WEBVTT

00:00:02.000 --> 00:00:07.919
all right time for a build 8700k

00:00:06.640 --> 00:00:12.160
check z370 motherboard

00:00:10.559 --> 00:00:15.519
check two RAM

00:00:15.519 --> 00:00:20.320
one stick hold on we'll be back

00:00:24.160 --> 00:00:28.279
where the devil is all the RAM

00:00:31.279 --> 00:00:39.120
and that's actually a terrific question

00:00:35.760 --> 00:00:41.200
dram has been in a global shortage

00:00:39.120 --> 00:00:48.399
allegedly due to collusion between the major producers for months so

00:00:45.360 --> 00:00:51.360
when Intel reached out asking for some

00:00:48.399 --> 00:00:57.199
creative ideas for a sponsored video we pitched using their optane caching

00:00:54.879 --> 00:01:02.239
modules as a way of topping up your system memory

00:01:00.399 --> 00:01:05.760
and for some reason pig green lit this

00:01:03.840 --> 00:01:09.880
so let's give it a shot

00:01:18.320 --> 00:01:27.680
so obviously we're still gonna need some system memory but the plan really is to

00:01:24.080 --> 00:01:28.560
take this and use it to replace some of

00:01:27.680 --> 00:01:32.880
this and the rationale behind this is actually a lot less crazy than you'd

00:01:32.000 --> 00:01:39.200
think in our testing we've seen that optane

00:01:36.079 --> 00:01:41.960
performs more like RAM than it does like

00:01:39.200 --> 00:01:45.920
an SSD which is to say that it's got ridiculously low response times compared

00:01:44.880 --> 00:01:50.720
to this all while being much cheaper than this

00:01:49.200 --> 00:01:56.000
and conveniently even though our optane module here plugs

00:01:53.600 --> 00:02:03.040
into an m.2 slot so this is something that is designed for an SSD

00:01:59.600 --> 00:02:06.640
Windows already has functionality that

00:02:03.040 --> 00:02:09.840
lets us use a regular old

00:02:06.640 --> 00:02:13.680
storage device as an extension of the

00:02:09.840 --> 00:02:16.400
system memory so we will give up quite a

00:02:13.680 --> 00:02:21.760
lot of our throughput compared to actual RAM

00:02:17.360 --> 00:02:24.400
but for small transactions that might

00:02:21.760 --> 00:02:31.440
not actually hurt us that much and it will certainly be faster than using a

00:02:26.879 --> 00:02:35.200
traditional NVMe or SATA SSD or

00:02:31.440 --> 00:02:37.200
heaven forbid a hard drive in this role

00:02:35.200 --> 00:02:41.519
to put this theory to the test we're gonna do something a little bit strange

00:02:39.200 --> 00:02:46.160
so we've actually installed two eight gig sticks of memory in our bench to

00:02:43.920 --> 00:02:52.640
give us the benefit of dual channel but then we're gonna use Windows built-in ms

00:02:48.879 --> 00:02:57.760
config utility to artificially

00:02:52.640 --> 00:02:57.760
limit our memory to four gigs

00:02:58.160 --> 00:03:01.680
without a plus sign in there

00:03:02.080 --> 00:03:09.360
so our four gig entry actually gives us just shy of three gigs of usable memory

00:03:08.319 --> 00:03:16.480
all right let's fire up some gaming benchmarks

00:03:12.720 --> 00:03:18.800
these numbers actually start out showing

00:03:16.480 --> 00:03:22.879
little concern for the lack of available RAM

00:03:19.680 --> 00:03:25.120
however in the open world far cry 5 we

00:03:22.879 --> 00:03:30.000
see a significant drop in 97th percentile frame rates as assets load in

00:03:28.080 --> 00:03:35.120
on the fly as for unigine's superposition we did see a small dip but

00:03:32.799 --> 00:03:40.040
that could just be due to run to run variation

00:03:36.640 --> 00:03:41.680
moving on to productivity things get

00:03:40.040 --> 00:03:48.400
painful adobe premiere took a whopping three

00:03:44.959 --> 00:03:51.920
times longer than our native 16 gig

00:03:48.400 --> 00:03:55.280
tests and blender is much the same story

00:03:51.920 --> 00:03:57.920
taking over two hours to complete where

00:03:55.280 --> 00:04:02.799
our non-limited testing gave us a result under 45 minutes

00:04:00.560 --> 00:04:08.720
let's hope for Intel's sake that things go a little bit better

