122.88TB model of the D5-P5336 series
Not so long ago, one could wonder how long it would take SSDs to overtake hard disk drives in capacity. It’s a while since that has happened, but the capacity of electronic storage based on NAND Flash memory continues to grow at a much faster rate than that of magnetic platters. While WD and Seagate have only just passed the 30TB capacity mark, SSDs now offer four times that. The price per unit of capacity, however, will be somewhere else.
Last summer, we wrote about the 61.44TB SSD released by Solidigm. If you don’t recognize the name, Solidigm is basically Intel’s former SSD division, which was bought out by Hynix years ago, who continues to run Solidigm as their sub-brand. That same Solidigm is now coming out with another apparently record-breaking SSD with a capacity of 122.88 TB, so double of the previous drive – and give or take four times the capacity of the largest HDDs currently existing, which are now around 30–32 TB.
This SSD belongs to the same D5-P5336 series as last year’s 61.44TB drive, so it is to some extent a new capacity SKU rather than a new drive. As you can probably guess from the not-so-attractive name, these are storage devices meant for servers and datacenter disk arrays, not really for PCs.
Solidigm uses a proprietary NVMe 2.0 controller that has a PCIe 4.0 ×4 interface (so it can’t use PCIe 5.0 yet, which is a pity, performance-wise) in the drive. QLC NAND is used again, these should be Solidigm’s 192-layer chips, which should be descendants of the QLC NAND technology developed by Intel before it got rid of their NAND business.
According to the manufacturer, the Solidigm D5-P5336 achieves a sequential read performance of 7400 MB/s, while the sequential write performance is lower at 3200 MB/s. This could probably be a figure representing the sustained write speed in QLC mode instead of the speed for the “pseudoSLC” write cache we are used to from consumer SSDs (if the SSD had such a buffer, the speed would be higher). The random access speed is 930,000 IOPS in reads, but only 25,000 IOPS in writes. However, this is not a figure for writing the usual 4KB blocks, but for larger 32KB blocks, and for a queue depth of 256 items, so having a fairly deep buffer that allows for better optimization of writes (SSDs usually provide a queue depth of 32).
These storage devices are intended for applications where data is mainly read and less commonly written or overwritten, which is something already implied by the use of QLC NAND with its naturally lower erase endurance. According to Solidigm, however, these SSDs have an officially guaranteed endurance of 0.6 total writes per day for five years, for a total of 134.3 PB of total writes. This works out to some 1100 rewrite cycles per cell, which is high for a QLC drive. Consumer QLC SSDs are usually only rated for 300 cycles.
In addition to the usual disadvantages of QLC NAND, the high capacity of this SSD comes at the price of one more complication – the power consumption reaches up to 25 W under load (but should be less than 5 W in idle). Datacenter SSDs often have controllers with higher power consumption than regular M.2 modules for PCs. But in this case, the high power consumption is likely due to the large number of NAND chips that are used as well (the controller is quite likely to have more channels than the regular ones, server controllers are often 16-channel). This power consumption (and dissipation) will already require cooling to be addressed, which in servers is provided by forced drafts through the rack chassis.
The Solidigm D5-P5336 SSD family comes in three form factors: as the E1.L and E3.S modules, which are something used exclusively in servers, but also in the U.2 design, which is a 2.5″ form factor based on the format of the once-used laptop hard drives (and still extant 2.5″ SATA SSDs).
You can actually install this version in a PC in theory. Its U.2 interface used to be present on some older high-end motherboards (or workstation models), and if it isn’t, it can be brought out with an adapter from a PCIe 4.0 ×4 slot or from an M.2 slot with PCIe 4.0 ×4 connectivity. Of course, you need to provide cooling in the 2.5″ drive bay or mounting position, which is something most of today’s cases don’t account for. For example the currently popular mounting of 2.5″ drives behind the motherboard tray in the cable routing space would almost certainly lead to a severe if not catastrophic overheating issues with a 25W SSD drive.
But this storage is going to be very expensive, as we don’t expect the price per terabyte of capacity to be better than or even equal to HDDs or consumer SSDs. For example, a regular consumer QLC SSD (Samsung 870 QVO) costs around 250 EUR for a 4TB capacity (the cheapest QLC modules like Kingston NV2 can be found for slightly under 250 € but are not that much cheaper). Multiply that by thirty-one and you have a sort of lower bound for how much the price sticker on a 122.88TB SSD could ask. Realistically, the price will probably be well above that, because you’re paying extra for this drive being a specialty product, and in general all server SSDs are more expensive than those for ordinary consumer market. This also means there will continue to be room for magnetic/mechanical HDDs to exist on the market, precisely because of the price per terabyte advantage.
Sources: Solidigm, Tom’s Hardware
English translation and edit by Jozef Dudáš
⠀
⠀
