Home / Component / Memory / Early performance tests reveal significant throughput penalty for HUDIMM DDR5

Early performance tests reveal significant throughput penalty for HUDIMM DDR5

Intel, in partnership with various partners, has introduced the HUDIMM (Half-Unbuffered DIMM) standard to address the ongoing global memory shortage and rising hardware costs. This new specification aims to lower the barrier to entry for DDR5 platforms by simplifying the internal architecture of the memory modules. However, this new module type appears to have disadvantages, particularly in terms of performance.

The core design of HUDIMM involves transitioning from the standard two 32-bit subchannels (forming a 64-bit-wide bus) found on traditional UDIMMs to a single 32-bit subchannel. By populating only half of the memory banks, manufacturers can produce modules using fewer DRAM chips, directly reducing production costs. Initial releases are expected in 8GB and 12GB capacities, targeted primarily at entry-level office systems and budget-conscious builds.

Image credit: HKEPC

Despite the cost benefits, performance benchmarks conducted by HKEPC and Asus (via Wccftech) show a severe trade-off. The test was performed by selecting a standard UDIMM and converting it into a HUDIMM module by shielding half of the banks. The transition to a single 32-bit subchannel results in an approximate 45–50% loss in total bandwidth across read, write, and copy operations. While memory latency remained relatively stable at 85–87ns, the throughput of a dual-channel HUDIMM setup was roughly equivalent to that of a single standard UDIMM running in single-channel mode.

One redeeming feature is the support for asymmetrical dual-channel operation. This allows users to mix different module types, like HUDIMM and UDIMM, to achieve unique capacities while maintaining three active 32-bit subchannels. However, for users prioritising speed, the significant bandwidth reduction makes HUDIMM only a solution for systems where peak memory performance is secondary to overall cost efficiency.

KitGuru says: Given the performance trade-offs (and without knowing the pricing of HUDIMMs), do you think budget builders will be willing to sacrifice half their bandwidth to get a modern platform running, or will they stick to older DDR4 systems?

Become a Patron!

Check Also

Micron reportedly developing vertically stacked GDDR to rival HBM in AI Workloads

Micron is reportedly working on a major architectural shift in graphics memory that could change …