Apple recently astounded developers
(some of whom had made plans, including hotel reservations) by moving WWDC from San Jose to San Francisco and postponing the affair by a month. This after the date had supposedly been set a year in advance. Ostensibly this was done in order to place into developers’ hands copies of the new version of OS X, code named “Panther”. However, there is apparently much more going on at One Infinite Loop than meets the eye. What follows was obtained by means neither Nellie nor the Spy are prepared to disclose of course, but readers can be assured that the source is reliable, and that we stand by our information.
The Northern Spy has learned
that Apple Computer Corporation is preparing a series of new machines based on novel chip technology developed by a little-known Idaho company called Monk, Inc.
The new architecture
is based on a novel form of memory organization in which each bit has ten states latching from 0 to 1V in .1V increments plus or minus .02V (to ensure recognizably discrete steps). That is, bits code decimal directly, rather than via binary. These bits are organized in five bit nibbles and in turn into two nibble bytes (plus a flag bit that can be used for such purposes as parity, sign, etc.) Note therefore that a byte stores values from zero through one less than ten billion, making a single byte a useful data store for signed and unsigned whole types.
Apparently in early testing, some wanted to term the new bit and five-bit units the “dit” and “munch”, respectively, but these were thought much too coy, and more conventional terminology has prevailed for the nonce, though it remains to be seen whether the new units will in fact be renamed in practice.
Early sampling has produced stable chips with a 60% yield testing to 10GHz and another 25% that work reliably at 5GHz, but these speed figures may treble by the time Apple has production machines rolling off the assembly lines.
As in more conventional machines, two bytes constitute a word. A new IEEE/ISO-30401 real format employing a word is in an advanced stage of preparation, but Apple is expected to go with the preliminary specifications, as they have in several other recent instances. In the new format, the low nibble (five figures) of one byte (and its flag bit for the sign) are used for the exponent, and the other three nibbles (fifteen figures) and the second flag bit are employed for significant decimal digits and their sign. If the flag bit of the high byte is set to nine or ten, the representation is fixed point instead. The result: higher precision, better representation of real world data without worrying about round off errors, and no need for space-wasting BCD. (Other meanings of the ten states of the flag bit are beyond the scope of this article.)
Because a given bit stores values, the Monk CPU has a much lower part count for a given functionality, freeing up valuable real estate for other purposes. Consequently, initial production units embody ten complete processors that can be employed in parallel. These share the use of 1000 registers, any of which can be used for data, addresses, counters, or stacks. All addressing modes are valid on all registers. During processing, addressing modes are held in a special “zip” word (one for each processor) and employ a variation of POP, wherein three bits are used to hold the register number of each of up to five operand registers, three the actual mode (direct, indirect, double indirect, memory, pre- or post-incremented, stack, and combinations thereof), and the remaining two are general-purpose flags. Each processor also has a binary flag register to handle carry, overflow, negative, and zero conditions, supervisory bits, and a two-bit decimal portion to record up to a hundred interrupt levels.
The chip takes advantage of the latest in pipelining technologies, including the Trans-Mountain algorithm, which allows different types of data to pumped through the same pipeline, so long as they are separated by a suitable buffer of null data. Initial production runs have 10M of on-board cache, against portion of which any of the ten processors can write a draft (a.k.a. a cheque), and optionally, lock the draft obtained against the other processors to ensure data integrity.
Externally, there are 20 data lines per processor, thirty signalling lines (ten per processor and ten global), and 100 addressing lines. The latter means that the Monk chip can handle 10 to the one hundredth bytes of physical memory, though with the high price of initial RAM, production machines are expected to have only a few terabytes installed.
A new type of external storage medium
may eventually render the latter relevant, however, by making virtual memory as fast as the installed kind. The medium, called a data cube, can be inserted into the data reader in any orientation, and the device automatically determines the appropriate direction from which to read. The following quote is from my upcoming book:
“About a half cent on each side, and translucent, the data cube is a nanomachine-grown carbon matrix with a doping molecule at regular interstitial intervals, each of which can be switched to one of six orientations and one of two excited states–more than enough for a ten-value bit. As in a computer’s own ten-state-bit memory, these are combined in five-bit nibbles to store numbers up to a hundred thousand at each memory location, or employed in two-nibble bytes for five additional significant figures. Read and written by paired red and green lasers, the data cube offers rewritable archival storage for up to a hundred terabytes, depending on its quality.” (1)
Obviously, it is possible to do much better for total storage simply by making the medium somewhat larger. These are, after all, preliminary models. But because there are no moving parts once the cube is latched into position (the laser beams are guided electronically) access times are expected to be comparable to those of conventional RAM, and the latter may soon be obsoleted.
So much for the hardware. What value does Apple add to the mix? At the top level sits OS 10.4.1, which has few surprises except, according to my spies, for the reintroduction of several more orphaned OS 9 features. The new OS is code named “Calico” but actually has a tortoiseshell, though not many will select this over the conventional shell as it appears slower at first.
The big difference under the hood is that Apple now has an integrated mix of BSD and Linux, each with aliases to items stored in its opposite number’s appropriate directory. This means that Linux or UNIX programs and scripts that depend on the location of various items will all work correctly without porting. This new base is called Flex-UNIX or “Flux” for short. However, reliable sources advise that Apple will not introduce the new OS to the broad market, allowing it to run atop just any Linux. This makes sense, for though there would appear to be nothing stopping people from porting the Darwin kernal, the new chips (made in a factory co-owned by Apple and McCain (Monk’s parent), will so blow away anything Intel has on the boards for the next ten years that there would be little point.
On the tools side,
Cocoa is being extended with a new compiler that uses context switches and semantic analysis to elaborate any combination of Assembler, C, C++, Objective C, Java, Modula-2, or Kylix in a single source file. Excited tool developers refer to the result as the first ever DWIM compiler. Languages can be added to the mix with drop-in front end modules, but Apple is apparently planning to leave the field clear for developers to expand the concept by publishing all the specifications for the module interface.
The Actual Boxes
Apple has also apparently decided to move on from the grey look of the current boxes. A sample I was told is the leading candidate for the design of the retail box has a pastel yellow hue, and Apple is internally using the code designation “Golden Delicious.” However, several industrial designers are bidding for this part of the contract, so the final appearance remains to be seen.
Perhaps the most interesting prototype in the line is a miniaturized version, smaller than the old Newton, but larger than a pocket device. No batteries are required as the device generates the .001W needed to remain functional by a combination of ambient heat energy and motion. It is worn retro-geek style in a leather pouch attached to the side of one’s belt. This project is more closely guarded, but the Spy can say that the “Cider” as it is currently known, is a combination miniature Mac/PDA/book reader/cell phone, for the latter of which functions Apple is apparently considering establishing a proprietary network that would be free to customers for life.
Miscellaneous notes to close off
include that Apple has brought on several new partners to make this venture a reality. These include Osterizer Corp. (designer of the multiprocessing system) Greyhound (supplier of the bus) and the Lewis Carroll trust (which financed development of a new, larger, white, pointer device designated ML8 ML8.)
Projected release date: April 1, 2004.
— The Northern Spy