Actually, since Intel Pentium 686 (I think), there is an upper limit of 64GB of RAM. Likewise, most OSes can address that as well, however it depends on various conditions and has its own limits.
Also, as you say memory controller is integrated with CPU, so modern topology of motherboards usually has memory connected directly to CPU, bypassing chipset. Also, it's why most topologies connects at least some of PCI(e)'s to CPU directly as well (giving GPU access to system RAM). This is also, accidentally, why good VGA requires good CPU as well for it to perform satisfactorily.
Anyway, what's really limits the MAX_RAM on the system is motherboard. 4GB requires 32 physical paths for addressing (excluding data, it's just for addressing) from CPU to memory slots. 8GB requires 33, 16GB requires 34 and so on. Intel's CPUs currently have 36 addressing pins (at least).
But laying pathways is expensive. Cutting down on this work lowers costs and in result prices.
So basically only limit is in motherboard...
However, sometimes documentation lies (unintentionally). It may be a good idea to try (if you have access to free ram for testing). OTOH, this was more common in days past. I don't know how that works in new/newest products.
EDIT: of course, I talk about POSSIBLE MAX. If manufacturer will not implement them in the chip, they won't be available. That's why there's max amount of memory supported by CPU usually way lower than 64GB. Anyway, what matters is CPU limit (pins and IMC to be exact) and motherboard (physical paths CPU-RAM sockets).
1"This is also, accidentally, why good VGA requires good CPU as well for it to perform satisfactorily.". Can you explain that a bit more? BEcause it feels wrong to me. – Hennes – 2015-02-12T10:09:36.167
What I mean by that is even if you invest in reeeeeally fast VGA it will not give you same result when CPU will be 1st gen i3 as compared to latest gen i7. And by that I mean really GPU-intensive work, not overall system performance. It is, of course, mitigated by giving GPU it's own RAM, but as is well known, no amount of RAM is ever enough. So if it's needed, the GPU often is given access to system RAM. And that is where we're back to CPU. The better CPU the faster it can give access to RAM, as it is not loaded with system management (CPU is called "CENTRAL" not without reason). – AcePL – 2015-02-12T10:19:53.633
True. But dropping speed to the dedicated graphics cards does not reduce it all that much. E.g. PCI-e v2 vs PCIe-v3 (same number of lanes) dropped performance by about 4%. Same for going from 16 to 8 lanes. 4% is not nothing, but it is not enough to be noticeable. – Hennes – 2015-02-12T10:30:36.903