16 GB DDR4-2400 PC4-19200 ECC registered RDIMM, 1Rx4 on eighteen 8 Gb x4 packages, 1.2 V, 288-pin; eighteen ordering codes under the M393A2K40BB1 and M393A2K40CB1 revisions
| Capacity | 16 GB |
|---|---|
| Memory Type | DDR4 |
| Speed | 2400 MT/s |
| Form Factor | RDIMM |
| Rank | 1Rx4 |
| ECC | ECC |
| Voltage | 1.2 V |
| Module Height | Standard |
| Condition | New · Used |
Samsung 16 GB DDR4-2400 ECC registered RDIMM at standard height, one rank of eighteen x4 devices on a 288-pin board at 1.2 V. PC4-19200 at CAS latency 17, 14.17 ns at the 1,200 MHz clock, and 19.2 GB/s peak per 64-bit channel.
Of the four common DDR4 bins, 2,400 is the one where rank count starts deciding whether a channel keeps its clock. At 2,133 a board will usually carry four ranks and stay at speed without comment; at 2,400 the same four ranks are where conservative firmware begins to ease back, and nothing in the operating system reports the change. A single-rank sixteen-gigabyte module is the least expensive way out of that corner, because two of them per channel occupy two ranks rather than four and no capacity is surrendered to get there.
Two die revisions carry it here, M393A2K40BB1 and M393A2K40CB1, with nine ordering codes each. Eighteen codes in one table at the foot of the page, and one set of specifications above them.
| Part number | M393A2K40BB1-CRC and M393A2K40CB1-CRC — two revisions of one module, with the eighteen ordering codes listed under Ordering codes (catalogue codes M393A2K40BB1 and M393A2K40CB1) |
|---|---|
| Capacity | 16 GB per module, single module |
| Organisation | 1Rx4 — one rank of eighteen x4 devices at 8 Gb each, 144 Gb of silicon delivering 128 Gb of data plus the eight check bits that travel with every 64-bit transfer; 2G x 72 at module level |
| Speed | DDR4-2400, PC4-19200, 2,400 MT/s, 19.2 GB/s peak per 64-bit channel |
| Timings | CL17 at 1,200 MHz, 14.17 ns; JEDEC 2400T bin |
| Voltage | 1.2 V nominal |
| Form factor | 288-pin RDIMM at standard height, 31.25 mm, bare PCB with no heatspreader |
| Register | On-module register buffers the command and address bus, so the board has to be specified for registered memory |
| Error handling | ECC across the 64-bit data path with x4 device granularity, which is what lets a controller retire a failing device and carry on |
| Manufacture | Samsung; assembly site and date code vary by production run and are printed on the module rather than fixed by the part number |
| Condition | New and tested pulls, both stocked |
The 288-pin keying, the register and the capacity are all fixed inside the part number; the label adds production detail around it. Read as one string, the two get confused, and the fields that matter for compatibility sit in the part number alone.
| Printed as | Means | Why it matters at purchase |
|---|---|---|
| M393 | Standard-height registered module | The 31.25 mm board; the low-profile version of the same capacity carries M392 |
| 2K | Sixteen-gigabyte class on 8 Gb devices | Fixes both capacity and device density, which is why eighteen packages are sufficient |
| 40 | x4 device width | The width that gives chipkill granularity; an x8 build of the same capacity gives it up |
| 1Rx4 as printed on most listings | Single rank, x4 devices | One rank against a channel budget, so four modules per channel remain inside the usual table |
| PC4-19200T | DDR4-2400, bin T, which is CL17 | The letter states the timing grade; a listing that quotes only the clock has left latency open |
| -R | Registered, not unbuffered | A different electrical interface from an unbuffered module of the same capacity, and it will not train in that slot |
| BB1 / CB1 | Successive die revisions of one design | Records which silicon generation built the assembly; both are listed with the same organisation, clock and voltage |
| CRC | The 2,400 speed block | Neighbouring blocks in this family are other clocks, so the block letter has to be checked rather than guessed |
| 00, 0Q, 4Y … | Ordering suffix, two characters throughout | What the purchase order quotes; the only field that differs inside one revision |
| YYWW | Date code, year and week | The only service history a pull carries, and the first field to read when judging remaining life |
DDR4 server boards generally allow four ranks per channel. Two 2Rx4 modules reach that limit and the channel is full; four of these reach it just as well, but only after four modules are installed. Between those two statements sits the whole argument for the single-rank build, and the clock at which it matters is 2,400, not 2,133.
The reason is the timing window. Raising the clock shortens every interval the controller has to close, and the number of loads it has to close them across does not shrink with it. Two modules of 2Rx4 place four ranks of thirty-six packages on the channel; two modules of 1Rx4 place two ranks of eighteen. Same capacity, half the rank count, and the controller keeps the extra slack at the exact point where the faster bin consumed it. Boards that quietly drop a fully populated 2,400 channel to 2,133 will often hold 2,400 with this part in the same slots.
Population arithmetic follows directly. Eight channels with two modules each is 256 GB at two ranks per channel, and twelve channels gives 384 GB at the same rank count. Reaching the same capacity with 2Rx4 modules means the same number of modules in the same slots and twice the ranks, which is why the choice is not really about gigabytes.
Eight gigabytes per device is what makes one rank enough. A x4 rank is seventy-two bits wide, so it needs eighteen devices; at 8 Gb each those eighteen packages already hold sixteen gigabytes of data. The 4 Gb alternative needs thirty-six packages and a second rank for the same capacity, which is the build this module competes with rather than the one it replaces.
BB1 and CB1 are successive silicon generations of the same raw card, and BB0 appears against the 2,133 block. The letter records which devices were used; it does not change the rank, the width, the capacity, the timing grade or anything the board trains against. Two modules from different revisions in one channel run as designed, and pairing by revision is a bookkeeping preference rather than a technical requirement.
