NVMe vs M.2

Yeah, QLC is pretty much a joke. QLC drives are barely 10-20% cheaper at most. When the QVO launched it was the same price as the EVO LOL.

So you're saying QLC is a bit like the people who can't grasp that different products exist for different usage scenario's at different price points?

I had to buy 3 NVMe drives earlier this month, I know the anticipated usage of each, my MBPr needs more space (1TB), a general productivity/gaming build (500GB) and a larger cache drive for my UnRAID box (2TB). Two of those are going to see reasonably light usage and the third will see sporadic 50GB dumps to it and docker/VM usage that's light. As mentioned above the 660p and P1 are the same underlying hardware (intel's firmware tends to include fixes not found in other partner firmware based on previous partnerships) and for non heavy IO, they're well suited - your average desktop user using real world office/productivity workloads is simply not going to get out of the SLC buffer. So it comes down to price. If - as you state - the price difference was 10-20% then i'd have just ordered a bunch of 970 Evo's or WD Black's, but a 970 Evo is near enough £100 for 500GB, I paid £49 for a 500GB P1, £85.99 for a 1TB 660p and 149 for a 2TB 660p, now maths was never my strongest subject, but it kind of looks like QLC is half the price - or less as you scale up the capacity. In my usage scenario's for those specific devices, I would gain absolutely nothing from spending a minimum of twice as much other than in irrelevant synthetic benchmarks and longevity that's equally pointless as the drive/devices will be replaced long before it becomes relevant.
 
Well basically every time they increase the bits per cell as a % the gains get smaller.

SLC to MLC = doubling of capacity. 2 bits instead of 1 bits per cell.
MLC to TLC = extra 50%, 1.5x capacity, 3 bits instead of 2 bits per cell.
TLC to QLC = extra 33%, 1.33x capacity, 4 bits instead of 3 bits per cell.

The gains so far on QLC are just for the manufacturers as the price of the drives is way too close to TLC to be considered worth it, each time they increase the bits per cell there is regressions to performance and endurance. First gen TLC proved to be problematic for samsung, the 840s and 840 evo drives had some nasty issues, later solved in newer gen TLC drives.

As far as TLC vs MLC goes, if the drive is big enough then the performance even without SLC cache is still close. Also samsung still has enough confidence in the nand to match the warranty of MLC at 5 years.

QLC is very different tho, warranty is reduced to 3 years so the confidence in the nand isnt there, and the performance is a huge drop off to the point its slower than spindle drives. Maybe they will improve QLC in future with technology improvements but for now I do consider it something to stay away from.

I do wonder how far they will go, will they still keep adding bits even if they only gaining 10% or less?

In terms of me saying the price is close, I was comparing the QVO drives to the EVO drives. 860QVO vs 860 EVO. When I brought my 1tb 860 EVO they both were within a few £ of each other.

The prices you quoted are more sensible, although they still first gen QLC products.
 
Last edited:

I bought one of those Intel drives despite being aware of the limitations of QLC as I only needed something for storing my media on for quicker processing in Lightroom, however after it took nearly two hours just to encrypt the drive ready for initial use, I sent it straight back to the retailer.

My 5 year old 2x512GB RAID0 volume performs vastly better, I would not considering using one of these QLC drives for any purpose.
 
I bought one of those Intel drives despite being aware of the limitations of QLC as I only needed something for storing my media on for quicker processing in Lightroom,

Sounds reasonable so far...

however after it took nearly two hours just to encrypt the drive ready for initial use, I sent it straight back to the retailer.

This is where you went wrong. We went from simple storage to encrypted storage and seemingly ignored the hardware AES256 encryption built in via HDD Password in UEFI as well as Pyrite 1.0/2.0 (no OPAL sadly). That would take 20? seconds to enable on a suitable system. This sounds more like you caused a full write? You know, the kind of thing someone who claims to be aware of QLC’s limitations would avoid?

