Lexar NQ780 4 TB

[W1zzard]

The Lexar NQ780 4 TB offers plenty of storage for all your games and videos. It is also priced very reasonably, at just $240, which means one TB of storage will cost you $60. Results from our testing show that performance is pretty good, too, as long as you don't exhaust the SLC cache.

Show full review

 

[FreedomEclipse]

Should see if you can get a 4TB FA200 from Acer to go head to head with this lexar.

 

[_roman_]

No DRAM Cache. Worse as expected on the other test.

Of course SSDs are faster. Faster as IDE HDD?

Moving 100GB long term install which is not much will be a pain with that drive. I would not even want to image backup a 2TB NVME with ~300MB/s.
I see the 394MB/s as best case scenario. Best nvme slot on the mainboard, no other influences and no long term firmware or drive issues

 

[Wooden Law - Black]

No DRAM Cache. Worse as expected on the other test.

The big problem here is the use of old QLC flash rather than the new 232L chips (or, better yet, 276L like Micron N69R). But the choice of using a DRAM-based controller without the DRAM is hilarious and doesn't improve the situation.

 

[tsunami2311]

meh not price resonable imo, $200+ for 4 tb is bad

 

[W1zzard]

meh not price resonable imo, $200+ for 4 tb is bad

https://www.newegg.com/p/pl?N=100011693%20600545605&Order=1

Not exactly impressive options for less than $200

 

[zmeul]

as far as I understand the controller on this can use DRAM cache, and yet none can be found on this unit
nah, I'm good

 

[Chrispy_]

Booo, it's QLC at TLC prices.

QLC is fine for a games library but c'mon, where's our goddamn discount, there's 33% more data per chip so where's our 33% discount, baby?

I'm not going to suffer all the performance/endurance/power downsides of QLC for nothing y'know. Only a moron or the ignorant masses would do that.

 

[QUANTUMPHYSICS]

I'm disappointed SSD hasn't come further by now.
Yes I'm happy that 1TB and 2TB SSD is cheap(er). I'm happy with the 4TB pricing - some available under $200.
But where are the 8TB and 10 TB modules?
By now they should be under $500.
I will continue to wait as prices drop. A 2TB OS drive and a 4TB game drive are enough to get me by for now.

 

[lexluthermiester]

For $240 this isn't terrible. The fact that it's QLC rule this drive out as a boot/OS drive, but as an additional mass storage type drive, not bad.

I'm disappointed SSD hasn't come further by now.

We need a break-through in either NAND formulation or in some other form of persistent/static flash memory.

 

[Scour]

144L Intel QLC? Remaining stock?

I guess for a good price and not high-performance actions it will find it´s customers

In Germany the NQ790 4TB starts at 210€, the 1st TLC-SSDs start at 220€

 

[boppingXQWB]

Is this Intel or Longsys? Multiple places in the review where it says Intel NAND. Thought maybe Longsys was an Intel brand-name for NAND chips but after a quick web search, it doesn't seem like they are related.

 

[W1zzard]

Is this Intel or Longsys? Multiple places in the review where it says Intel NAND. Thought maybe Longsys was an Intel brand-name for NAND chips but after a quick web search, it doesn't seem like they are related.

Longsys buys the NAND dies, probably untested. Then they test/sort and put them into their own package (the black plastic substance that you see), and of course they print "Longsys" on them

 

[chrcoluk]

@W1zzard page 6, can delete this post after you fix. GB instead of MB

This is expected for a QLC SSD. Filling the whole capacity completes at 394 GB/s on average, which is a very low result when compared to other SSDs with TLC

Also on summary I think you meant 870 QVO?

When comparing against other QLC SSDs like the Kingston NV3, Crucial P1 and Samsung 870 EVO

 

[EL_H]

@W1zzard page 6, can delete this post after you fix. GB instead of MB

Also, one should be ~25% capacity.
"These speeds are sustained until 928 GB have been written, which means the drive will fill 68% of its capacity in SLC mode first."
@W1zzard

 

[W1zzard]

"These speeds are sustained until 928 GB have been written, which means the drive will fill 68% of its capacity in SLC mode first."

928 GB in SLC mode = 928 GB x 3 actual usage, which is 68%. This keeps coming up, will make it more explicit in the future.

 

[lexluthermiester]

928 GB in SLC mode = 928 GB x 3 actual usage, which is 68%. This keeps coming up, will make it more explicit in the future.

What I'd like to see is the use of MLC mode as a cache. It'd still be fast enough but would offer a much larger write buffer.

 

[Chrispy_]

What I'd like to see is the use of MLC mode as a cache. It'd still be fast enough but would offer a much larger write buffer.

I've also wondered why controllers don't tier the cache through all the different modes, but I guess it would add controller/firmware complexity, increase write amplification on the already-pitiful number of NAND PE cycles and pSLC is probably enough for 99.x of customers. In the end, cheap/simpler/reliable usually dominates the market while expensive/complex/high-maintenance tech gets relegated to extremely niche use cases.

Even though it's easy enough to exhaust the cache in a synthetic test, very few real-world scenarios exist where you would be streaming data to an SSD at the maximum speed of pSLC for uninterrupted chunks of data that are larger than one third (TLC) or one quarter (QLC) the total drive capacity. If that's your workflow you shouldn't be using a single SSD in the first place.

 

[lexluthermiester]

but I guess it would add controller/firmware complexity, increase write amplification on the already-pitiful number of NAND PE cycles and pSLC is probably enough for 99.x of customers.