Samsung lists the CB1 revision as end of life on its own product page, with the same sixteen gigabytes, the same 1R x 4 organisation of eighteen 8 Gb devices, the same 2,400 clock and the same 1.2 V supply as the BB1 build. That status describes the catalogue rather than the hardware: the line has moved to the secondary market, and what circulates is undeployed stock plus tested pulls from retired fleets.
Against the 2,133 grade the step is 19.2 GB/s versus 17.1, about twelve per cent, for an absolute latency that moves from 14.06 ns to 14.17 — close enough that no workload will notice. In eight-channel terms it is 154 GB/s against 137, and in twelve-channel terms 230 against 205.
Going up to 2,666 buys another eleven per cent of bandwidth and costs a little more latency, 14.25 ns, but it also narrows the window again. That is the point at which even a single-rank module can be asked to give up the clock on a densely populated board, and where the sensible answer is usually to stay at 2,400 and spend the difference on capacity. The three bins do not differ much in what a benchmark reports; they differ in how much population a board tolerates before it derates.
The module sets its own ceiling and the platform sets the other, and the channel runs at the lower of the two:
Each of the three trains without configuration, and none is damaged by the mismatch in rating. What the part number describes is the module; what the machine does with it is a separate decision taken in firmware.
Eighteen codes carry this module, nine under each revision, and the nine suffixes repeat exactly between the two families. The family code fixes height class, density, device width and revision, CRC fixes the 2,400 speed block, and the last two characters name the build. Every tail in this block is two characters, which is the opposite of the 2,133 block where lengths run from one to three, so a comparison that assumes a fixed width will work here and fail there.
Because the suffix set is identical across BB1 and CB1, a code read off one family transposes to the other without ambiguity — but only after the revision letters are swapped as well. What a purchase order should name is the full code, since the two families are stocked as one line and either satisfies the specification.
| Code | Build | Build tail | Condition | Quote |
|---|---|---|---|---|
| M393A2K40BB1 | BB1 catalogue | Family code without a speed block; the BB1 build listed here is the PC4-19200 / 2,400 MT/s one | New / used | Enquire |
| M393A2K40BB1-CRC | BB1 | Base code, no suffix appended | New / used | Enquire |
| M393A2K40BB1-CRC00 | BB1 | 00 | New / used | Enquire |
| M393A2K40BB1-CRC0Q | BB1 | 0Q | New / used | Enquire |
| M393A2K40BB1-CRC0Y | BB1 | 0Y | New / used | Enquire |
| M393A2K40BB1-CRC30 | BB1 | 30 | New / used | Enquire |
| M393A2K40BB1-CRC3Q | BB1 | 3Q | New / used | Enquire |
| M393A2K40BB1-CRC40 | BB1 | 40 | New / used | Enquire |
| M393A2K40BB1-CRC4Q | BB1 | 4Q | New / used | Enquire |
| M393A2K40BB1-CRC4Y | BB1 | 4Y | New / used | Enquire |
| M393A2K40CB1 | CB1 catalogue | Family code without a speed block; the CB1 build listed here is the PC4-19200 / 2,400 MT/s one | New / used | Enquire |
| M393A2K40CB1-CRC | CB1 | Base code, no suffix appended | New / used | Enquire |
| M393A2K40CB1-CRC00 | CB1 | 00 | New / used | Enquire |
| M393A2K40CB1-CRC0Q | CB1 | 0Q | New / used | Enquire |
| M393A2K40CB1-CRC0Y | CB1 | 0Y | New / used | Enquire |
| M393A2K40CB1-CRC30 | CB1 | 30 | New / used | Enquire |
| M393A2K40CB1-CRC3Q | CB1 | 3Q | New / used | Enquire |
| M393A2K40CB1-CRC40 | CB1 | 40 | New / used | Enquire |
| M393A2K40CB1-CRC4Q | CB1 | 4Q | New / used | Enquire |
| M393A2K40CB1-CRC4Y | CB1 | 4Y | New / used | Enquire |
Every module leaves with one year of cover counted from dispatch, new stock and tested pull alike, and the cover is recorded against the ordering code and the batch rather than against the invoice. Pulls pass a matched-set screen before packing and the screening record ships inside the box. Shelf stock goes out in three to seven working days with tracking, and the smallest order accepted is two modules.
No figure is printed here because the line is quoted per order: two conditions, eighteen codes and a per-piece number that moves with quantity. Send the code, the quantity, the condition and the platform, and the reply comes back the same working day.
Tell us who you are — we reply within one working day with a firm quote, stock status and lead time. Volume discounts available.
16 GB DDR4-2933 PC4-23400 ECC registered RDIMM, 2Rx8 on eighteen 8 Gb x8 packages, 1.2 V, 288-pin; six ordering codes under M393A2K43CB2-CVF with BY, CO, BQ, C0 and GQ tails
16 GB DDR4-2400 PC4-19200 ECC registered RDIMM, 2Rx8 on eighteen 8 Gb x8 packages, 1.2 V, 288-pin; eighteen ordering codes under the M393A2K43BB1 and M393A2K43CB1 revisions
16 GB DDR4-2133 PC4-17000 ECC registered RDIMM, 2Rx8 on eighteen 8 Gb x8 packages, 1.2 V, 288-pin; nine ordering codes from M393A2K43BB1-CPB through the CPBQ tail
16 GB DDR4-2666 PC4-21300 ECC registered RDIMM, 2Rx8 on eighteen 8 Gb x8 packages, 1.2 V, 288-pin; six ordering codes under M393A2K43BB1-CTD with 6Q, 6Y, 7Q, 7Y and 8Q tails