Kingston NV2 1 TB M.2 NVMe SSD

[lexluthermiester]

I didn't test a review sample, I purchased the drive myself from a German retailer

Oh! Interesting twist. Maybe it was a regional thing for a short time?

 

[kaktus1907]

Can you guys review SSD HIKVision E2000 ‘CRIUS’ M.2 NVME 2TB?
It's a peculiar drive that has a 8 channel Maxio MAP1001A-F2C controller with 2*1GB DRAM modules. No Maxio controller drive in TPU review sheet yet.

 

[W1zzard]

No Maxio controller drive in TPU review sheet yet.

Good suggestion.. I have Colorful CN600 and CN700 here, just no time to review yet

 

[Dinosaullo]

Guys between this SSD and the WD SN570, which one is more worthwhile? By the tests in the review it seemed that the NV2 excelled in everything, but i see many people recommending the WD. Thanks

 

[lexluthermiester]

Guys between this SSD and the WD SN570, which one is more worthwhile? By the tests in the review it seemed that the NV2 excelled in everything, but i see many people recommending the WD. Thanks

The WD performs a little better(1%) but is a bit more expensive($10) depending on where you buy it. Both use TLC NAND from Toshiba, which is good quality. The Kingston has a 3year warranty, where as the WD has a 5year. You're effectively paying $10 for two extra years of warranty. So it comes down to personal preference. I like the Kingston as is seems the better value, but that's just me..

BTW, Welcome to TPU! Don't be a stranger!

 

[CatalinT]

Hello. The review says it has 2 temp sensors. My 1tb smi controller NV2 shows just 1 temp in hwinfo...Is there another way to test ? Thanks

 

[CatalinT]

Can we call this thread dead ? And close it since nobody answers ?

 

[lexluthermiester]

Hello. The review says it has 2 temp sensors. My 1tb smi controller NV2 shows just 1 temp in hwinfo...Is there another way to test ? Thanks

Can we call this thread dead ? And close it since nobody answers ?

The question you're asking is fairly technical. It's not that anyone is being rude, but as a rule, if we don't have an answer, we tend not to chime in. The second sensor may not be visible to the external system if it's just wired to the SSD controller for internal monitoring. There could be a variety of other reasons why that second sensor doesn't show up in any given utility, including that the utility in question may not be programmatically able to see it.

 

[CatalinT]

The question you're asking is fairly technical. It's not that anyone is being rude, but as a rule, if we don't have an answer, we tend not to chime in. The second sensor may not be visible to the external system if it's just wired to the SSD controller for internal monitoring. There could be a variety of other reasons why that second sensor doesn't show up in any given utility, including that the utility in question may not be programmatically able to.

I think the model i have with smi controller is worse than the phison one. Like everybody says one for reviews and one for retail. In my country i cant find the phison one anywere so i guess phison is the better one. Thanks for the answer and sorry for my way of say it above.

 

[Wirko]

I think the model i have with smi controller is worse than the phison one. Like everybody says one for reviews and one for retail. In my country i cant find the phison one anywere so i guess phison is the better one. Thanks for the answer and sorry for my way of say it above.

It's a thread that belongs to a review by TPU, and I don't think those are ever closed.

 

[lexluthermiester]

Like everybody says one for reviews and one for retail.

That really doesn't happen very often. W1zzard stated earlier that he bought his review sample himself at retail, so Kingston didn't send him anything. What's likely going on is that Kingston, like a great many other companies, are feeling the effects of the chip shortage and are doing their best to keep their stock flowing. This likely means sourcing NAND controllers from multiple vendors. It's not an attempt at anything shady, it's just getting the job done.

Does that make sense?

 

[CatalinT]

That really doesn't happen very often. W1zzard stated earlier that he bought his review sample himself at retail, so Kingston didn't send him anything. What's likely going on is that Kingston, like a great many other companies, are feeling the effects of the chip shortage and are doing their best to keep their stock flowing. This likely means sourcing NAND controllers from multiple vendors. It's not an attempt at anything shady, it's just getting the job done.

Does that make sensi

That really doesn't happen very often. W1zzard stated earlier that he bought his review sample himself at retail, so Kingston didn't send him anything. What's likely going on is that Kingston, like a great many other companies, are feeling the effects of the chip shortage and are doing their best to keep their stock flowing. This likely means sourcing NAND controllers from multiple vendors. It's not an attempt at anything shady, it's just getting the job done.

