Having seen the spec sheets and benchmarks regarding the various NAND SSDs that have trickled onto the market, I’ve noticed that the enterprise SLC SSD tend to have write speeds that trail behind consumer SSDs using pSLC cache.
I don’t understand how pSLC can outpace real SLC, but suppose the consumer SSD is designed with a fixed percentage cap which can be used in pSLC mode (e.g., the new Crucial T750 reserves up to 11%), would artificially limiting the partition on the SSD to that size effectively make pSLC always on, thereby turning the TLC or QLC SSD into a pSLC SSD?
There are two objectives I believe this would achieve:
Extending the SSD lifetime by an order magnitude
Example: the Crucial T705 allocates up to 11%, or a ninth, of the capacity to pSLC mode. Constraining the writable area to that size should hypothetically extend the DWPD by 9 times (the wear characteristics of writing only in pSLC mode notwithstanding).
Getting pSLC write speeds, which are faster than what enterprise SLC SSDs typically offer
I believe some enterprise drives can actually be configured to run entirely in pSLC mode, as SLC is no longer manufactured, or rather, anything modern “SLC”, ie 128 layer or 192 layer SLC is “pSLC” anyway. I remember reading in some whitepaper for some drive, I don’t recall which, when, or where, that there were options to configure it pre-sale for pSLC deployment, so I doubt utilities to do this are widely available, and may not be available at all.
I believe it would provide the write speeds you’re after most of the time, but I think there’s a solid chance that, because it will try to demote data to TLC/QLC eventually, you may encounter inconsistent performance. It also isn’t going to effectively make it always use pSLC, because the drive will constantly try to demote data to TLC/QLC, you aren’t going to get the data retention, latency, and reliable performance of fully pSLC drives.
Interesting… Quick search didn’t turn up exactly this with enterprise SSDs, but did turn up configuring embedded NAND (e.g., eMMC and UFS) with such a feature.
I remembered seeing something to that effect when reading a bit deeper into Kioxia’s new SLC offerings (or that it’s really MLC in pSLC mode). It’s possible that it’s also true of those from Micron and Solidigm. I can’t remember where I found it specifically.
Tech Insights phrasing suggests that it’s non-native SLC. “Key features of KIOXIA XL-FLASH include 128 Gb die, SLC mode, 16 plane die design, compatible FLASH protocol and 4KB page size to fit into SCM application in the memory hierarchy.”
I still find it odd that consumer SSDs with pSLC cache can absorb writes quick enough to nearly saturate the PCIe interface while enterprise SSDs with “SLC” struggle to get three-quarters of the way there. Perhaps consumer SSDs are making a Faustian bargain for that performance. I’d be curious to know what is sacrificed in the process.
The reason Crucial can do it on a consumer SSD is they understand their flash memory since its their designs so there is more control like Samsung and Kioxia. If you buy consumer SSDs from non-OEMs, they either use a controller using internal cache or push customers into installing their toolkit for a system memory cache.
Consumer SSDs change their flash suppliers often, it may of been the reason Corsair had a bunch of models that suffered varying levels of failure rates. Corsair SSDs have used Toshiba(Kioxia), SK Hynix, Crucial and SanDisk(pre-WD) chips.
(577.0 KB)