Hacker News Reader

Up to a certain point and It’s actually very dependent on the CPU.

Take Epyc processors. On certain ones, after certain RAM amount, populating all the slots causes the cpu to kick the RAM speed to a lower tier.

You’re then limited to capacities of two sticks.

Weird, but it has to do with power requirements. Abutting above the threshold had to be buffered, which increases latency.

View on HN

It's the signal integrity and applies to desktop class boards, as well. A fully populated four DIMM Intel or AMD board will need to run at a slower mega transfer rate than a two DIMM board (or a four DIMM board with only two DIMMs populated)

I have a fully populated server with 2x7K62 and 16x64GB (3200 mhz) for my home lab. Do you know how to check if I am affected by this?

Easiest way would be to put two dimms in one channel, check your mem spped, put 1 dimm in 2 channels, check mem speed, then repopulate as you had, check mem speed.

Sometimes having more, but slower ram, is still better than having less, but faster ram.

As always, measuring your real world use cases is the ultimate test.

Reply on news.ycombinator.com