Here’s another list of big computing projects that use distributed computing across the Internet. Idle time on people’s computers and all that.
Discounting Bitcoin mining (which isn’t really a single-goal driven project), the biggest one in the teraflops column seems to be … ergh. The table is a mess, it appears to be a mix of “.” and “,” decimal notation. Some of the largest in terms of CPUs seem to have abnormally small teraflop numbers.
Weather, SETI, folding, Math challenges, etc.
One of my faves types that doesn’t appear to be on the list are hash table computations. By enumerating all the basic hashes in a crypto function, you can make it almost trivial to break some older systems.
I wonder if there are any problems of current importance / interest where the computing power is clearly the rate limiting step. Like the theory and code for the system is in good shape, understanding is there, but we just lack the computational horse power to get it done.
If computing power was enhanced by orders of magnitude tomorrow, we’re still not solving the protein folding problem (ISTM). Not at any deep level. The understanding is not yet in place. Does the converse situation exist?
In many cases we do have a good understanding of the underlying problem, but the difficulty is in how to make it tractable at all - finding techniques for simplifying that remain at all valid. For things like protein folding there is a perfectly good theory - it is called quantum electro-dynamics. It will provide as close to a perfect solution as you might like. Well understood. And computationally infeasible. But it is used for simpler chemistry, where the number of atoms is small. Such ab-initio techniques are sufficiently accurate that an entire new field of chemistry has evolved. It is possible to perform experiments that could not be performed in the lab ever. This can include chemistry in extreme circumstances. The problem is that the computations require the system to calculate the interactions of at least every outer shell electron of every atom with every other outer shell electron, and you must do this for all possible energy states. (Since these states are infinite one of the limiting parameters is the number of states you will calculate for, and turning this down reduces accuracy.) As an approximation, the calculations are On[sup]5[/sup]. That is a very brutal limitation.
The list of things that Disheaval posted are all approximations. Indeed, it possible some of these approximations might be validated by comparing them to the results of the QED. But if you take every one of the issues listed, and expand it out to the full set of possibilities allowed by the physics, you will end up with a list of a few tens of thousands (more with the surrounding water) of electrons all interacting with one another via the rules of QED. In that sense we have a perfect understanding of the problem, and could write a program to exactly simulate a protein folding right now. However with our current computational capability the sun would go cold before we had any useful results. Hence the need to simplify the problem, something which thus far still leaves us way short of computational feasibility.
