EDIT : some clarifications
-------------------------
My question makes me feel like a noob, but as I never seems to need ultra specialised cities at the level I play (prince), I may be a noob at specialised cities.
It would seems to me that even in a GP farm you would need some production :
In a GP farm you would want to produce buildings; normaly at least : granary + health and happy structure ... it helps that some of those (grocer-market, temple-cathedrale) also give you some specialist slots(merchant, priests).
You also need to produce buildings with the only aim of producing specialists slots:
scientist : library, univeristy; engineers : forge, and factory; artists : theatre, other (I do not remember). You need a lot of slots if you want to micromanage or have a word in the GP you want produced (thats why I included factory and others). You will need these building as wonder produced slots are limited by... wonders. So only one per map, so if you plan only on those, every such wonder-race lost is a specialist slot lost.
So here you have many (3+) slots for every kind of specialists, and with a miraculous by-product some of the buildings are usefull economically speaking : multipliers for science, for gold and for hammers.
Of course you can whipp all these buildings
, but it seems to me that it can be a little counterproductive to whipp so much, as it represents a enormous number of population sacrified (I do not want to do the math but even with the pop-rush bug and the good ressources, it still makes a lot of hammers ==> lot of pop). And so an enormous time lost for producing GPP. It thus ******s the early GP creation. And that does not count producing (whipping) the national epic that gives +100%GPP.
Furthermore: 1) all those specialists are going to produce some commerce($)/gold(G)/beakers(S)/hammer(P)/culture(C) as a by-product, and more with the good civics and wonders; 2) each yield is less than non GP farm cities of the same size but it is NOT NEGLIGIBLE if you have a lot of specialist as a GPfarm should have to produce lot of GPP; 3) by default your river and sea tiles will produce commerce, some tiles may produce hammers and the trades routes will again produce commerce. 4) if you can't have the Kremlin, rush-buying the late buildings will be very expensive.
So we are here in a position were you have a city with a lot of yields ($,G,S,P,C) that are not fully exploited, and I hate underexploiting my "ressource" ie here : yield.
IMO it means that it is still interesting (even maybe necessary) to build more buildings in a GP farm : power central (improve P), banks (use $ or G), observatory and monestries (S), media tower (C?) in my GP farm. so this again means producing hammers.
If you want to really have a GP farm and do not mind wonder produced specialists, you may want to build wonders such as Great library : +2 scientists... ie +6GPP(scientists)+2GPP(wonder)=+8 GPP (+3source for a scientific GP) x150% = +20 GPP/turn = not negligible if you are really aiming to speed up GP production.
All those facts means that IMO you would at least want some tiles in your GP farm that can be used as heavy-production tiles.
SO why do everybody say that all building in a GP farm can be whipped and that a full grass + food ressource is the best emplacement?
Using a heavy-production tile or two (more hammer for a long period ==> more buildings ==> better use of beakers/commerce ..)==> it can amount to gaining one or two tech in it's own right, or permits to build units in place of a commerce city... is it not worth loosing the n+1 GP if n is sufficiently high?
Furthermore, if you do not have to pop-rush so much, it means that more pop can be used as specialist, accelerating a bit the GP creation.
In this case it may even be more profitable for your whole civ to have a less performant GPfarm that gives you only one GP less : the number n+1 one : the one that needs 4000 GPP.
==> is it worth uberspecialising only to (maybe) gain 1 GP or 2 in the whole duration of the game 400-1000turns?
Does anybody agrees ?
or maybe having the (n-X) to n GPs coming each a bit sooner is still an enormous advantage in itself ?