Crucial P2 QLC version tested against TLC version

[CAT-THE-FIFTH]

TH retested the drive now:
https://www.tomshardware.com/uk/feat...ap-downgrade/2

When tasked with light loads, the new QLC variant of the P2 delivers similar performance to its predecessor, scoring just a few points below in the Quick System Drive Benchmark.

However, when tasked with the Full System Drive Benchmark tests, the QLC variant fell flat with roughly 40% lower performance. Latency, the most important metric for an SSD, was also three to four times higher than any other competitor.

SLC cache has increased to 135GB:

One improvement, at least to some degree, is the increased SLC cache capacity. The TLC variant’s SLC cache measured only 24GB, while the QLC replacement’s cache measured roughly 135 GB. Still, while the capacity has increased in an attempt to offset the impact of the much slower QLC flash, the speed of the P2 degraded severely both within and outside of the SLC cache. Some of this goes back to the reduced number of packages, which hampers interleaving, and thus parallelism.

Our TLC-based P2 wrote at roughly 1.85 GBps while the QLC-based P2 wrote at 1.16 GBps before degrading. Once degraded, the TLC variant's sustained write speeds measured roughly 450 MBps, which isn’t great, but acceptable. However, the QLC variant averaged USB 2.0-like speeds of just 40 MBps after the SLC cache was full.

TH,don't like it now:

We no longer recommend Crucial’s P2 for those seeking a value-level SSD. Rather, we would go as far as to place it on our ‘do not buy’ list.

I think it is a little harsh,as if cheap enough it still has its uses. However,it does show TLC drives to appear to be a better balance of performance.

[DanceswithUnix]

I think it is a little harsh,as if cheap enough it still has its uses. However,it does show TLC drives to appear to be a better balance of performance.

More levels per cell will always make for a slower and more delicate drive.

It is a real shame they benchmarked the 500GB drives. Those are no longer the price sweet spot, I've been buying 1TB drives for a while now and am eyeing up my first 2TB drive purchase. I wouldn't expect a 1TB version to actually be a good drive, that's asking too much for such a low end spec, but bigger drives are always faster and hopefully the amount of SLC cache will scale with drive size as well. So a 1TB drive might be more usable. But then I also wonder if that is just simply quite a slow controller.

I've had a few Crucial drives so far, there is a 1TB SATA MX500 in this machine atm, and they have been pretty decent. So disappointing to see such a poor performer.

[Jonatron]

Is there a list of companies who do and don't bait and switch SSDs? I don't want a QLC drive, I can pay extra for TLC.

[DanceswithUnix]

Is there a list of companies who do and don't bait and switch SSDs? I don't want a QLC drive, I can pay extra for TLC.

I'm not sure that would be all that useful. What you really want is avoiding a bait and switch to *bad* drives.

We have been through all this before. The last time it was avoiding TLC drives when MLC ones were so much faster and had better wear rates.

But now? The WD SN850 and Samsung 980 Pro drives are stupidly quick, and are TLC. We are used to TLC and respect it as a choice. So while QLC is currently the slow option, there will come a time when someone will just bundle twice as many channels on a device without skimping on buffer RAM and hey presto a fast QLC drive. I suspect it will need at least a 1TB drive to make QLC viable though. TLC drives suck if you only have 250GB, so it makes sense if QLC sucks at the 500GB point. I note that the Samsung QVO drives are faster at 4TB than the 1TB models, but still not actually quick: https://www.anandtech.com/show/15887...lc-refreshed/3

My last drive purchase was a 1TB SN750 which hit the spot for me for decent performance for a bit more money than one of the cheaper drives.

[CAT-THE-FIFTH]

Samsung and Western Digital so far don't seem to do the stealth changes to NAND. WD did make some stealth changes to the controller in its WD Green series. Maybe also Seagate especially with their Firecuda drives? The latter come with free data recovery too!

More levels per cell will always make for a slower and more delicate drive.

It is a real shame they benchmarked the 500GB drives. Those are no longer the price sweet spot, I've been buying 1TB drives for a while now and am eyeing up my first 2TB drive purchase. I wouldn't expect a 1TB version to actually be a good drive, that's asking too much for such a low end spec, but bigger drives are always faster and hopefully the amount of SLC cache will scale with drive size as well. So a 1TB drive might be more usable. But then I also wonder if that is just simply quite a slow controller.

I've had a few Crucial drives so far, there is a 1TB SATA MX500 in this machine atm, and they have been pretty decent. So disappointing to see such a poor performer.

I have an MX300 and MX500 so far in my rigs,and have purchased the BX500 sometimes for upgrading some older systems.

[Xlucine]

Scaling up from the 500GB drive, then the 270GB cache on the 1TB model makes it a pretty compelling choice. You'd have to really try to exceed that, especially when it costs the same as competing 500GB drives (and you'll find the cache limits on those far easier).

It's dropped to £60 on scan, while the P5 (should be TLC) is £80. It's a good time to buy an NVMe SSD - at least some PC hardware still has a functioning market!