If for example you dumped data to the drive and then converted it using bit-locker, that would be bad. From personal experience on 8TB drives, you’re looking at 1.5 days per 8TB and that’s with a CPU with AES-NI. On an empty drive writing as you go, you probably won’t notice, but to actually encrypt existing data, that takes time.

QLC is just like any other storage product in the market, understand the usage case and buy appropriately. Get that usage case wrong or do something silly, and you will regret it. For all but a tiny percentage of home users, they’ll never notice or care as it’s fast and cheap.
 
Yeah, QLC is pretty much a joke. QLC drives are barely 10-20% cheaper at most. When the QVO launched it was the same price as the EVO LOL.

QLC is hardly a joke, it's driving down the cost of NAND, alas some QLC drives at the moment aren't competitive against TLC drives, though the Intel 660p 2TB is pretty competitive. The Intel 665p should be even more competitive.

The endurance of QLC has yet to be quantified, though the real endurance will probably be a multitude of what is warranted, i'd also expect the endurance of QLC to increase as the manufacturing process matures.
 
Neither drive in the OP would be my choice, the 1TB Sabrent Rocket is the best 1TB drive for performance per pound currently. It's got a decent controller and TLC NAND.

As for the chap trying to encrypt the 660P, you're doing it wrong. Massive writes to QLC are an issue, you'll exhaust the SLC cache and writes will plummet.

I've recently bought the 2TB 660P, it has 250GB ish of SLC cache when empty and its more than adequate for fast storage. I just wont try and fill it in one go. It was £170 for 2TB which is pretty good value to me, despite the limitations of QLC.
 
Neither drive in the OP would be my choice, the 1TB Sabrent Rocket is the best 1TB drive for performance per pound currently. It's got a decent controller and TLC NAND.

As for the chap trying to encrypt the 660P, you're doing it wrong. Massive writes to QLC are an issue, you'll exhaust the SLC cache and writes will plummet.

I've recently bought the 2TB 660P, it has 250GB ish of SLC cache when empty and its more than adequate for fast storage. I just wont try and fill it in one go. It was £170 for 2TB which is pretty good value to me, despite the limitations of QLC.

Couldn’t agree more, if I wrote it myself - if you scroll up a little, I pretty much did ;)
 
QLC is hardly a joke, it's driving down the cost of NAND, alas some QLC drives at the moment aren't competitive against TLC drives, though the Intel 660p 2TB is pretty competitive. The Intel 665p should be even more competitive.

The endurance of QLC has yet to be quantified, though the real endurance will probably be a multitude of what is warranted, i'd also expect the endurance of QLC to increase as the manufacturing process matures.

The rated endurances of ssd's has never really been tested, so to say it will probably be more is a big statement to make, my 850 pro died at 45 erase cycles and 20TB of a 150TBW rating.

The cost reduction directly from QLC vs TLC is not very much, if a QLC drive is cheaper than that difference it will mean one of the following.

1 - The other drives have bigger profit margins, and a smaller margin is accepted on the QLC products to get market share and consumer acceptance.
2 - There is other factors such as the nand doesnt meet quality expectations of higher tiered products, so the lower spec lowers the price e.g. lower warranty period.
3 - The drives are sold at a loss to gain marketshare.

The intel 660p is only about 20% cheaper than reputable TLC drives, do you consider that significant? Because I dont. No way do I think a 20% saving is worth a 8 fold drop in write performance and at least a 50% drop in endurance. Thats a very poor trade off. Hence me calling it a joke.

Now if QLC was at least half of the price of TLC, then maybe just maybe I would think its an ok product but I still would never buy a first generation QLC drive, completely untested in people's homes.

The sad thing from this is TLC will probably be phased out like MLC has been done, its a classic race to the bottom. If TLC does survive it will probably get its price bumped and earmarked as a premium instead of mainstream product.

QLC manufacturing is about 25% cheaper than TLC assuming equal yields.
 
