The number before "R" is rank count - independent sets of DRAM chips a module presents to the controller; the number after "x" is each chip's data width in bits. Lenovo Press's own ThinkSystem ordering tables list 1Rx4, 2Rx4, 1Rx8 and 2Rx8 as real, currently-shipping part variants of the same 32GB capacity - and the width choice changes RAS capability, not just a label: standard-width modules support Single Device Data Correction where narrower 9x4 modules documented by Lenovo do not.
Updated ยท vendor documents read 16 September 2026
Rank notation packs two numbers into a short code. Lenovo Press's own ThinkSystem memory-summary ordering tables give concrete, currently-listed examples of all four common combinations at one capacity: "ThinkSystem 32GB TruDDR5 6400MHz (1Rx4) RDIMM", "(2Rx4) RDIMM", "(1Rx8) RDIMM", and "(2Rx8) RDIMM" are all real, distinct shipping parts - same brand, same capacity, same speed, different rank/width organization.
| Code | Ranks | Chip width |
|---|---|---|
| 1Rx4 | 1 (single rank) | 4 bits per chip |
| 2Rx4 | 2 (dual rank) | 4 bits per chip |
| 1Rx8 | 1 (single rank) | 8 bits per chip |
| 2Rx8 | 2 (dual rank) | 8 bits per chip |
| 4Rx4 | 4 (quad rank) | 4 bits per chip - seen on Lenovo's 256GB 3DS RDIMM parts |
Lenovo Press documents a real RAS (reliability) consequence of the width choice, not just a naming difference: "Standard RDIMMs and 3DS RDIMMs support two 40-bit subchannels (that is, a total of 80 bits), whereas 9x4 RDIMMs support two 36-bit subchannels (a total of 72 bits)... The extra bits in the subchannels allow standard RDIMMs and 3DS RDIMMs to support Single Device Data Correction (SDDC), however 9x4 RDIMMs do not support SDDC."
A narrower chip organization can trade away a specific error-correction capability - that is a documented reliability property, stated by the vendor, not a speed claim we are making about one width versus another.
This is platform-specific rather than a fixed industry ceiling. Lenovo Press gives a worked example instead of a maximum: for a particular 2-socket Intel Xeon Scalable configuration, "Each channel is populated with two dual-rank (2R) DIMMs, so the total number of ranks per channel is 4R." That is an illustration of how ranks stack across DIMMs sharing one channel, not a stated ceiling for every platform - see our population-rules page for the channel and DPC rules this interacts with.
A single stated capacity - 32GB, for example - is not one fixed internal layout. The same nominal capacity can be reached with fewer, wider ranks or more, narrower ranks, which is exactly why Lenovo Press's own ordering tables list 1Rx4, 2Rx4, 1Rx8 and 2Rx8 all at 32GB rather than picking one. A buyer comparing two listings at the same capacity and price is not necessarily comparing the same internal organization, and without the rank code neither listing tells you which one you're getting - which is the practical reason this notation exists on a spec sheet at all, separate from any performance question.
At the higher end of the capacity ladder, rank count tends to climb along with capacity rather than staying fixed. Samsung's own current DDR5 RDIMM lineup lists capacities up to 256GB per module; reaching that ceiling on a 3DS-stacked part is part of why Lenovo Press's own 256GB TruDDR5 parts appear at 4Rx4 and 8Rx4 rather than at 1Rx4 or 2Rx4 - more physical DRAM inside the package tends to mean more ranks presented to the controller, not a fixed rank count that simply gets denser per rank. None of this is a claim about speed; it is only about how many independently-addressable rank groups a given capacity is built from.
Some platform features narrow which rank/width combinations are eligible at all. HPE's ProLiant DL360 Gen11 documentation states: "If different-rank memory is mixed, the quantity of each part number must be the same; only x4 options with Memory Fault Tolerance." A reliability feature like Memory Fault Tolerance can rule out x8-organized modules entirely, independent of capacity or speed - another reason rank/width is not a cosmetic detail on a spec sheet.
