The WIZ chip
--------------

The WIZ chip is very simply a chip full of WIZes.

It is not more complex than that. They are not coordinated in any specific way. A WIZ chip has no circuitry other than WIZes, and a mesh-like bus structure to interconnect them all as part of the planetary WIZ mesh-network.

Every WIZ on the planet has a unique serial number. The instruction "A -> B", where A and B are WIZ serial numbers (rather than slot numbers) copies data from one WIZ to another, whether in the same WIZ chip or another chip half-way across the planet.

Suppose then that some WIZes run "subroutines" and by copying data to them we are supplying "arguments" and then later by reading data from them we are getting "answers". For example, to compute Q = sine(P), where WIZ serial number 2839300349 is running a sine algorithm:

  P -> 2839300349   // copy data from our slot "P" to a WIZ running a sine algorithm
  2839300349 -> Q   // copy data (answer from sine) to our slot "Q", ie, Q = sine(P)

Likewise, we can have many WIZes running many other arithmetic functions. And as these serial numbers are unique across the planet, we can access subroutines in any WIZ anywhere on the planet.

Note that every WIZ runs independently. And they can all run simultaneously. Thus everything said earlier about parallelism, ILP, and OoO, within a single WIZ, applies equally well to the entire network of WIZes.

For example, if we need to compute Q = sine(P), S = squareRoot(R), and U = log(V), and we have three WIZes running these three algorithms, we could use a "natural" order like this:

  P -> sine       ; sine -> Q          // compute Q = sine(P)
  R -> squareRoot ; squareRoot -> S    // compute S = squareRoot(R)
  V -> log        ; log -> U           // compute U = log(V)

This will have three long waits on each of the computations. Instead we could do it "out-of-order", like this:

  P -> sine ; R -> squareRoot ; V -> log   // START all three WIZes
  sine -> Q ; squareRoot -> S ; log -> U   // now pick up their answers.

This version produces as much as a 3x faster result, as all three functions are running simultaneously in three other WIZes.

If we don't need to use the results of any of the functions right away, the following is even better:

  P -> sine ; R -> squareRoot ; V -> log
  etc -> etc  // more instructions here, until any one of the three functions is needed
  sine -> P ; squareRoot -> S ; log -> U

Thus we have one coherent paradigm, "A -> B", at every scale of expansion. When we get entire planets running a single massively parallel algorithm, with a serial number for each planet, we will see "mars -> saturn" as just another ZOZ instruction. ;-)