DO NOT buy these NVMEs. Chinese SSD test

Hello friends in one of our videos on the Russian Channel we tested five sad SSD drives from China and got quite interesting results people in the comment section suggested us to also test nvme drives okay then the test subjects have finally arrived from China this time we picked up more capacious 512 gigabyte drives here we have the best selling m.2 nvme ssds on AliExpress kingspac golden for glitter knitak and i300 also I picked up the cheapest and vmesisd I could find in a local store it is called Patriot p300 and we also got an SSD from MSI I know this is not an ad this is the only manufacturer who just sent us their SSD for a test it's more expensive and we will see how it performs and somehow it seems to me that it will not perform much better than the rest you can see the prices for all the ssds on your screen right now basically they are all more or less the same give or take up to 50 US dollars except for MSI station and of course Samsung because it is a two terabyte premium drive that we used for reference let's see how such as this perform in a long distance race how much they heat up whether they need heat sinks how they behave when felt and of course take a look at their insights we will also test a couple of pcie adapters for m.2 from China this is okay today we'll conduct a full-fledged test of popular Chinese and Viennese solid-state drives here is some basic info on pcie to nvme adapters these standards are very similar the adapters are very cheap on AliExpress you might need them if you have run out of m.2 slots on your motherboard first remember about the keys the m.2 slot can have two of them and they determine which type of Drive can be inserted into them SATA and vne or either second nvme Works through four PCI Express Lanes therefore basic adapters with one lane will cut down the performance by a lot thus the top end Samsung 980 Pro connected via one lane performed at 800 megabytes per second for both reading and writing this is very close to the SATA standard whereas the real performance of this drive is about 6 Gigabytes per second via pcie 4.0 and More Than 3 gigabytes per second with pcie 3.0 therefore look for the adapters with 4 or 16 Lanes the ladder will still have only four logical pcies so there will be no real difference between them as you can see they do not cut down the performance compared to the m under two slots on the board since both in fact work through the same pcie Lanes connected to the processor or chipset it is also worth remembering that for ordinary PCS the number of pcie lanes is limited to a couple of dozen and if you connect both a graphics card and a couple of drives there's a chance that the actual operating speeds will be lower than they should be in our test we had to pull out the graphics card and use on-board Graphics moving on what is the main downside of a fast SSD it is Heating in order to process data at the rate of a couple of gigabytes per second fast controllers are needed which are prone to heating a lot but is it critical well let's check it out and let's start with Samsung if you constantly record data on it at a maximum speed in the Ada 64 test with large data chunks then after a few minutes you will see the saw teeth on the graph and the time it took to record two terabytes reach 32 minutes according to our pyrometer the controller has heated up to 80 degrees Celsius and smart utility reports a high temperature after the test now let's equip it with a simple heat sink from AliExpress without a heat pipe a very cheap one in fact just a bar of aluminum and the situation changes drastically the graph becomes smoother and there are no Peaks and the recording time has decreased by 20 to 26 minutes judging by the built-in sensor the drive heated up significantly less to 63 degrees let's check if a more advanced heatsink with a heat pipe by snowman will change the situation no we see the same 26 minutes and a smooth graph so if you have a top and hot nvme SSD it does need a heat sink but not necessarily an advanced linear but what about those basic Chinese drives do they need it all of our six subjects could not reach the temperatures above 68 degrees even in the Ada 64 extreme test on average they heated up to 50 or 55 degrees at the same time the tests showed the absence of thermal throttling although a different problem turned up all the cheap drives showed incorrect temperatures in smart 40 or 42 degrees which never changed only the more expensive MSI drive has a working thermal sensor but this is not particularly important it is not to hide thermal throttling anyway the real temperatures are too low for that and in real workload the temperatures will be even lower and therefore there is no point in wasting money on a heat sink in the case of inexpensive in the review section on AliExpress there are many comments that say that they do buy heat sinks which I believe is rather excessive now let's move on to the tests to begin with here's the data from Crystal disk info everything seems to be normal all ssds except MSI are formally new and have never been turned