HPE ProLiant is the perennial number two of the server install base, and its owners face the same badly served question as Dell's: every site answering 'what RAM fits my DL380' is selling its own memory. This hub answers it neutrally — each ProLiant model below carries rules verified against HPE's official QuickSpecs, wired to our live cross-brand price tracker, so the compatibility answer arrives attached to the actual market rather than a configurator's markup.
The platform list splits on the 2026 fault line: Gen11 machines (DL380, DL360) run DDR5 on 4th/5th-gen Xeon Scalable at shortage-premium prices, while the Gen10 fleet — the DL380 and DL360 workhorses plus the ML350 tower — runs DDR4 from the affordable side of the market. The Gen10 wave is also the one flowing off leases into the secondary market, where its 24-slot chassis meet the tested-pull DDR4 that the same decommissioning wave supplies.
HPE's memory vocabulary deserves one demystifying note: 'HPE SmartMemory' is HPE's brand for the standard registered ECC modules it validates — the silicon is the same Samsung, SK Hynix and Micron output as everyone else's. SmartMemory carries HPE-specific authentication that some management features read, but the platforms run standard spec-compliant RDIMMs of the documented type and grade; production fleets wanting full iLO amenities can weigh that against the cross-brand prices in our tables.
The QuickSpecs are precise about speed ladders, and the model pages carry each platform's exact figures: Gen10 machines split 2666/2933 MT/s across 1st and 2nd-gen Xeon Scalable CPUs, while Gen11 DDR5 platforms step down a grade as channels take their second DIMM — the same physics as every current platform. Faster-rated modules downclock harmlessly everywhere, which our compatibility tables mark per module so grade-shopping on price stays safe.
HPE's population discipline follows the industry pattern with its own accents: channels populated evenly, capacities uniform within a channel, and the QuickSpecs' supported-configuration tables as the arbiter of what HPE actually validates. Where the QuickSpecs state a rule, the model page's rules box quotes it; where they are silent, the page is silent too — this cluster's standing policy is that unverified claims simply do not render.
For the used-market buyer, Gen10 ProLiants are arguably the sleeper pick of the off-lease wave: the same Xeon Scalable internals as the Dell R740 crowd, iLO management that hobbyists either love or license around, and street prices that frequently undercut the equivalent Dell chassis. The memory play is identical — abundant tested DDR4 pulls, the standard memtest86+-on-arrival ritual, and matched modules per channel.
Using this hub: know your model, go to its page — verified rules box, live compatibility table sorted by $/GB, and upgrade scenarios computed from today's listings. Comparing platforms, the list below shows each model's compatible-module count and best live $/GB side by side. Everything inherits the site's rules: 1024-based $/GB, out-of-stock modules excluded from rankings, four byte-identical freshness signals bound to the real refresh, and a source link on every spec.
Reading HPE QuickSpecs well is half the buying skill, and the model pages do the reading for you — but the pattern is worth internalizing: HPE states memory speed per processor tier (not per platform), states population validity in supported-configuration tables, and buries chassis-variant limits (like DIMM-count ceilings on drive-dense variants) in footnotes that bite capacity plans. When a spec here surprises you, the linked QuickSpecs paragraph is the arbiter.
The Gen10-versus-Gen11 economics mirror the industry's DDR4/DDR5 fork, with one HPE accent: the Gen10 installed base is so large that its memory market — both new remaining stock and decommission pulls — is among the deepest anywhere, keeping upgrade prices honest even mid-shortage. A Gen10 fleet with empty slots is sitting on the cheapest capacity expansion in the HPE universe, and the model pages price it live.
For production fleets weighing SmartMemory against cross-brand parts, the practical split is feature-dependence: if your operations lean on the iLO memory-health amenities tied to authenticated modules, the premium buys something you use; if your monitoring lives in the OS and standard ECC reporting, it does not. Either way the platforms run spec-compliant registered modules, and either way the live tables here tell you what the market version costs.
The 1U/2U choice inside each generation changes nothing about memory compatibility — DL360 and DL380 share rules within their generation — so fleet standardization on one module part number across both form factors is realistic and worth the procurement leverage. Pick from the live table's cheapest reliable tier, validate once, deploy everywhere.
And the towers: the ML350 carries full rack-class memory rules into office environments, which makes it this hub's reminder that 'server memory' is not a datacenter-only concern. Small businesses running one tower get the same verified-rules treatment as a thousand-node fleet — arguably they need it more, since nobody down the hall is a memory specialist.
On warranty and support interactions, the perennial ProLiant question deserves a plain answer: installing third-party memory does not void an HPE server warranty — warranty law and HPE's own terms cover HPE components regardless — but support may attribute memory-adjacent faults to non-HPE modules first. Fleets weigh that support-friction cost against the price delta in the tables; homelabs, running without support contracts anyway, weigh nothing at all.
Capacity planning across a mixed Gen10/Gen11 estate benefits from one framing: treat the two generations as separate memory economies with separate budgets. DDR4 spend on Gen10 machines buys cheap runway; DDR5 spend on Gen11 buys expensive necessity. Blending them into one 'memory budget' hides the fact that a gigabyte of the former costs a quarter of the latter — and that misallocation is where shortage-era budgets quietly leak.
The scenarios on each model page are deliberately conservative: they never exceed the platform's slot count, price only from in-stock listings, and note the channel-balance guidance the QuickSpecs provide. Treat them as baselines to beat, not quotes to accept — the market moves every few hours, and the table above each scenario is the tool for beating it.
One last practical pointer for readers arriving from search with a model number in hand: the fastest route through this hub is the platform list above — each entry's live module count and best $/GB are drawn from the same tracker that powers the model pages, so the number you click through on is the number you land on. Nothing here is a teaser figure; it is the market, sampled a few hours ago at most.
The platforms run standard registered ECC modules of the documented type and speed — SmartMemory is HPE's validated-and-branded version of the same manufacturers' silicon, with authentication some management features read. Third-party modules of the correct spec work; fleets that want every iLO feature can price the premium against the live tables.
Gen10 (and Gen10 Plus) runs DDR4 on 1st/2nd-gen Xeon Scalable; Gen11 moved to DDR5 on 4th/5th-gen. The boundary is absolute — modules cannot cross it — which makes 'upgrade the Gen10's DDR4 now or buy into Gen11's DDR5' a genuine economic fork during the shortage, one our upgrade-economics pages help price.
Almost always CPU generation or population: 1st-gen Xeon Scalable caps DDR4 at 2666 MT/s where 2nd-gen runs 2933, and DDR5 Gen11 platforms drop a grade at two DIMMs per channel. Both behaviors are documented in the QuickSpecs and stated in each model page's rules box.
Yes — Xeon Scalable internals, deep DDR4 slot counts, and street prices that often undercut equivalent Dell units. Two checks before buying memory: which CPU generation the unit carries (it sets the speed ceiling), and iLO licensing state if you care about remote management amenities.
HPE's supported configurations favor uniformity — identical modules per channel, capacities not mixed within a channel. Where the QuickSpecs for a specific model state mixing rules, the model page quotes them; the universally safe pattern remains matched part numbers in the counts the documentation validates.