WEBVTT

00:00:00.960 --> 00:00:07.279
as ssds or solid-state drives have evolved some of the components have

00:00:04.799 --> 00:00:11.679
become much more sophisticated like the controllers on board that have gone from

00:00:09.360 --> 00:00:16.560
rudimentary single core affairs to multi-core processors with huge amounts

00:00:14.320 --> 00:00:21.600
of RAM and complex algorithms built into their firmware

00:00:17.840 --> 00:00:23.760
but the foundation of nearly every SSD

00:00:21.600 --> 00:00:30.480
the nand flash memory that actually stores your data has actually gotten

00:00:26.720 --> 00:00:32.880
progressively worse in some big ways

00:00:30.480 --> 00:00:38.640
we went from high speed and super reliable slc flash which draws only a

00:00:35.920 --> 00:00:44.960
single one or zero to each cell to dual air mlc to triple layer tlc to finally

00:00:42.840 --> 00:00:52.800
today this is the Intel 660p the first consumer SSD

00:00:49.520 --> 00:00:55.920
with qlc flash which is notable for its

00:00:52.800 --> 00:00:59.199
ability to store 4 bits per cell that

00:00:55.920 --> 00:01:02.160
means 16 separate voltage levels

00:00:59.199 --> 00:01:07.360
this gives it fantastic affordability especially for an NVMe SSD but my mama

00:01:04.960 --> 00:01:09.840
always told me if it seems too good to be true

00:01:08.320 --> 00:01:13.430
it probably is so let's take a look at the pros and

00:01:12.479 --> 00:01:21.439
cons

00:01:21.439 --> 00:01:30.960
for today only you can pick up pc building simulator on chrono.gg for 15

00:01:27.759 --> 00:01:32.799
usd so check it and all of chrono.gg's

00:01:30.960 --> 00:01:36.799
games out at the link in the video description

00:01:34.159 --> 00:01:43.439
let's start with the bad stuff first up is that qlc nand has lower

00:01:39.840 --> 00:01:45.840
endurance than tlc which means that all

00:01:43.439 --> 00:01:48.560
other things being equal it doesn't last for as long

00:01:47.280 --> 00:01:52.880
why well i'm glad you asked because it's

00:01:50.560 --> 00:01:57.680
science time with pictures this is really important

00:01:54.799 --> 00:02:01.280
every time a cell gets written to a voltage pulse is sent through the

00:01:59.439 --> 00:02:05.439
control gate which creates an electric field which agitates the electrons

00:02:03.520 --> 00:02:12.480
causing them to move through the silicon dioxide layer towards the floating gate

00:02:08.319 --> 00:02:14.480
that their layer wears out a tiny bit

00:02:12.480 --> 00:02:19.760
every time a program erase command is sent to the cell causing some of these

00:02:16.640 --> 00:02:22.560
shifting electrons to get stuck inside

00:02:19.760 --> 00:02:29.360
of it which means that it will have and more importantly continue accumulating a

00:02:25.440 --> 00:02:31.599
progressively more negative charge

00:02:29.360 --> 00:02:36.319
now this gets compensated for by applying ever slightly higher positive

00:02:34.560 --> 00:02:41.200
voltage to the cell to get it to the desired voltage state

00:02:38.400 --> 00:02:45.680
now at some point the voltage levels start bordering those required by the

00:02:43.680 --> 00:02:50.400
adjacent states to the point where it takes too long to distinguish what's

00:02:47.920 --> 00:02:54.239
what and that block will get taken out behind the barn old yeller style

00:02:53.360 --> 00:03:00.400
now the fewer layers that there are inside

00:02:56.640 --> 00:03:02.879
of a cell the wider the spare voltage is

00:03:00.400 --> 00:03:08.720
in between the states so then as you can imagine with qlc the point of cannot

00:03:06.400 --> 00:03:14.000
deal with this anymore gg cell comes sooner than with tlc and

00:03:11.040 --> 00:03:18.720
especially mlc or slc much sooner

00:03:15.519 --> 00:03:19.599
second bad stuffs qlc is also slower

00:03:18.720 --> 00:03:26.000
which like wait a second why why is that denser processor

00:03:23.760 --> 00:03:30.319
transistors are better increasing the aerial density of a hard drive platter

00:03:27.840 --> 00:03:35.200
is better shouldn't more data density in nand flash be better

00:03:32.799 --> 00:03:41.040
unfortunately no it's actually the opposite you see with hard drives

00:03:38.000 --> 00:03:43.760
cramming more data into the same surface

00:03:41.040 --> 00:03:50.319
area increases the read and the write speeds because the platter rotates at a

00:03:46.640 --> 00:03:52.720
constant speed usually 5400 to 7200 RPM

00:03:50.319 --> 00:03:56.879
or so which means that the more densely the bits are packed the more of them

00:03:54.959 --> 00:04:03.840
pass under the head in a given amount of time more bits is more data so denser in this

00:04:01.120 --> 00:04:07.280
case is better that is as long as you can keep your read error rate under

00:04:05.599 --> 00:04:13.120
control but that's a whole separate discussion back to ssds for now when an

