Info My first QLC SSD experience

[mikeymikec]

I imagine most people are aware of the existence of QLC SSDs and their lacklustre performance, but I thought I'd add my experience to the pile.

A Lenovo V15 ADA laptop (Ryzen 5 3500U, 8GB RAM) with a 500GB SSD has this model SSD in: Crucial mtfdhba512qfd. According to notebookcheck.net this is a QLC drive (I'd consult the specs direct from Crucial but it requires a login for ... reasons).

I was transferring about 200GB of data from a USB mSATA drive I have used plenty of times before which is really handy for high-speed transfers. To begin with, the transfer speed was what I'd expect from my mSATA drive (~300MB/sec), but then it soon dropped to 60-70MB/sec (which I've never seen this mSATA drive do, and I've used it for plenty of transfers). I'm aware that SSDs' write performance suffers once the write cache has been exhausted so as soon as I saw the drop by that much I assumed it must be a QLC drive, but the thing that really surprised me was that evidently read performance also suffers greatly in this scenario. The USB SSD's usage % was about 20%, the internal SSD was at 100% with response times exceeding 2000ms. Normally, a Windows system with a decent SSD would allow me to do such a transfer while I also try to open a basic app like Windows calculator, and it would open it in say 1-2 seconds. This laptop however was acting like it had a HDD in, I'd say the calculator app took about 10 seconds to start during the data transfer.

I'm not 100% sure if this is mainly a QLC issue or maybe a DRAM-less SSD issue, but I'm inclined to think it's QLC related because I've sold this model of laptop many times (though usually with 250GB SSDs) and I've never seen that sort of performance drop-off before.

 

[Justinus]

I'm not 100% sure if this is mainly a QLC issue or maybe a DRAM-less SSD issue, but I'm inclined to think it's QLC related because I've sold this model of laptop many times (though usually with 250GB SSDs) and I've never seen that sort of performance drop-off before.

Direct to QLC write speeds are hot garbage, but it's not exclusive to QLC. There are TLC drives with similar issues due to using crappy nand/configurations and bottom of the barrel controllers. I have a TLC ssd that exhausts its couple GB of pSLC write cache and drops to 70-80MB/s direct to TLC writes. This is another reason to thoroughly vet what SSD you are choosing (when making an explicit decision about buying one). There's such a wide, wide variance in direct to TLC and QLC write performance due to several variables

I miss the days when manufacturers explicitly listed the direct to TLC write speed on the spec sheet. Last I saw that was 960 Evo Era. It's becoming less of a thing with dynamic pSLC caching but it's still an important data point for many drives as very few will use all available nand and most still have an upper limit around 30-100GB write cache, so your transfer still could have exhausted the cache.

 

[mikeymikec]

My reason for posting was more about system responsiveness than the transfer speed. For me anyway, a decent SSD's ability to handle multiple workloads is a major reason why I like them so much; it seems to me that they can take what I would normally regard as a hammering and still carry on performing basic tasks at a normal speed.

 

[Justinus]

Ah, well the responsiveness during sustained load would be due to the dramless nature of the SSD while the extremely slow write speed is due to the direct to QLC writes after exhausting the pSLC.

Latency goes through the roof on dramless drives when there's a bunch of stuff going on.

 

[igor_kavinski]

I wonder why some techie doesn't open an online store? They could put filters like QLC/DRAMLESS and sort by TBW rating etc. on SSDs and other important criteria for other hardware. That would really cut down on the time we have to waste researching stuff.

 

[mikeymikec]

I wonder why some techie doesn't open an online store? They could put filters like QLC/DRAMLESS and sort by TBW rating etc. on SSDs and other important criteria for other hardware. That would really cut down on the time we have to waste researching stuff.

In my experience manufacturers can often be obstructive. For example, trying to find out whether a cheap SSD is DRAM-less is tricky, or say Samsung referring to TLC as "3-bit MLC", or say a manufacturer deciding to revise the specs of a product way after release.

I found out that the Samsung 980 is DRAM-less due to a product review. The specs aren't all that forthcoming about it.

With regard to printers I once asked Brother why their printer specs don't explicitly say that a given printer lacks duplex printing/scanning/copying. Their response was that the industry standard is to say what it has. It's also becoming increasingly common for the official specs to be extremely light on details.

I find the strategy overall to be a little strange because for example the Samsung 980 is officially marketed as the "smart choice for NVMe performance" with "breathtaking speed"... so uh, why do we need the PRO? I would have thought that especially a comparison chart should show up some nice red X's next to 'on-board cache', or Brother's duplexing capabilities etc, in order to drive customers to the more expensive products which surely they profit more from. Instead, they often make the user infer from lack of information or use terms like "HMB(Host Memory Buffer)" or SMR.