Highest possible rating

Lets say that advancements in computer processing speed in the future allowed chess to be strongly solved such that a 32 piece tablebase existed such that the optimal move and theoretical result of all positions were known.

Lets assume for the sake of argument that the theoretical result of the starting position is a draw and that black has at least some room for error.  in other words, assume that black does not have to play precisely for every move during the entire game to hold a draw.

Lets also assume that the computer could somehow press for a win in the most effective possible fashion in a theoretically drawn position (i.e. by giving its opponent the highest probability of making an error).

What would the FIDE rating in classical chess of such a computer?

Here is the thing...I don't think that the computer would win every single time against houdini or even the world's top grandmasters.  Occassionally, Carlsen might play the computer to a dull draw with the white pieces...same thing with houdini.  Maybe this tablebase would score 80% against houdini which would only allow a rating of about 3500 (assuming houdini is 3300).

So is 3500 the highest possible rating even with perfect play?

please give your opinion and reasoning

There is no strict ceiling. It depends on what the rate of win/draw were. 

Example, if Carlsen were able to draw 5% of the time, then assuming that Carlsen's rating were fixed, the "god computer" rating would approach the rating that would give 95/5 win/draw ratio the more games were played between the two, which would be about 3400.

Of course, it could be much higher, depending on what that probability were. So if it were 99/1, then it would be more like 3700. 

Interesting question, but the answer seems sort of elementary.

I understand Daniel.  So what score do you think Carlsen could achieve against a "god computer" as you call it?  1% only?

Impossible to say. It depends on too many factors. What the structure of the "solution" to chess looks like. Also, there could be multiple "god computers," given that there are a vast number of "optimal" moves, if it steered the game in one direction or another, there may be increased or decreased probability that Carlsen plays a losing move. I'd guess that a "psychologically sophisticated" computer with a 32 piece tablebase would probably win more than 99% of the time -- but admittedly, this is a totally unfounded estimate (my gut intuition).