00:04:10.400 --> 00:04:17.919
SSD cell is accessed a distinction between the multiple voltage levels

00:04:15.280 --> 00:04:22.639
needs to be made and the number of states that you need to sift through

00:04:20.160 --> 00:04:27.600
goes up exponentially with the number of layers that it holds

00:04:24.560 --> 00:04:29.840
so the more layers the more states and

00:04:27.600 --> 00:04:34.320
the longer it takes to get a reading for example it takes 25 microseconds to read

00:04:32.880 --> 00:04:42.080
for slc 50 for mlc 75 for tlc and 100 for qlc

00:04:40.000 --> 00:04:45.759
and it's the same story with rights except the performance drop-off is even

00:04:44.320 --> 00:04:50.560
worse finally big problem number three

00:04:48.960 --> 00:04:57.360
here we're getting into this drive specifically the advertised write speeds

00:04:54.320 --> 00:05:00.000
are kind of hacks because there isn't

00:04:57.360 --> 00:05:05.280
really a right way to do it you see this drive treats part of its qlc cells

00:05:04.320 --> 00:05:12.320
as slc flash as a kind of cache so remember slc

00:05:09.759 --> 00:05:17.600
that's the fastest kind and the exact size of this cache scales depending on

00:05:15.199 --> 00:05:20.800
how much you've filled up your drive so so then

00:05:19.520 --> 00:05:26.400
by this point in the video you're probably thinking all right well then

00:05:23.120 --> 00:05:29.840
qlc has a severe case of the no buenos

00:05:26.400 --> 00:05:31.840
and this here 660p drive is a nomi gusta

00:05:29.840 --> 00:05:34.880
but uh no

00:05:33.199 --> 00:05:40.479
everything that we just said was in theory now it's time for a little bit of

00:05:37.600 --> 00:05:43.520
reality so to put those hacks right numbers to the test and see just how bad

00:05:42.560 --> 00:05:49.680
the whole slc cache with slower qlcnn behind it

00:05:47.360 --> 00:05:53.759
deal affects things we hit this drive with a full range right through hd tune

00:05:51.919 --> 00:06:00.080
pro and guess what halfway through the performance

00:05:56.160 --> 00:06:02.319
plummeted to below that of a hard drive

00:06:00.080 --> 00:06:08.639
and then stayed there for consecutive runs after a brief spike on each run

00:06:05.360 --> 00:06:10.400
but then we observed this behavior only

00:06:08.639 --> 00:06:14.800
after more than half of the drive's capacity was written to something that

00:06:12.639 --> 00:06:19.199
was supposed to happen earlier on in the test which got us thinking

00:06:16.880 --> 00:06:24.400
maybe the caching algorithm was actually working in the background shuffling data

00:06:21.600 --> 00:06:28.240
over to the slower qlc cells so we tried running the cache flusher utility from

00:06:26.240 --> 00:06:32.639
Intel's SSD toolbox during the test and observed its

00:06:30.319 --> 00:06:35.600
progress bar actually going backwards while our benchmark was showing

00:06:34.000 --> 00:06:40.000
throughput spikes when we ran the cash flush after doing a

00:06:38.080 --> 00:06:44.240
full capacity right it took about 20 minutes total and then restored the

00:06:41.759 --> 00:06:50.080
drive to its full write speed now when we tested Intel's own higher

00:06:46.319 --> 00:06:52.240
tier and noticeably more expensive 760p

00:06:50.080 --> 00:06:56.800
we also saw right performance drop during the first run after it ran out of

00:06:54.400 --> 00:07:01.199
cash and dropped further on consecutive runs but

00:06:58.400 --> 00:07:04.880
as you'd expect it remained much faster in the same scenarios than its lesser

00:07:03.120 --> 00:07:10.880
sibling okay so that's synthetic tests but what

00:07:06.880 --> 00:07:14.400
about a real life but edge case

00:07:10.880 --> 00:07:16.400
let's copy a 250 gig steam folder over

00:07:14.400 --> 00:07:22.880
to the 660p there we go same thing it starts off

00:07:19.360 --> 00:07:24.639
fast then it dips dips dips and plummets

00:07:22.880 --> 00:07:28.639
like occ's stock before their bankruptcy in 2013.

00:07:26.319 --> 00:07:32.000
when compared to a similarly priced 840 evo

00:07:30.160 --> 00:07:36.400
that thing managed to maintain a steady 485 megabytes per second

00:07:34.560 --> 00:07:41.840
and completed the copy in only eight minutes and 45 seconds this thing took a

00:07:39.120 --> 00:07:45.520
staggering 23 and a half minutes remember when i said qlc is slow without

00:07:44.479 --> 00:07:49.919
a cache well behold the worst case scenario

00:07:47.680 --> 00:07:55.199
slower than a hard drive meanwhile the tlc equipped 760p after starting off

00:07:52.639 --> 00:07:58.879
strong dropped to a more stable 560 megabytes per second and then maintained

00:07:57.360 --> 00:08:03.199
that completing the same transfer in 8 minutes and 20 seconds now let's get

