Wow, surprised to see this done with an ITX board, even one as good as that. I suppose VDDR delivery isn't as crippled with that form factor as the CPU but it still looks weird, you would think anybody shooting for a record overclock would be using a full-size PCB with lots of extra phases and MOSFETs to share the voltage required.

Then again, I don't tend to follow world records so as far as I know this could be pretty common.

BlueberriesWow, surprised to see this done with an ITX board, even one as good as that. I suppose VDDR delivery isn't as crippled with that form factor as the CPU but it still looks weird, you would think anybody shooting for a record overclock would be using a full-size PCB with lots of extra phases and MOSFETs to share the voltage required.

Then again, I don't tend to follow world records so as far as I know this could be pretty common.

I think the reason you see them using ITX boards is that the trace lengths are minimized for that last few MHz

I can see how the ITX form factor could be really beneficial... or more just the two slot configuration. 4 almost always compromises performance on at least two of the slots, just because of added complexity in trace layouts... a little easier to make 2 sets the same length versus 4. Doesnt always shake down that way in reality because of what winds up being most reasonable to produce... two slots winds up being a compromise for other reasons. But a premium two slot board should almost always have better RAM performance, assuming you stay in the sweet spot for two-stick configurations. The traces matter that much. The same two sticks are theoretically gonna be faster on the two-slot-only layout than in the two fastest slots of 4, assuming they use a trace layout that leads to one set being better, as they often do. Of course there are many different ways to terminate either and it comes down to engineering choices with any given board, though. Often it doesn't work that way simply because its not practical for other reasons. The cost of compact boards is already higher and requires more compromises, so even a top-end ATX board with an inferior (in terms of raw speed) 4 slot layout can still work better.

In this case I think MSI specifically designed this board to do insane RAM speeds with a two slot ITX as a calculated decision. It had to be conceived specifically for record breaking in that specific area, no holds barred. I wonder what compromises came along with that though. Im sure there are good reasons why its not usually done the way it had to be done with this board to get those speeds. But that's getting beyond the limits of what I know about motherboard design. I just know that any other ITX board coudnt have done this without losing out in eiter cost, vesatility, or both. For most people, a board built to do these speeds probably isnt worth it... and thats why it isnt done... its not like it couldn't have been done before, you know? Its a matter of balancing on the engineers' parts, not necessarily engineering innovation.

5000+ is as cool as it is stupid to me. Theres a part of me that would have a blast trying for that. And then there's the pragmatist in me that asks "Why? Just... why."

robot zombie5000+ is as cool as it is stupid to me. Theres a part of me that would have a blast trying for that. And then there's the pragmatist in me that asks "Why? Just... why."

I'm square in the latter camp. This benefits maybe a handful of actual workloads, and certainly none that are likely to be run on that platform. Also, for perspective, outside of pure bandwidth this will likely be noticeably slower than a DDR4-3200 C14 kit. Going by Anandtech's memory "Performance Index" (calculated by dividing data rate by CAS latency), which is a pretty decent predictor of real-life performance across workloads, this is way slower, scoring 180,9 (5608/31) vs. 228,6 (3200/14). It's barely faster than DDR3-1600 C9, and slower than DDR3-2133 C11 or DDR4-2666 C14.

In other words, this is like all most world records: (moderately) impressive on its own grounds, but useless in the real world. Unless your Z390 ITX board is running an extremely bandwidth-hungry server workload, that is.