A Skylake-based Intel Xeon Silver 4114 CPU can access up to 6 DDR4-2400 memory sticks at once. It has two memory controllers with three channels per controller for a total of six memory channels.

Does this mean that an ideal setup would have six ram sticks per CPU? Would memory access be slower with fewer sticks?


If you want to maximize the bandwidth of the memory subsystem, you'll want to install at least one DIMM per channel (6 in this case). If you want to install more than 6 DIMMs, you'll want to keep adding them in multiple of 6.

This is because each memory channel can be used independently, so if you only populate two slots, the CPU will write the data over two channels. If you populate all 6 channels, the CPU can write data to RAM over all 6 channels at the same time.


|improve this answer|||||

Of course this may vary from manufacturer to manufacturer but I've just looked at the specs for the HPE DL380 Gen10 and the Cisco C220 M5 and there's no speed penalty for using all six dimms per socket when using a silver 4xxx CPU - they'll all run at 2400 MT/s whether there's one or six DIMMs per CPU.

|improve this answer|||||
  • My concern was that fewer sticks would be slower. The clock would still be 2400, of course. But if the channels aren't full, perhaps less data can be accessed per clock tick. – Dave Nov 29 '17 at 16:24

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.