The End of Serial Computing?

How does this differ significantly from Open MPI or PT-Scotch or any of a number of other implementations of multicore threading? I've been doing this for a couple of years now with my laptop with a dual core...Oh. The difference is I don't use Windows- MS seems to still be working on 64 bit basics...

I have 9 wives. Does that mean they can deliver a baby in one month? Random instruction executing yields random answers. The larger number of concurrent executions, albeit randomly, will require at least equal tracking of which results match the original question. My experience is that very few real processor executions are "plain adds" that would fit the criteria of selection for instructions to be executed randomly concurrently which means all the rest of the executions are based on some variable variation in data, time, or external influence.Antique processors use last in first out stacking or first in first out stacking to handle the execution branching that helps determine what to execute next. If you do not follow a serial path, how do you know when the journey has ended?

Where I use parallel computing is in such things as Finite element analysis, where a problem can be broken down into relatively independent blocks which are solved simultaneously. There is always a "master" responsible for making sure the different parts get put back together correctly. The process is not random, but it is concurrent. Each processor still follows a linear path...

I do understand when advantages come from having multiples of processors for assigned subsets of the master. No challenges to that concept but you need to do some research to the major vendors that already have offered systems with hundreds of cores. The reality of those systems is the design of applications that can be partitioned. If it is a one off special task that may only be used a few times comparatively speaking to a commercial profitable usage i think you will only find a couple of dozen organizations potentially capable of using such a system and afford to buy it.

Let me take you down a different path a minute. If the processing world needs to advance to a new level,the sizes of databases seems to be the fastest growing problem that is common. Certainly faster, larger computation is attractive, but look at how many cores are in use that only have less than 15% core usage 97 % of the time. During that 3% busy time getting the information thru the system is the bottleneck. So if we could handle data while stored and accessed quicker, the cores could operate serially and in my words, normally controlled.For the last 43 years that I have been in this business I have thought why not move to a ternary logical system. Not just ones and zeros but ones, zeros, and minus ones (mones). If you understand binary you will quickly see the exponential gains from the three unit numbering system. The simple diode and transistors and memories can be manufactured to recognize +,0,- voltage levels.

Actually, my two processors on this computer (and the other three processors I have laying about) are only used to about 1% of their capacity, if that much- until I run one of my analyses.

Data has no intrinsic value. A value can be assigned to data ONLY when it has crossed the human interface and has been used to make a decision that either enhances (positive value) or degrades (negative value) ones welfare, wealth, health or happiness. The more data that one accumulates, the more costly in time and resources it becomes to recover at a critical moment. I am a unrepentent data collector myself- but I generally wind up with so much data collected, that it becomes more costly to mine it for critical information than to regenerate the basic information from scratch. Just moving it about from one point to another accomplishes nothing, unless it just, by chance, happens to land in front of the decision maker at an opportune moment.

I definitely appreciate the value of your tertiary, as opposed to binary, scheme- only I would label the third digit "maybe", and assign it a value of 0.5. The only problem with my scheme is that one could not tell whether two maybes added up to a true or a false...I can understand how to get your three values from a transistor, but I do not know how you can get this out of a diode...

This is not about random instruction execution. It is about multiple coordinated paths that can process concurrently.

Think of it in terms of a Pert chart. There is a start and end, but in the middle, independent paths toward the end occur. Within a path, things need to go sequentially. But the other independent paths, run in parallel.

At some point they will have some coordination and rejoin.

The longest path in the process is known as the "Critical Path" because it determines the minimum time possible from beginning to end of the process.

Multiple wives and multiple babies clearly has a critical path time of about 9 months.

So while your single wife would take 81 months to have 9 babies, 9 wives could have same number of babies (if not the same babies) in 9 months.

(And I agree with another poster: How does this differ from multicore multithreading, or parallel processing (SIDD - same instruction, different data)?)

Rufus

".. 9 wives...deliver a baby in one month". No, but the average rate of baby production is 1 baby/month. Ffej

A radial redesign? Why not, it worked for tires!

It would need some variation of the Critical Path Method to plan the weaving of the threads through the cores.