We checked every RAM listing currently in our catalogue - model, full model and series text - against rank notation patterns like 1Rx4/1Rx8/2Rx4/2Rx8, and found zero confirmed matches. That is not a claim that none of these modules have a documented rank organization; it means sellers overwhelmingly do not print it in listing titles, and our own part-number decoders for the brands we carry explicitly mark the rank segment as undocumented (rather than guessed) where the manufacturer's own grammar does not confirm it. Rather than show a table of unverified rank claims, the live picks below are simply our cheapest current listings, unfiltered by rank.
Which rank/width combinations your specific server supports - and any feature-linked restriction like the HPE example above - is set by the platform, not by this page. See our OEM-verified platform pages for models we have checked directly.
No in-stock listings in our catalogue currently carry confirmed rank notation, per the gap explained above - these are our cheapest current listings generally.
| Brand | Module | Capacity | Type | $/GB | |
|---|---|---|---|---|---|
| SK Hynix | HMAA8GL7MMR4N | 64GB DDR4 | RDIMM | $4.61 | Buy Now |
| A-Tech | M393A2G40DB0-CPB | 16GB DDR4 | RDIMM | $5.44 | Buy Now |
| Samsung | M393A8K40B22-CWD | 64GB DDR4 | RDIMM | $6.23 | Buy Now |
| Supermicro | B07DFLTWN3 | 64GB DDR4 | LRDIMM | $6.23 | Buy Now |
| Samsung | M393A2G40EB1-CPB | 16GB DDR4 | RDIMM | $7.12 | Buy Now |
| MICRON | MTA36ASF4G72PZ | 32GB DDR4 | RDIMM | $7.19 | Buy Now |
| Samsung | M393A2K43CB2-CTD8Q | 16GB DDR4 | RDIMM | $8.75 | Buy Now |
| SK Hynix | HMA84GR7CJR4N-VK | 32GB DDR4 | RDIMM | $8.84 | Buy Now |
The number before R is the rank count (how many independent sets of DRAM chips the module presents to the controller); the number after x is the data width of each chip in bits. Lenovo Press's own ThinkSystem ordering tables give real shipping examples: "ThinkSystem 32GB TruDDR5 6400MHz (1Rx4) RDIMM", "(2Rx4) RDIMM", "(1Rx8) RDIMM", "(2Rx8) RDIMM" all appear as genuinely distinct, currently-listed parts.
Per Lenovo Press, it changes RAS (reliability) capability, not just a label: "Standard RDIMMs and 3DS RDIMMs support two 40-bit subchannels (that is, a total of 80 bits), whereas 9x4 RDIMMs support two 36-bit subchannels (a total of 72 bits)... The extra bits in the subchannels allow standard RDIMMs and 3DS RDIMMs to support Single Device Data Correction (SDDC), however 9x4 RDIMMs do not support SDDC." A narrower width buys back some capacity/cost at the expense of that RAS feature.
It's platform-specific and documented per server, not a fixed industry number. Lenovo Press gives a worked example rather than a maximum: "Each channel is populated with two dual-rank (2R) DIMMs, so the total number of ranks per channel is 4R" for a specific 2-socket Intel configuration - that is an example of stacking, not a stated ceiling.
Yes, and it can be tied to specific features. HPE's ProLiant DL360 Gen11 documentation: "If different-rank memory is mixed, the quantity of each part number must be the same; only x4 options with Memory Fault Tolerance." A feature like Memory Fault Tolerance can narrow the field to x4-organized modules specifically.
Sometimes, under documented conditions - see our mixing-rules matrix for the full picture. It is not a blanket ban the way mixing module classes is, but it is also not unrestricted.
Not currently in a way we can verify. Checking every listing's model text, full model text and series field against rank notation (1Rx4/1Rx8/2Rx4/2Rx8-style patterns) across our entire RAM catalogue returned zero confirmed matches - sellers largely don't print rank organization in listing titles, and our own part-number decoders for the brands we carry explicitly mark the rank segment as undocumented rather than guess at it. We're stating that gap directly below instead of showing a table we can't stand behind.
Specification claims on this page were checked against the pages below on 16 September 2026. Where a widely-repeated figure did not survive that check, we report what the source says now and say so in the text rather than repeating the familiar number.