Does that make sense?

I will say nothing is shady until i will see a side by side comparation of this two versions

 

[lexluthermiester]

I will say nothing is shady until i will see a side by side comparation of this two versions

You can make a determination yourself. Run some drive benchmarks and compare to this review. If the results are close enough, there you go.

 

[chrcoluk]

It would be great if you could measure the size of the pseudo-SLC cache once again when the drive is 80% full. Is it as large as physically possible (so ~66 GB) or is it at 20% of 88 GB (so ~18 GB)? Or something else?

I think that there must be some technical reason, not just product segmentation, to limit the pSLC cache and not let it expand over the entire flash memory. Maybe it's because of controller limits (small internal RAM etc.) or related to the size of the host memory buffer. Phison's blog gives a small hint:

Mapping tables has always been my concern on dramless, as I believe where its stored cannot be wear levelled well so DRAM greatly improves endurance, this flaw might be long resolved but not mentioned in reviews hence my decision to keep buying DRAM SSDs.

If HMB has managed to take on 100% DRAM tasks like mapping table caching then review guides should state this.

Replace "five times more expensive" with just "more expensive" and my point stays the same - I did try to compare the NV1 to the most expensive, "obsessive enthusiast" level of devices in practical applications. In the shop I usually buy from, NV2 is, at the moment, the least expensive m.2 NVME drive or very close to being one, about five times cheaper than the most expensive drives of similar capacity. So what would I get for more money, no matter if just a tiny bit or a lot more? No practical improvement in any of the use cases the drive is intended for - it being a consumer device. I did not count SATA drives, a choice I should have mentioned, since I have no use for anything other than m.2 NVME. The car analogy you mentioned holds true for me - I don't care for "fun", whatever it would mean for a mass storage device. I want it to work for the intended purpose, a hybrid Toyota is an infinitely better car than a Maserati if you live in a city.

Don't take it as me trying to be snarky, I'm genuinely trying to understand the rationale behind buying unnecessarily expensive products. Does a typical consumer spend his days transferring large files or installing software all day long? This is a simple drive for people with simple, limited needs. If a person needs unparalleled performance for an income generating machine, there is a whole world of extremely fast and reliable storage. As for QLC not being cheaper than TLC - there's a lot going on with pricing a product. Manufacturing and sales volumes come to mind, TLC is probably being phased out of manufacturing because QLC has more density and therefore allows for bigger margins. Corporations never ever, in any market segment, pass their savings to consumers, "whatever the market will bear" pricing strategy is alive and well. Your problem therefore seems to be with corporate pricing practices, not with products.

Again and honestly not being sarcastic: What do you do with consumer hardware to be limited by it's capabilities? "Objectively better" doesn't mean anything if there's no need for it being better, it just becomes a waste of money a person spends on underutilized capacity.

Yeah same with SMR not been cheaper than CMR.

To prevent the better tech phase out been successful as consumers we need to avoid buying the QLC/SMR stuff.

 

[CatalinT]

Some benhmarks.
System - Msi B450 Tomahawk Max, R3600 stock, 2X8GB 3200 ram.
Nvme disk is working in Pcie 3.0 x4 mode and has some airflow blowing on to it.
View attachment 269674
View attachment 269675 View attachment 269676
View attachment 269677 View attachment 269678
View attachment 269679
View attachment 269680

At least Nand is the same Toshiba 112-layer 3D TLC BiCS5 as in review.

Hello. My drive has 300 gb free of 900gb. It is running in pci 3.0 mode like yours but there is a HUGE difference in write speed...Any ideea why ? Thanks

Hello. My drive has 300 gb free of 900gb. It is running in pci 3.0 mode like yours but there is a HUGE difference in write speed...Any ideea why ? Thanks

Same test but now the drive has 463gb free out of 931gb.This makes a HUGE difference in write speeds.

 

[Wirko]

Hello. My drive has 300 gb free of 900gb. It is running in pci 3.0 mode like yours but there is a HUGE difference in write speed...Any ideea why ? Thanks


Same test but now the drive has 463gb free out of 931gb.This makes a HUGE difference in write speeds.

