Why do motherboards only support oh so much RAM?

Question

How come some common PC motherboards can only support up to 16 gigabytes of RAM, and then there are some that are more expensive motherboards that can support far more quantities of RAM? How come you can't just use as much compatible RAM as you have, as long as there are enough channels?

Answer 1

tldr - It is a result of physical constraints, and memory controller constraints. The memory controller can only handle so much. There's a limit to how much ram you can pack into a single stick, and how many sticks per motherboard.

The more complex answer involves a few elements.

1. Does the memory controller recognise sticks of more than a certain size. This might be product segmentation in some cases or simply the lack of sticks of the size at the time. I have a system that won't recognise a 2gb stick, but recognises a 2x 1gb sticks fine, despite the ram being the correct type. My older and still common DDR3 memory tops out at 8gb per stick. The DDR4 used with current enthusiast boards, and many future skylake boards may top out at 16gb

2. Does the memory controller have enough channels to handle more ram. This would be the difference between a 'mainstream' board with 4 slots, and a enthusiast board with 6-8.

3. Does the board have enough physical slots - the same processor in a full ATX board would happily swallow up 4 sticks of ram (so 32gb or even 64gb) while a mATX or mITX board would take 2. This would mean the half the ram, even if you had the same processor with the same maximum amount of ram.

4. Support for "rdimm" and "lrdimm" - mostly on servers but this has a tradeoff - you tradeoff overall ram speed for significantly more ram - up to something like 2x or more per stick. Anandtech has an awesome article on this.

Its really a matter of design, and in some cases validation.

Answer 2

(note: parts of this post will be Intel-centric, I don't follow AMD platforms)

PC memory is divided into "ranks". A rank is a group of 8 chips (9 for ECC) that cover the complete width of the memory bus.

Normal desktop memory is "unbuffered" and this limits the ammount of memory chips that can be connected to a memory channel. You can have up to two ranks on a module and up to two modules on a channel. "Buffered ECC" server memory can go to higher rank counts but you pay for it in both performance and pricetag, and it's only supported on the higher-end server platforms.

Then there is the size of the chips themselves. In general newer platforms support bigger chips. The first DDR3 platforms only supported 2 gigabit chips (leading to 4 gigabyte dual-rank modules). Later DDR3 platforms* support 4 gigabit chips (leading to 8 gigabyte dual-rank modules). DDR4 platforms generally support 8 gigabit chips (leading to 16 gigabyte dual-rank modules).

Then there are space constraints. Laptops will usually only have one slot per channel, the same is true of some small form factor or bargin basement desktop boards.

Then there is the fact that vendors don't always tell the whole truth. There are many systems where the advertised maximum ram is based on 4 gigabyte modules but where 8 gigabyte modules do in-fact work.

Then there is the fact that the high-end desktop platforms (LGA1366, LGA2011, LGA2011-3) have more memory channels than the mainstream desktop platforms. The high-end desktop stuff also moved to DDR4 sooner.

Put all of this together and you end up with a wide range of advertised maximum memory.

* Sandy bridge and later officially, unofficially also X58 LGA1366