There is no point going for an NVMe drive over a "normal" SSD if all you are planning on doing is putting games on it. Games will load faster being on SSD compared to HDD for certain but the difference NVMe>SSD is negligible and only becomes apparent in benchmarks.

NVMe is still priced higher than equivalent capacity SSD as well so it totally makes sense to go for SSD as well as being able to keep the 2 ports open, you could effectively run 2 SSDs and split the library across them.
I strongly disagree with that as I had a noticeable improvement with a good NVMe compared to a good SSD. It does depend on the game but its not always unnoticeable or negligible. I wouldn't go back to an SSD after trying a NVMe. Its far more then just benchmarks. Its not so much the loading but the write speeds that made a noticeable difference for me.
 
I strongly disagree with that as I had a noticeable improvement with a good NVMe compared to a good SSD. It does depend on the game but its not always unnoticeable or negligible. I wouldn't go back to an SSD after trying a NVMe. Its far more then just benchmarks. Its not so much the loading but the write speeds that made a noticeable difference for me.

You're disagreeing in the wrong context though. OP has stated this is for a games drive. When (please tell me) do you find yourself doing intensive writes to the drive DURING playing a game?

Yes NVMe will write faster so is a bonus in other scenarios but certainly not for gaming. You can't even claim writing the game to the drive as it downloads is an excuse as virtually no broadband will even reach the speed of a HDD write never mind SSD.
 
The rated endurances of ssd's has never really been tested, so to say it will probably be more is a big statement to make, my 850 pro died at 45 erase cycles and 20TB of a 150TBW rating.

The cost reduction directly from QLC vs TLC is not very much, if a QLC drive is cheaper than that difference it will mean one of the following.

1 - The other drives have bigger profit margins, and a smaller margin is accepted on the QLC products to get market share and consumer acceptance.
2 - There is other factors such as the nand doesnt meet quality expectations of higher tiered products, so the lower spec lowers the price e.g. lower warranty period.
3 - The drives are sold at a loss to gain marketshare.

The intel 660p is only about 20% cheaper than reputable TLC drives, do you consider that significant? Because I dont. No way do I think a 20% saving is worth a 8 fold drop in write performance and at least a 50% drop in endurance. Thats a very poor trade off. Hence me calling it a joke.

Now if QLC was at least half of the price of TLC, then maybe just maybe I would think its an ok product but I still would never buy a first generation QLC drive, completely untested in people's homes.

The sad thing from this is TLC will probably be phased out like MLC has been done, its a classic race to the bottom. If TLC does survive it will probably get its price bumped and earmarked as a premium instead of mainstream product.

QLC manufacturing is about 25% cheaper than TLC assuming equal yields.

Samsung have a slightly skewed reputation when it comes to SSD’s, they usually win the benchmark wars, but the firmware has sometimes required some pretty nasty fixes historically. The 830, 840 and 850 all had various issues and failures on 850 pro’s seem higher than 850 evo’s in my (admittedly small) sample size. Very few drives ever in a home environment get hammered to the point they hit the TBW lifespan, indeed some of the testing done shows generally with older drives, lifespan is well above the rated TBW. Flash lifespan will have been tested to be certified, and OEM’s will leave a safety margin, though obviously failures can happen on any flash drive.

You keep going on about QLC being 20% cheaper, obviously if that were the margin, it’s insignificant, but two of us have purchased 2TB 660p NVMe drives for £149 and £170, I may be loosing my faculties, but where are you seeing branded TLC drives for the 20% increase you keep referencing? That’s £178.80-£204 which at the low end is 1TB 970 EVO Polaris money on a good day, not 2TB.
 
You're disagreeing in the wrong context though. OP has stated this is for a games drive. When (please tell me) do you find yourself doing intensive writes to the drive DURING playing a game?

