If you’re doing everything through the CPU, as is the case with EPYC, you won’t have room to differentiate between motherboards

This is of course nonsense, what you see as a “solution” that makes motherboards more complex and more expensive.

Let’s say you’re serving everything via the CPU and neglecting the chipset (which can save a lot of cost, and is very important for budget boards).

Now let’s use your example of a dummy consumer CPU with 48 paths and a second example where the CPU has 20 paths (like now) and then 16 paths for chipset connections.

If you have a CPU with 48 threads and no chipset, you can make motherboards as cheap (or as expensive as you want).

After all, you can issue a very cheap board with only one x16 slot and one x4 M.2 slot and you can’t simply use the other lanes, while on the other hand you can also issue an ultra-luxury board with multiple PCIE slots, think x16/16x/x4 or x16/ x8 / x8 / x4 (or a combination of these) with also a 3x and x4 M.2 slot. and all possible variants in between. Specifically when you are not tied to a chipset, you as a motherboard manufacturer have great flexibility and can differentiate very well in terms of motherboard hardware.

On top of that, differentiation doesn’t expire in other areas either, in addition to the above, you can still compete on the Phys used for (several) nics, whether or not it’s Thunderbolt, the number of USB connections, VRMs used, and the presence of RGB (unfortunately) And everything else you currently see on motherboards, too.

Whereas if you use the example of 20 lanes for the CPU and 16 lanes for the chipset connections you will get an expensive PCB, because you will lose an extra 16 lanes for the copper lanes, you need the chipset (one of the most expensive components on the motherboard) and you are less flexible than in the example above. More lanes for chipset connections will only increase the cost of cheaper motherboards.

[Reactie gewijzigd door Dennism op 22 augustus 2021 17:22]