on before although it is worth mentioning that any drive's mileage can be zeroed out so to say now let's look at the controllers and memory chips all of our ssds don't have a dram buffer which is of course a huge downside however they are using the technology called hmb which uses your system's Ram as a buffer which is supposed to help but of course it doesn't quality nvme drives of such volumes have dram chip of 512 megabytes hmb uses 16 to 64 megabytes of your computer's Ram which is extremely little and the computer memory is physically far away as a result expecting high performance from such drives is a no-go but of course not the buffer alone is responsible for this as for the controllers closer Patriot and golden fur are based on the good old sm-2263xt controller from 2017. at that time it was not bad at all but the question is where are they getting it from today the controller itself has four channels and in the case of a 500 gigabytes SSD all of them are occupied that is we see the maximum possible performance which means that the one terabyte version will not be any faster as for the memory chips here's where things are getting interesting clither comes with a 96 clear SanDisk golden Fern is even worse a 64 layer Micron both of these chips are obsolete they are four or five years old but at least this is TLC and not a more recent but slow and less reliable qlc Patriot turned out to be a surprise though a fairly recent 128 layer High next chip gives it a chance to win the race moving on the next in line are King spec and knee tag equipped with the maxio mep-1202 controller this is a purely Chinese peace attack from the late 2020 which similarly has four channels and is faster than the above Solution by silicon motion with the memory chips everything is more interesting while new attack is equipped with a not so old 100 12 layer Sanders chip in the case of King spec the diagnostic utility just froze trying to determine what it is one of the chips is okay it is a 128 layer ymtc chip but apparently there are problems with the other one given that this is a young Chinese manufacturer we may have come across a defected or bent chip a little spoiler as a test of this particular SSD were most excruciating and finally the most expensive Drive in our collection MSI and as expected it looks fairly decent for its price comes with the Faison E15 controller which is essentially a top-end e19 except with disabled pcie 4.0 support working in tandem with it is the new 176 layer Micron memory with four channels and here we can draw intermediate conclusions that low prices come at a cost these Chinese ssds are so cheap not because premium Brands set high prices to get a bigger margin but because these Asian shops operating in some basements are using cheap or obsolete controllers and memory chips well let's run some test to see how they perform these are the performance results for our ssds showed in Crystal disk Mark although it is more demonstrative we will draw our conclusions using another Benchmark that gives more detailed information we're running at a benchmark on empty drives to see Peak Performance of each of them and here unlike with SATA the spread is much more significant when testing SATA we almost always ran into the bus limits and the max performance varied little 450 to 550 megabytes per second here only need AC and MSI got somewhere close to the theoretical limits of four lanes of pcie 3.0 we got 3.5 gigabytes per second and that only in read mode kinspack shows the worst result here 800 megabytes per second a little more than the limitations of SATA on average the cheap drives showed about 1.5 to 2 gigabytes per second the difference with SATA drive is quite noticeable it would seem but as you may have guessed there are nuances to determine those nuances let's run a large data chunks test in Ada 64. yes it's still synthetic it's hard to imagine someone writing 500 gigabytes of data at a time but this test is indicative of how the controller handles data packing and its maximum performance in continuous workload and the results turned out to be extremely interesting firstly the general principle of operation of all tested bufferless nvme ssds is similar to the side ssds but differs in one important detail more on that later secondly in some cases SATA drives turned out to be even faster let's try to figure out why all the drives start off quite well and write down about 15 to 25 percent of the volume quite fast knee tag and MSI are in the lead 2 gigabytes per second and this is understandable these Solutions have newer controllers and the memory chips used in them are not complete junk either clicks are Patriot and golden first stick together shown about 1300 to 1500 megabytes per second that is caused by the albeit good but still obsolete controller and memory chips but after continuous recording of about 100 gigabytes of data a steep drop down happens as the leader does not change the MSI Drive yields 850 megabytes per second The Outsider doesn't change either case back is at the bottom with 60 megabytes per second the rest are around 250 to 300 megabytes