Yes NVMe will write faster so is a bonus in other scenarios but certainly not for gaming. You can't even claim writing the game to the drive as it downloads is an excuse as virtually no broadband will even reach the speed of a HDD write never mind SSD.
Some games have auto saves and with a SSD you can be running forward only to get a split second pause while it saves. Depending on the game the NVMe removes or vastly reduces those small pauses creating a smoother experience. Same for quick and manual save they are much faster. Strategy games can benefit a lot as well, when I was playing Sim City 4 with large cities the NVMe had a noticeable impact on save speed. Same for Stellaris with auto save set every 5min there was a noticeable benefit on giant maps with large empires at end game. With a NVME I could play without a split second pause every 5 min.

With mmo’s its common for players to run combat logs streaming raid data to a file to go analyse later. When I play Star Trek Online I have all my combat data written into a file both to analyse and to upload for DPS runs.

Not that I do it myself but if you record\stream your gaming sessions the writes of the NVMe reduce the stress on the system and help created a smoother experience.

Lastly some games like MMO’s its common to play in a window or have the game on 1 screen while doing other things on the 2nd screen and NVMe's are more flexible for multitasking.

A lot of it does depend on what games you play. I do have many games where I cannot tell the difference between an SSD or NVMe but that is not true for all games. Its kind of like 16GB+ half the time it wont do anything but at times it makes a large difference.
 
Last edited:
Samsung have a slightly skewed reputation when it comes to SSD’s, they usually win the benchmark wars, but the firmware has sometimes required some pretty nasty fixes historically. The 830, 840 and 850 all had various issues and failures on 850 pro’s seem higher than 850 evo’s in my (admittedly small) sample size. Very few drives ever in a home environment get hammered to the point they hit the TBW lifespan, indeed some of the testing done shows generally with older drives, lifespan is well above the rated TBW. Flash lifespan will have been tested to be certified, and OEM’s will leave a safety margin, though obviously failures can happen on any flash drive.

You keep going on about QLC being 20% cheaper, obviously if that were the margin, it’s insignificant, but two of us have purchased 2TB 660p NVMe drives for £149 and £170, I may be loosing my faculties, but where are you seeing branded TLC drives for the 20% increase you keep referencing? That’s £178.80-£204 which at the low end is 1TB 970 EVO Polaris money on a good day, not 2TB.

Well given I can only find 660p drives for £190-240 might be something to do with it. I am only checking big name retailers. To exclude outliers. Looking at pricespy shows one hit out of over 50 below £190 and its some seller I never heard off. Price history graph seems to indicate it launched cheaper then had a sharp increase. You guys it seems got on a promotional rate. Also you talk about whats practical for dumb end users, I think on that basis its reasonable to compare it to SATA drives given NVME offers no noticeable improvements. I would consider a reputable SATA TLC superior to a QLC NVME anyway.

On samsung's history I do agree, it is shaky, the pattern seems to be first gen on new tech, so 840 and 840 evo were first gen TLC drives for them and 850 series SSD's first gen v-nand. I also observed more hits on google for failed 850 pros vs 850 evos, consider that there is probably at least four times as many evo's sold vs pro's thats telling. Also those very limited tests that were done to test SSD endurance the issue with them was they didnt emulate real world endurance. The tests are better than nothing, but I dont consider them conclusive.

I do agree QLC will improve in future by the way, my issue is the potential displacement of TLC as a mainstream product which I dont want to see and that I think its risky to use first gen QLC products. The samsung QLC pricing is a complete joke, but I will acknowledge I didnt check other vendors and if one was able to find a 2tb ssd for under £150 "and" they not using it as a system drive, it might be worth a punt. The manufacturing saving is 25%, so as I said if the retail margin is bigger then that then something else is also at play. So yeah that £149 buy whilst seemingly not possible now probably was worth a punt for a WORM type use.
 
Last edited:
It's available for £19x from very large and well known retailers, the same places have branded 2TB TLC NVMe for roughly twice the price, not 20% more, please stop referencing fictional numbers, it discredits what is otherwise a reasonably considered post.

