Multi-channel memory architecture

Lore & Background

Multi-channel memory requires a capable memory controller, which may be on the motherboard or chipset or integrated into the CPU. Memory modules must be installed into matching banks, each belonging to a different channel. For dual-channel use, a matched pair of modules may usually be placed in the first bank of each channel, with a different-capacity pair in the second bank. Modules rated at different speeds can be run in multi-channel mode, but the motherboard will run all modules at the speed of the slowest module. Some motherboards have compatibility issues with certain brands or models when used in multi-channel mode, so identical pairs are generally advised. A matching pair must match in capacity, speed, memory timing (including CAS latency), and number of memory ranks and bus width of each chip. Only clock and timing are absolutely required to match, as the channels work in lockstep. Size and rank count requirements have loosened over time, starting with Intel's Flex Memory in 2004, which allows two modules of different size to work in a combination of dual and single channel. AMD introduced an equivalent feature some time before 2015. Mixing of bus widths remains flaky, especially for ECC setups. Some systems such as Broadwell-EP do not support Flex Memory but allow capacity in the form of 'summing'.

Reader's Guide

Multi-channel memory architecture has been a key technology for increasing memory bandwidth since the 1960s. Dual-channel configurations theoretically double memory bandwidth and reduce latency compared to single-channel, though this should not be confused with double data rate (DDR) memory. Dual-channel was originally conceived as a way to maximize throughput by combining two 64-bit buses into a single 128-bit bus (ganged mode), but due to lackluster performance gains in consumer applications, modern implementations use unganged mode by default, maintaining two 64-bit buses with independent access to support multithreading. The technique has scaled to higher channel counts: quad-channel memory is supported by modern high-end desktop and workstation processors such as the AMD Ryzen Threadripper series and Intel Core i9 Extreme Edition lineup, while server processors from AMD Epyc and Intel Xeon platforms support from quad-channel up to 12-channel layouts. Triple-channel architecture was used in the Intel Core i7-900 series on the LGA 1366 platform with DDR3 memory. With DDR5, each DIMM has two independent 32-bit sub-channels, analogous to unganged operation. GPU VRAMs also use multi-channel concepts, with memory controllers serving one or two chips via sub-channels. The technology continues to evolve, with Flex Memory allowing mixing of different-sized modules in some configurations, though mixing module types is generally not documented as supported for triple-channel and above, especially on server hardware.

Did You Know?

• The technique goes back as far as the 1960s, used in the IBM System/360 Model 91 and CDC 6600.
• Dual-channel should not be confused with double data rate (DDR); the two technologies are independent.
• The AlphaStation 600 (1995) had a chipset supporting eight-channel memory, but the backplane limited operation to four channels.