00:04:06.000 --> 00:04:08.720
with optane

00:04:10.040 --> 00:04:15.280
unsurprisingly our gaming results

00:04:13.040 --> 00:04:21.280
don't really change much though we do gain back most of our 97th

00:04:18.320 --> 00:04:26.000
percentile frames on far cry 5 compared to the 3 gig test and unigine

00:04:23.440 --> 00:04:30.320
superposition did pick up a cool 33 points of probably run to run

00:04:28.639 --> 00:04:34.639
variants moving on to productivity things look

00:04:33.600 --> 00:04:40.320
well actually a lot better we shave about a minute off of adobe

00:04:39.199 --> 00:04:45.040
premiere though we didn't get that close to true

00:04:42.560 --> 00:04:50.240
RAM but as for blender wow

00:04:46.320 --> 00:04:51.280
blender took a whole hour off the render

00:04:50.240 --> 00:04:57.600
time i mean to be clear that still 25 and a

00:04:54.639 --> 00:05:01.840
half minutes longer than if we had all just RAM

00:04:59.040 --> 00:05:07.360
but considering that this render calls for a minimum of 12 gigs of RAM and

00:05:04.320 --> 00:05:09.840
takes about 12 and a half we know that a

00:05:07.360 --> 00:05:15.680
considerable amount of our working data was sitting in our optane module

00:05:13.120 --> 00:05:19.680
but that's only part of the story let's take an even closer look at these

00:05:18.560 --> 00:05:23.759
results what you're looking at now is a scatter

00:05:21.919 --> 00:05:29.039
plot of frame times and the corresponding disk activity where we can

00:05:26.000 --> 00:05:31.360
see that most of our spikes in

00:05:29.039 --> 00:05:36.560
instantaneous frame times so these would be perceived as hitches

00:05:33.600 --> 00:05:41.520
are actually caused by spikes in activity and this is especially evident

00:05:39.199 --> 00:05:46.639
when we look at far cry 5. we can also look at how the relationship

00:05:43.600 --> 00:05:49.280
works between drive activity and CPU

00:05:46.639 --> 00:05:56.560
usage during our premiere and blender tests where we can see well mostly a

00:05:53.199 --> 00:05:59.240
mess but but squint harder look at this

00:05:56.560 --> 00:06:07.919
the dips in drive activity actually correspond to spikes in CPU usage

00:06:03.360 --> 00:06:12.000
meaning our CPU is being kept fed so the

00:06:07.919 --> 00:06:13.600
less a drive thrashes the more our CPU

00:06:12.000 --> 00:06:18.400
gets to work nowhere is this more apparent than our

00:06:15.600 --> 00:06:24.400
blender test where our non-optane result had the CPU working at below 50 capacity

00:06:22.400 --> 00:06:30.319
most of the time so what can we take away here

00:06:27.520 --> 00:06:35.120
well for one octane memory modules aren't going to be replacing system RAM

00:06:33.360 --> 00:06:38.639
anytime soon but

00:06:36.400 --> 00:06:42.880
think of the potential out of necessity it's common practice

00:06:40.880 --> 00:06:48.000
for computing workloads to be coded around the availability of modern

00:06:45.280 --> 00:06:52.240
hardware so we needed to create a pretty contrived experiment to show such a

00:06:50.080 --> 00:06:59.440
clear benefit today but imagine if developers or creative

00:06:55.919 --> 00:07:02.400
types or researchers were free to create

00:06:59.440 --> 00:07:07.840
much larger data sets knowing that they'd have a cost-effective way to work

00:07:04.960 --> 00:07:12.080
around bottlenecking their CPU we actually saw this in action in our

00:07:09.520 --> 00:07:17.360
optane 900p video so for today we're still recommending

00:07:14.880 --> 00:07:20.960
optane modules for accelerating slower storage devices like hard drives

00:07:20.160 --> 00:07:25.840
but this was a really fun exercise in future

00:07:23.440 --> 00:07:31.280
crafting and it really shows off the potential of the underlying optane

00:07:28.880 --> 00:07:34.240
memory so thanks for watching guys if you disliked this video you can hit that

00:07:32.800 --> 00:07:39.680
button but if you liked it hit like get subscribed maybe consider checking out where to buy the stuff we featured at

00:07:37.440 --> 00:07:43.599
the link in the video description again to accelerate other storage not to

00:07:41.759 --> 00:07:46.240
replace your system memory while you're down there you can check out our merch

00:07:44.960 --> 00:07:53.240
store which has cool shirts like this one and our community forum it's kind of

00:07:48.639 --> 00:07:53.240
a cool place to hang out and talk tech