[CAT-THE-FIFTH]

TH also highlighted another issue - drive latency which is not very good either(and some of the copy tests under the SLC cache limit also had slow performance,so perhaps smaller file operations have issues?).

Some feedback from Reddit:
https://www.reddit.com/r/NewMaxx/com...al_p2_now_qlc/

Having used it for a bit now, performance is quite bad, probably worth a buy only if it's at least ~20% cheaper than the SN550 given how much slower it is, especially when writing direct to NAND, which will happen a lot given how small the SLC cache is.
Direct to QLC write speeds on my 1TB version are only about 65MB/s, exponentially slower than what I experience with a WD SN550 1TB.
At 50% filled it only managed to cope with <20GB of writes before it drops down to ~65MB/s. Can't really see much of a use case for the P2, even as a game drive it will bog down with a bigger game update and is unable to keep up with a 1Gbps internet connection.
Did another test too, copying 100GB takes about as long (20 minutes, give or take a few seconds) as a WD Blue 2TB 5400rpm SMR HDD, which in a way is pretty impressive, impressively bad.

Only upside so far is that I got it for cheap on an Amazon sale (1TB was on sale for £71).

Having said that at it closing onto £50 for a 1TB,so it is cheap,so might be expecting too much!

It's a good time to buy an NVMe SSD - at least some PC hardware still has a functioning market!

TBF you can extend that to PSUs,motherboards,cases and literally every component apart from the latest CPU or any kind of GPU.

[Rob_B]

Terrible performance even knowing the limitations. I held off as there are 256GB WD/Unbranded Samsung drives for ~£25 so prices are heading in the right direction.

[watercooled]

I'm not sure that would be all that useful. What you really want is avoiding a bait and switch to *bad* drives.

We have been through all this before. The last time it was avoiding TLC drives when MLC ones were so much faster and had better wear rates.

But now? The WD SN850 and Samsung 980 Pro drives are stupidly quick, and are TLC. We are used to TLC and respect it as a choice. So while QLC is currently the slow option, there will come a time when someone will just bundle twice as many channels on a device without skimping on buffer RAM and hey presto a fast QLC drive. I suspect it will need at least a 1TB drive to make QLC viable though. TLC drives suck if you only have 250GB, so it makes sense if QLC sucks at the 500GB point. I note that the Samsung QVO drives are faster at 4TB than the 1TB models, but still not actually quick: https://www.anandtech.com/show/15887...lc-refreshed/3

My last drive purchase was a 1TB SN750 which hit the spot for me for decent performance for a bit more money than one of the cheaper drives.

Missed this thread initially!

Point regarding MLC > TLC avoidance acknowledged, but 3D TLC is more equivalent to planar MLC in many ways given how much larger the 3D cells are, including performance and endurance. Avoiding planar TLC but being accepting of 3D TLC is not contradictory for that reason. Planar TLC was and remains quite slow, even by SATA standards. 3D MLC is so massively overkill it's barely used even for enterprise drives nowadays. Enterprise-rated 3D TLC such as Micron's FortisMax (vs the consumer FortisFlash) comes with incredible endurance ratings. We're talking around 10,000 P/E cycles.

Increasing the number of NAND planes can help to a degree, but doubling the number of planes only gives you double performance at best, when QLC seems substantially slower than 1/2 TLC's performance. Plus plane sizes often increase as die density increases.

Edit: Found the data I had in mind: https://www.anandtech.com/show/16491...-at-isscc-2021
Compare program throughput of equivalent TLC and QLC.

[DanceswithUnix]

Increasing the number of NAND planes can help to a degree, but doubling the number of planes only gives you double performance at best, when QLC seems substantially slower than 1/2 TLC's performance.

Well yes, the QLC packages seem to be about (slightly worse than) a quarter of the speed of TLC.

That just says to me that they need to hit 16 planes on the package, or alternatively 8 planes and double the channels although M.2 form factor makes adding more nand packages hard.

[watercooled]

That all adds to cost though, and a small cost advantage per stored byte is the only thing QLC has to offer over TLC. You only get a maximum theoretical 33% uplift in usable capacity, although in practice less than that is usable because of over-provisioning requirements, and the raw cost of QLC-capable silicon is higher per cell because of the considerably added complexity of the peripheral circuitry.

Shave down those cost margins any further and the only advantage starts to evaporate. It's what makes me dubious about the prospects of PLC, and even WD seems to be tempering expectations around that. I dread to think what program speeds would be like for PLC if the current trends continue! <10MB/s per device?

[Ulti]

Hopefully I'm not going off topic, but what's a realistic way of replicating these tests with some cheaper or older SSDs I've got to see how they perform.

I can see what programs they're using which hopefully are free and I can just use too. I see they also mention prefilling the drive to 50% capacity - what's an easy way to do this?

[DDY]

Hopefully I'm not going off topic, but what's a realistic way of replicating these tests with some cheaper or older SSDs I've got to see how they perform.

I can see what programs they're using which hopefully are free and I can just use too. I see they also mention prefilling the drive to 50% capacity - what's an easy way to do this?

I use Dummy File Creator:

https://www.mynikko.com/dummy/