Interesting. Around 700 MB/s is what W1zzard got once the SLC cache was exhausted. So your previous results might mean that a 2/3 full drive is unable to use the cache, or exhausts it in a very short time (much earlier than ATTO finishes its job), or something like that.

 

[lexluthermiester]

Interesting. Around 700 MB/s is what W1zzard got once the SLC cache was exhausted. So your previous results might mean that a 2/3 full drive is unable to use the cache, or exhausts it in a very short time (much earlier than ATTO finishes its job), or something like that.

That sounds about right. There are limits to how far and effect the SLC cache can be.

 

[DarrenW]

Given that the NAND does seem to be the same, I think the change in controller can be forgiven.

I bought NV2 1TB from AliExpress for just $57.99. The nand is Intel 144L QLC. The BOM of Kingston entry level consumer SSD is not stable.

 

[CatalinT]

So i guess its safe to say that the 2Tb version is the better choice if someone wants better performance.
Its weird that i have to keep my drive half empty if i want better performance....

 

[lexluthermiester]

So i guess its safe to say that the 2Tb version is the better choice if someone wants better performance.
Its weird that i have to keep my drive half empty if i want better performance....

Yes. But you can always partition less space so that the NAND controller always has enough unused space to have an SLC cache. For example: With a 1TB drive, only partition 900GB of it and leave the remainder unallocated. With a 2TB drive, partition 1800GB and leave the rest unallocated. You lose that space but maintain the SLC cache buffer performance. This doesn't work on all drives, but the newer DRAMless models which employ SLC caching schemes, like the Kingston, are highly likely too benefit from such a partitioning tactic.

 

[CatalinT]

I bought NV2 1TB from AliExpress for just $57.99. The nand is Intel 144L QLC. The BOM of Kingston entry level consumer SSD is not stable.
View attachment 272247

It was about time for them to switch to qlc trash...At least the price is good but in my country i paid 65 euro for the tlc toshiba nand version...

 

[W1zzard]

Intel 144L QLC

Oh wow that sucks .. somehow I don't think it'll reach the quoted specs

http://imgur.com/ghjf1tK

Any lawyers here? could be a fun project .. I doubt any reasonable judge will buy the "up to" argument

 

[Wirko]

Yes. But you can always partition less space so that the NAND controller always has enough unused space to have an SLC cache. For example: With a 1TB drive, only partition 900GB of it and leave the remainder unallocated. With a 2TB drive, partition 1800GB and leave the rest unallocated. You lose that space but maintain the SLC cache buffer performance. This doesn't work on all drives, but the newer DRAMless models which employ SLC caching schemes, like the Kingston, are highly likely too benefit from such a partitioning tactic.

But what do you do if SLC caching (apparently) stops working while your SSD still has 300 GB free out of 900 GB total?

I do have a theory, stupid maybe, but here it is. The free space became very fragmented after a lot of writing and erasing of files. The fragmentation may be internal, not even visible to the OS. On the other hand, the pseudo-SLC caching requires a large unbroken chunk of storage space to work. A few GB, maybe tens of GB. That unbroken space didn't exist until @CatalinT removed ~160 GB. On the second benchmarking attempt, it was there, and pSLC could ingest the data very fast.

@W1zzard : testing the behaviour of the pSLC cache once again when the drive is 80% full would be a great addition to your set of SSD tests. It's a situation very much related to real-life use. It may be revealed that SSDs manage their pSLC poorly, meaning that they offer very little caching, or none at all, in this state.

 

[CatalinT]

Write speed with Kingston 20% full and copying from Samsung 970 evo plus 1tb.

 

[lexluthermiester]

But what do you do if SLC caching (apparently) stops working while your SSD still has 300 GB free out of 900 GB total?

Good question. I don't have an answer for you.

I do have a theory, stupid maybe, but here it is. The free space became very fragmented after a lot of writing and erasing of files. The fragmentation may be internal, not even visible to the OS. On the other hand, the pseudo-SLC caching requires a large unbroken chunk of storage space to work. A few GB, maybe tens of GB. That unbroken space didn't exist until @CatalinT removed ~160 GB. On the second benchmarking attempt, it was there, and pSLC could ingest the data very fast.

That is strange. There's got to be a logical answer.