow's response to my message, I still think he has it wrong. Dell's offering or not offering a sub-$1000 PC is in no way comparable to the architectural and technology transitions that hurt the early Super Computer manufacturer's such as Cray.
1. Dell will respond to its customers. When they ask for lower cost/lower performance PCs based on entry level processors, MD will provide them.
2. The low cost PC phenomenon is not being driven by higher levels of integration (something on the horizon from National Semi and others), but rather by availability of clones from AMD and National Semi, their willingness to ship dollars in order to penetrate a giant market, and the desire of Compaq to shore up its distribution channels against Packard-Bell-NEC and others even at a loss, and the blow-out prices they promoted necessary to clear the stuffed channels.
3. The prices of entry level PCs are already stabilizing closer to $1000+ with monitor as evidenced by Compaq's new price points and the drop-off of blow-out promotions in the channel. Check out the Compaq web site for details: only one unit under $1,000, and that's with a 13 inch monitor. Other consumer PCs running at 266MHz are well above the $1200 point, and with minimum expandability: max 64mg, max one bay, max 2 slots availability.
4.Dell has announced two systems based on the Celeron II at 266 and 300MHz, priced ~ $1000 and $1200 quite competitive with the Compaq machines.
5. The high level of integration Bilow touts is occurring only in processors using older technology, and aimed at the appliance and embedded applications, NOT at desk top corporate or consumer markets. There are several reasons for this: a. There is enormous development expense associated with developing the high level of integration Bilow is talking about. Until the technology is well established and a common set of functions defined which will provide all the functionality the market desires, without throwing in a lot of technology that the market doesn't want, or worse, and lot of technology the market has decided not to use either because of something better or just a lack of need, it is prohibitive to design, manufacture and market such highly integrated devices. b. Of course the story of the computer industry is one of increasing levels of integration at the silicon level. The CPU in the 60s was the size of a large refrigerator, not it is one chip.
There has yet to be achieved a level of standardization in any aspect of the technology such that all the computers functions could be frozen and laid out on a single slice of silicon. For example, Memory technologies continue to proliferate, and the value and performance to be derived continues to advance, and the price performance curve continues its steady progress. We have many memory technologies in use in current systems and new technologies on the horizon. What to do, what to do? Integrate them all? Select just one or two and make the customer choose? The same story applies to almost any other of the many PC technologies. Serial Bus, RS 232, Ethernet, cable modems or USRX. Until we have fixed all the elements we will continue to have a high degree of after market, customer and application driven selection of the technologies to be used. Only those applications which can be tightly specified in terms of performance requirements and functionality can be safely integrated entirely in silicon, assuming adequate volumes to support the design and manufacturing cost.
So what are Motorola, National Semiconductor, IBM and others talking about when the talk about the PC on a chip? Mainly embedded and hand-held applications that integrate a specific set of features for a particular application, for example the UPS handheld data shipment data entry computer, smart appliances, automotive controllers, hand held messaging systems, internet appliances, set-top boxes and possibly the thin client PC.
Even robust consumer applications require a degree of performance, expandability and configurability that prevent total integration. That is why many home users require very high performance and configurability from their systems, as well as state of the art technology. Try running a recent home education or game program on a 200 MHZ machine without 3d graphics or MMX: new math and language programs and some of the 3D games that arrived last Xmas come to mind.
Better yet try editing your home videos on any thing less than a high end machine, or making your own CD/photo album. Try surfing the web reviewing audio, graphic or saving voice to your disk. IMO we have not yet run out of reasons to upgrade, or applications in the corporation and at home that will use all the horsepower we can get. Rather we are at the beginning. I do agree however that there is a place for low cost entry level computers, just as there is a place for low cost personal computing and communications appliances with relatively fixed functions.
The thin client comes to mind as a replacement for the green display terminal for administrative applications, data entry, data access etc. No one I am aware of, not even IBM who is a key proponent, believes they will replace the PC or significantly cut into the volume.
6. It seems odd to say that the old supercomputer companies were destroyed by the higher level of integration. As the articles he refers to note, they went into a tail spin as a result of the collapse of the military market. There were eclipsed entirely by the arrival of new technologies based on standardized components used in mass markets. The volume/price curve of the standard technology destroyed the economics for the old supercomputer. 1000 8086s or 286s or now Pentiums, linked together with massive memory systems based on standardized memory components provide far greater performance at a lower cost than proprietary, low volume high speed components based on expensive, one off technologies such as SOS. of course the new low cost technologies require a new architecture: massively parallel hardware and new memory system architectures. They also require new software algorithms which obsoleted the approach that CRAY was following. The ultimate supercomputing systems which have emerged, are possibly less integrated than those they replaced. They are also much more scalable, and have much lower entry costs,based as they are on high volume low cost standard components.
In other words, I think Bilow's arguments are leaking a bit around the edges. Better head for shore. Doug |