Good points.

Taking my earlier thought to the next level, I'd love to see drive makers offer a user configurable option to control what cycling is used and to what degree. For example, If I wanted to limit the first 1/4 of the drive to MLC(for a total of 512GB in the case of this drive) and the rest TLC(for a total of 2304GB), OR set the first 1/2 for MLC mode(1TB) and the second 1/2 to QLC(2TB). This kind of write control scheme would be very nice, very useful and is easily done.

 

[Chrispy_]

Good points.

Taking my earlier thought to the next level, I'd love to see drive makers offer a user configurable option to control what cycling is used and to what degree. For example, If I wanted to limit the first 1/4 of the drive to MLC(for a total of 512GB in the case of this drive) and the rest TLC(for a total of 2304GB), OR set the first 1/2 for MLC mode(1TB) and the second 1/2 to QLC(2TB). This kind of write control scheme would be very nice, very useful and is easily done.

I think they're more dynamic than that, so fixing the behaviour to specific capacities of pSLC and pMLC etc won't actually increase overall performance in any real-world situations. Drives will use pSLC until they're out of space, at which point they drop down to native QLC NAND writes, but even a drop in sustained writes to just 90% of max speed is enough to let the drive start writing some of that earlier pSLC data to QLC. Conversion to QLC happens any time the drive is not 100% writing at max NAND transfer speed.

You'd need to do timed runs, but given that the underlying bottleneck is always the write speed to QLC NAND, I suspect reaching the end result will always be fastest with the minimum number of re-writes.

By that, I mean that the total time to fill the drive entirely ultimately comes down to the slow native QLC write speed, and every addition re-write of data converting from SLC to MLC to TLC is an additional process that delays that total time to fill the drive. So pSLC all the way until it crawls is probably faster for those heavy sustained writes than anything else, even if it's a jarring wall that your SSD runs into 1/4 of the way through the process.

On the other hand, being able to choose whether I have a 4TB QLC drive, a 3TB TLC drive, a 2TB MLC drive, or a 1TB SLC drive would be absolutely awesome.

 

[boppingXQWB]

Longsys buys the NAND dies, probably untested. Then they test/sort and put them into their own package (the black plastic substance that you see), and of course they print "Longsys" on them

Thanks for the explanation!

 

[Wooden Law - Black]

I've also wondered why controllers don't tier the cache through all the different modes, but I guess it would add controller/firmware complexity, increase write amplification on the already-pitiful number of NAND PE cycles and pSLC is probably enough for 99.x of customers.

We have this mode now: Micron advanced write technology (AWT) on the 2600. For the 2600 it breaks the cache in pSLC and pTLC, but I think that we can see pMLC, pQLC and pPLC for future SSDs reading the patent.

According to the patent, the choice for the usage of a specific mode is based on the frequency access of blocks which could be a capacity threshold of the SSD, read counts, error rate of the wordline, etc.

 

[Chrispy_]

We have this mode now: Micron advanced write technology (AWT) on the 2600. For the 2600 it breaks the cache in pSLC and pTLC, but I think that we can see pMLC, pQLC and pPLC for future SSDs reading the patent.

According to the patent, the choice for the usage of a specific mode is based on the frequency access of blocks which could be a capacity threshold of the SSD, read counts, error rate of the wordline, etc.

Oh wow, that's brand new, as in - it was launched hours after this NQ780 review was published!

The only review I've read so far is https://www.thessdreview.com/our-reviews/nvme/micron-2600-gen4-dramless-ssd-review/5/ and they don't test the headline feature.
/facepalm.

I'll look forward to @W1zzard 's review. This AWT feature should show up clear as day in the usual write-intensive graph - and the area under the graph is what matters. The pTLC cache on the 2600 is touted by Micron as being bigger than the pSLC cache, and it implies they're fixed ratios depending on the available free space, so for an empty 2TB drive, rather than 500GB of pSLC it would be (at most) 375GB of pSLC and 375GB of pTLC.

I'm trying to think what data transfer size would benefit from AWT in this scenario. If I assume a best-case 1:1 ratio of pSLC and pTLC caches, and use 6GB/s, 800MB/s and 350MB/s as numbers for pSLC, pTLC, and QLC write speeds respectively, some quick math gets us this:

• 0-500GB is obviously fastest on the traditional pSLC drive, because it's all writing at 6GB/s pSLC speeds for the entire amount.
• 600GB would write in 368s on the pSLC drive and 344s on the AWT drive.
• 700GB would write in 655s on the pSLC drive and 468s on the AWT drive.
• 750GB+ both drives are back at native QLC write speeds.

So it looks like AWT just trades performance between 375GB-575GB written to deliver better performance between 575GB-750GB written. The larger the pTLC cache is in relation to the pSLC cache, the smaller the margin of benefit becomes compared to a pSLC-only drive.

 

[EL_H]

928 GB in SLC mode = 928 GB x 3 actual usage, which is 68%. This keeps coming up, will make it more explicit in the future.

Sorry, my bad. I was used to reading these figures as the amount of user space used as SLC cache. I believe this 928 GB is in binary; judging from the graph, it ends around 3725 GB or about 25% just as expected for QLC.

 

[tsunami2311]

https://www.newegg.com/p/pl?N=100011693%20600545605&Order=1

Not exactly impressive options for less than $200

impressive or not 2tb should be well under $100 at this point and 4tb should no more then $150 and even those prices high, then all NAND make purposely kick price high sorta like certian Nvidia jacket man