First gen is always perceived as higher risk, it often has quirks, but it's usually not so inherently flawed as to make it unusable. The market will do what it has always done, TLC will remain as a enterprise/prosumer product for a generation or three and priced accordingly, QLC will be a consumer product, as capacities increase, so does the SLC cache, on a 2TB drive we already have 280TB of SLC, how many average desktop users do you think regularly dump 280TB to a drive? Big and cheap is what matters in most cases, those that need something else will pay a premium.
 
This is where you went wrong. We went from simple storage to encrypted storage and seemingly ignored the hardware AES256 encryption built in via HDD Password in UEFI as well as Pyrite 1.0/2.0 (no OPAL sadly). That would take 20? seconds to enable on a suitable system. This sounds more like you caused a full write? You know, the kind of thing someone who claims to be aware of QLC’s limitations would avoid?

The encryption is really a moot point as it would perform the same if I did a non encrypted full format. Yes I chose not to use the build in UEFI encryption because it does not meet my security or compatibility requirements.

I was aware of QLC limitations in the general sense, that does not mean I trawled over benchmarks trying to replicate this exact scenario with this exact drive... one does not expect a 2019 SSD to perform worse than a usb pen drive made 10 years ago under any circumstances! Theoretically an empty 1TB QLC drive should have a 500GB TLC cache, however after reaching only 15% encryption the rate plummeted to <50% of the typical TLC drive throughput. By comparison it takes less than 1/4 of the time to encrypt my 1TB EVO 850 array.

You come across as quite defensive of the product, all I said was that I would not consider one based on my experience which you acknowledge as being a valid one.
 
Last edited:
I was aware of QLC limitations in the general sense, that does not mean I trawled over benchmarks trying to replicate this exact scenario with this exact drive...

You can't have been that aware because it sounds like you bought a drive and immediately put it into its worst case scenario (ie: whole drive written over) which is well publicised as QLC's main weakpoint.

Most people will just buy the drive, slowly fill it with media/photos/games/movies and never bump into QLC's issues.
 
Theoretically an empty 1TB QLC drive should have a 500GB TLC cache, however after reaching only 15% encryption the rate plummeted to <50% of the typical TLC drive throughput. By comparison it takes less than 1/4 of the time to encrypt my 1TB EVO 850 array.

QLC uses an SLC cache as best I'm aware, the 660P does anyway:

https://www.anandtech.com/show/13078/the-intel-ssd-660p-ssd-review-qlc-nand-arrives

At best you have 76GB on a 500GB drive. Which almost exactly matches the point where your encryption rate plummeted.
 
You can't have been that aware because it sounds like you bought a drive and immediately put it into its worst case scenario (ie: whole drive written over) which is well publicised as QLC's main weakpoint.

Most people will just buy the drive, slowly fill it with media/photos/games/movies and never bump into QLC's issues.

I didn't expect it to be as badly designed as it was. Using an SLC cache is poor design because the performance of SLC vs MLC is so slight.

But like I said it's moot, I'm not saying don't consider it, I'm saying that it doesn't work for me.

When I can have an almost constant 500MB/s vs the 15% 1.8GB/s and the 85% 200MB/s for the same price I know which I'm going to choose.
 
Last edited:
I didn't expect it to be as badly designed as it was. Using an SLC cache is poor design because the performance of SLC vs MLC is so slight.

But like I said it's moot, I'm not saying don't consider it, I'm saying that it doesn't work for me.

When I can have an almost constant 500MB/s vs the 15% 1.8GB/s and the 85% 200MB/s I know which I'm going to choose.

Im no particular fan, I just have one as it was cheap, but it’s not as badly designed as you say, you just didn’t buy the right drive for your niche use case. For the vast majority QLC is fine.

I don’t think any TLC SSDs use an MLC cache do they? Never seen it mentioned. They all seem to use SLC.
 
Back
Top Bottom