SSD DRAM Cache vs DRAM-less (HMB) Design: Real-World Impact on Sustained Performance
Two NVMe drives with near-identical rated sequential speeds can behave very differently once you throw a large, random-access workload at them. The difference usually traces back to whether the drive has its own dedicated DRAM chip or is borrowing a sliver of your system RAM instead.
Every SSD needs somewhere to store its flash translation layer (FTL) — the mapping table that tracks which physical NAND location corresponds to each logical block address the operating system asks for. Without fast access to this table, every read or write would require the controller to search through NAND itself for the mapping, which is drastically slower than reading it from memory. A DRAM-cached SSD keeps this entire mapping table in a dedicated onboard DRAM chip, giving the controller near-instant lookups regardless of how random or how large the workload is.
What DRAM-less drives do instead
DRAM-less drives remove that onboard chip to cut cost, reduce power draw, and shrink the board, which matters for compact form factors like M.2 2230 drives used in handhelds and small laptops. Instead of onboard DRAM, most modern DRAM-less NVMe drives use Host Memory Buffer (HMB), an NVMe protocol feature that lets the drive borrow a small allocation of the host system's own RAM — typically somewhere in the range of 32MB to 200MB depending on the drive — to store a working subset of the mapping table. This recovers a meaningful amount of the lookup performance a dedicated DRAM chip would provide, but it depends entirely on the host OS and driver correctly allocating and maintaining that HMB region, and it's a smaller, shared resource rather than a dedicated one.
Some budget or older DRAM-less drives skip HMB support entirely and instead cache only a small mapping table directly in the controller's own limited SRAM, falling back to reading larger portions of the mapping table from NAND itself when a requested address isn't cached. This is the configuration where performance degrades most severely under demanding workloads, because NAND reads for mapping lookups compete directly with NAND reads and writes for actual user data.
Where the difference actually shows up
For typical desktop use — booting an OS, launching applications, loading game levels, general file management — the practical difference between a DRAM-cached drive and a competent HMB-based DRAM-less drive is often small enough that most users wouldn't notice it in a blind test, because these workloads don't sustain large amounts of random small-block I/O for long enough to expose a mapping-table bottleneck. Sequential read and write speeds, which dominate marketing benchmarks, are also fairly similar between the two designs on modern controllers, because sequential access patterns are inherently easier to predict and cache regardless of DRAM presence.
The gap widens under sustained heavy random I/O: large database workloads, extended video editing with scrubbing across a long timeline, running multiple virtual machines from the same drive, or any workload that keeps the drive under continuous mixed read/write pressure for extended periods. In these scenarios, a DRAM-less drive without HMB, or with limited HMB allocation, can show noticeably higher latency spikes and lower sustained IOPS than a DRAM-cached equivalent, because the mapping-table lookup overhead compounds with the same write amplification and garbage collection pressure that already slows sustained writes as a drive fills up.
Why this matters more for capacity planning than for peak-speed shopping
Marketing sequential speed numbers on a spec sheet won't tell you which architecture a drive uses — that information is usually buried in a manufacturer's technical datasheet or independent teardown reviews that identify the controller and check for an onboard DRAM chip. For a boot drive that mostly handles OS files and application loading, a well-implemented DRAM-less HMB drive is a reasonable cost-saving choice with minimal practical downside. For a drive that will handle continuous heavy write workloads, large file transfers, or serve as working storage for professional editing, the dedicated DRAM cache remains the safer choice, and the price premium over a comparable DRAM-less drive is often smaller than the premium for stepping up a full capacity tier.