per second and this behavior is close to what we saw in the sad SSD test and the reason for this is simple SLC caching the principle is simple modern memory chips are based on Triple level cells or quad level cells or TLC and qlc which allows for a more efficient data storage you can store three or four bits of information in one cell rather than just one bit however writing data in three or four bit mode is quite slow therefore manufacturers of SSD controllers made a trick initially the data is written in fast single bit mode for as long as possible into these so-called single level cell cache or SLC in this mode the drive can easily reach the performance of several gigabytes per second which we can see in various benchmarks such as Crystal diskmark as SSD or Auto and at the beginning of the ada64 test but the problem is that in this mode the volume of the drive drops three or four times that is for a 500 gigabyte Drive you'll have like 100 or 150 gigabytes which can be written at such high speeds and what happens when the storage space Runs Out in single bit mode but you still need to keep writing the data down onto that drive that's right the controller has to pack the data recorded in a single bit cells and rearrange it into TLC and qlc while simultaneously fill in the free each cells with new data that's still coming in this task is challenging even for top-end controllers therefore the performance drops down significantly and stays that way until the very end of the real drive volume and this is exactly what we saw in the settings detest but in the case of nvme everything is more interesting since all the test subjects come with TLC memory it would make sense if the maximum write speed maintains up to a third of the drive volume however all of them drop the speed much earlier at about 20 percent of the volume here we can see how the smart and vme controllers are trying to soften the blur that succumb if they have freestyles in stock maintaining reasonable speed in TLC mode is possible for some time that is until you finally have to get to squeezing that SLC cache and only when the drive runs out of this extra stalker cells we see a second drop in performance which is extremely huge the controller has to squeeze the single bit cache and write down until C mode only at the second third of the volume which causes this catastrophic drop down that sad ssds had on one third this approach makes a lot of sense in the long run a good analogy here is Runners it is more profitable not to run out of breath at the very beginning but to start slowly save your breath and eventually get a higher average speed but in our case the runners are obviously not of the world class the first drop down made them as slow as the cheapest SATA ssds are and the second basically turn them into classic hard drives but there are two exceptions these are kinksback which took as much as two hours to record 500 gigabytes at a speed of 60 megabytes per second and MSI the first drop down for which was the least and in the second it was clearly seen that the controller still manages to pack the SLC cells on time synthetic benchmarks only confront our theoretical conclusions ssds have made a lot of progress in recent years and if a good five-year-old processor is still able to run almost everything without issues a five-year-old SSD controller is not a good pick actually most likely this is exactly why they're sold so cheaply in China along with old memory chevs and thus make their way into small local manufacturers of solid state drives synthetic tests are of course interesting but in practice hardly anyone would record 500 gigabytes at once let's take a look at a more realistic work scenario when such a drive is felt somewhere by half say about 300 gigabytes a gigabyte and you want to throw in 100 gigabytes more for example in Adobe Premiere project or a triple A game we will copy that much data from a two terabyte Samsung 980 Pro in order to make sure that this is not a bottleneck half of the SSD volume is already filled which means that we have 50 gigabytes administ for the SLC cash and this is clearly seen in practice in the case of glycer patriot and knee tag they wrote down some part of the volume into SLC cache at a good speed of 1.5 gigabytes per second and then altogether they dropped to 200 megabytes per second if you are downloading a modern video game you'll hardly notice this difference even if your connection is extremely fast but if you plan to work for example with 4k footages on a regular basis then such performance is only slightly better than that of regular hdds which can hardly be called adequate especially considering that knee tag hit more than 3 gigabytes per second at the beginning now let's talk a bit more about the ones standing out these are golden fur MSI and kingsback the leader was already obvious the MSI Drive maintained more than 800 megabytes per second until very end even when copying a 100 gigabyte file MSI as well as golden fur uses a dynamic SLC cache in other words the controller controls the volume depending on the free space in theory in the case of