00:08:01.759 --> 00:08:06.000
more realistic all of the theory crafting we've done so

00:08:05.039 --> 00:08:10.800
far failed to have any tangible effect on

00:08:09.120 --> 00:08:15.199
our other tests game load times were about the same

00:08:12.560 --> 00:08:18.960
between all three drives and when we ran the pc mark 8 storage subsystem bench we

00:08:17.520 --> 00:08:25.039
had to double check that we didn't accidentally test the same drive twice

00:08:21.919 --> 00:08:27.360
so a typical LTT video takes up about 24

00:08:25.039 --> 00:08:32.159
gigs for seven and a half minutes which takes about 20 minutes to render so just

00:08:29.680 --> 00:08:37.200
over one Gigabyte per second of right so we wouldn't have any performance drop

00:08:34.959 --> 00:08:42.240
unless we were filling up our 660p to the brim leaving no room for slc caching

00:08:40.399 --> 00:08:46.800
which by the way we don't recommend filling up any SSD

00:08:44.720 --> 00:08:54.080
also keep in mind that during normal usage your SSD is idle the vast majority

00:08:51.360 --> 00:08:57.279
of the time so that cash flush algorithm will have all the time it needs to

00:08:55.760 --> 00:09:03.279
quietly restore your SSD in the background so now that we've covered the

00:08:59.760 --> 00:09:05.360
worst case for qlc performance and how

00:09:03.279 --> 00:09:09.200
the slc cache can give it boosts of higher performance

00:09:07.120 --> 00:09:15.200
let's talk longevity when compared to Intel's own tlc 760p

00:09:12.560 --> 00:09:20.240
drive this thing is rated for one third the longevity

00:09:17.600 --> 00:09:24.959
but looking closer this actually assumes about a hundred gigabytes of writes per

00:09:22.160 --> 00:09:30.000
day that is basically equivalent to rendering out a couple of LTT videos and

00:09:27.600 --> 00:09:33.519
installing doom on your computer every day

00:09:31.440 --> 00:09:38.000
it's not a realistic use case for the average consumer

00:09:35.600 --> 00:09:42.480
so to sum things up if you were to install a 660p SSD in your system

00:09:41.040 --> 00:09:47.440
unless you're using it for something that it wasn't designed for like as a

00:09:44.720 --> 00:09:51.200
cache for your nas or your hard drive all things considered you would save a

00:09:49.440 --> 00:09:53.760
few bucks since this is Intel we're talking about

00:09:52.480 --> 00:09:58.560
they're not generally known for their aggressive pricing this thing out

00:09:55.920 --> 00:10:04.000
competes other NVMe drives on price and you'd probably never notice that

00:10:01.360 --> 00:10:07.839
you're running qlc because of the slc cash and you'd have a five year warranty

00:10:06.240 --> 00:10:12.240
to give you peace of mind unless of course the drive hits its

00:10:10.079 --> 00:10:17.440
total drive rights limit before that five years runs out so

00:10:14.880 --> 00:10:21.920
the enthusiast in me has some serious misgivings about this move for the nan

00:10:19.760 --> 00:10:25.279
storage industry but

00:10:22.880 --> 00:10:28.880
this product wasn't made for me that's optane

00:10:26.399 --> 00:10:33.600
and i haven't seen anything about this drive to suggest that it won't do a good

00:10:30.880 --> 00:10:38.880
job for its intended audience thanks to the firmware and controller

00:10:35.680 --> 00:10:41.279
trickery that we alluded to earlier

00:10:38.880 --> 00:10:45.519
i guess that's okay even if it makes me uncomfortable

00:10:43.920 --> 00:10:50.000
do you need to create a beautiful functional website without the hassle

00:10:47.760 --> 00:10:53.200
check out squarespace their all in one platform makes it easy to get up and

00:10:51.680 --> 00:10:56.959
running quickly they've got award-winning templates that you can use

00:10:54.800 --> 00:11:01.920
as starting points for a wide range of projects and if you're having any

00:10:59.360 --> 00:11:07.600
trouble squarespace offers webinars help guides or you can even contact their 24

00:11:04.720 --> 00:11:10.959
7 support via live chat and email if you already have a third party domain you

00:11:09.279 --> 00:11:14.959
don't have to give it up you can just transfer it over to squarespace and you

00:11:13.200 --> 00:11:19.279
get tons of great features including ecommerce to help you sell merch and

00:11:17.200 --> 00:11:24.079
easily manage your inventory and orders so head over to squarespace.com forward

00:11:21.279 --> 00:11:26.880
slash LTT and get 10 off your first purchase we're gonna have that linked

00:11:25.600 --> 00:11:30.880
below so thanks for watching guys if you disliked this video you can hit that

00:11:29.519 --> 00:11:35.760
button but if you liked it hit like get subscribed or maybe consider checking out where to buy the stuff we featured

00:11:33.760 --> 00:11:38.320
at the link below also down there is our merch store which has cool shirts like

00:11:36.959 --> 00:11:41.880
this one and our community forum which you should totally join