quality ssds this allows for a better long run performance at a cost of the drop down that comes faster this is exactly the case with MSI but not with golden fur the 150 megabytes that it hits is adequate for a cheap SATA solution but not for nvme kingspac which is my favorite now as usual showed the worst results imaginable 2.2 megabytes per second that is its final performance it tried very hard and sometimes it even managed to get back to 60 megabytes per second and in this case it is not the controller's fault knee tag has a similar controller and works much better apparently the ymtc memory chips in this case is what was left after binning it is quite obvious that if you use such a drive as a system drive a drop down to 2 megabytes per second will at best cause your operating system to work slowly and at worst to the blue screen of death at the end we will once again run test in Auto on drives selled by 400 gigabytes which is 80 and here again no surprises there's practically no difference with empty ones that is except for kingspac and this is expected all popular benchmarks usually write down not smaller than a couple of gigabytes which always fit into the SLC cache given the fact that Windows behaves the same way even with drives that are fairly full there will be no performance issues well with the exception of kingsback of course which even at 80 of the volume filled managed to drop down to 150 megabytes per second with large files into 20 megabytes with small ones and I think it is not necessary to explain that this will lead to a serious slowdown of your system at any attempt to record just anything so what's the conclusion a miracle did not happen saving a bug by picking up a cheap Chinese this is the only express is not a wise decision all of those drives are made according to a simple principle cut all the corners possible as a result in the best case scenario we will have relatively new base level controllers but for the most part solutions from five years ago with absolute memory chips and with a minimum of diversity there were only two types of controllers in the four ssds tested this is the main problem that is clearly visible in the case of kingspac even an official poster the company shows that the performance may vary by three times which clearly indicates a lottery in the components used and the manufacturer clearly doesn't bother to test the result in Frankenstein's as a result you may well receive a drive that will record data slower than the cheapest USB stack and this applies not only to King's back as I said above all no-name drives from China are similar so you can expect floppy drive performance in knee tag as well as golden fur and in general in any cheap SSD and this is not the worst here's a question for you where do they get these old memory chips and controllers best case scenario they buy up leftovers from manufacturers that need to get rid of obsolete components that no self-respecting brand will ever use anymore and the worst case demonstration samples or rejected chips as in king spec the status they are sold as rejected silicon Wafers that have not passed all necessary tests but some of the nand chips on them are possibly still operable small companies are happy to cut them into chips test them somehow package them and sell them cheaply to local SSD makers as a result you may well come across a drive that has one good memory chip and one not very good with deplorable performance and a great chance to lose your data and this is not an assumption go ahead and scroll through the reviews on King's pack and golden firm a serious performance variation is clearly noticeable and there are also complaints that the drive does not work out of the box or the windows stopped loading on it after a while while buying CPUs from AliExpress is indeed a good option these processors are definitely not counterfeit and sometimes are twice as cheap as in the local retail this is not the case for ssds and it would make sense if similar Solutions in local stores cost twice as much but no the Patriot Drive that we picked up at the local store costs about three dollars more offering the same old controller but new memory chips and the chance of that loss is significantly lower here and most importantly it comes with a 1 one year warranty from the retailer anyway here's the rating of the ssds tested in this video among those that I ordered from China glycera turned out to be the best it did not disappoint in the test with large data chunks and maintain high speed and copying without dropping below 250 megabytes per second does it make sense to risk your nerves and data in order to save five bucks I said just you answer this question in the comments in the description you will not see the links for the ssds tested we wouldn't want you to buy such garbage this was MK my name is Mikhail crushan I'm waiting for you in the comments bye

Channel: My Computer

Views: 244,659

Keywords: ssd, nvme, m.2, test ssd, my computer, KingSpec, Netac, ssd aliexpress, goldenfir, kllisre, What SSD

Id: RCtN6O-vXhE

Length: 19min 43sec (1183 seconds)

Published: Fri Jan 27 2023