WEBVTT

00:00:00.320 --> 00:00:07.520
so you've come looking for ddr5 i'll cut you a deal

00:00:04.640 --> 00:00:12.320
only double retail pricing it's a steal

00:00:09.519 --> 00:00:15.550
for me that is

00:00:18.320 --> 00:00:23.840
scalpers have taken the fledgling ddr5 market hostage look around you

00:00:22.480 --> 00:00:29.840
everything on newegg is out of stock right now and low end 16 gig modules of

00:00:26.880 --> 00:00:34.960
ddr5 are being sold for as much as 32 gigs of trident z is supposed to be

00:00:32.960 --> 00:00:40.239
meanwhile on ebay you're looking at anywhere from 700 to a

00:00:37.600 --> 00:00:43.760
thousand dollars for a 32 gig kit of the good stuff we're in a bad place right

00:00:41.840 --> 00:00:49.280
now but there's a light a light of justice because the fact that

00:00:46.160 --> 00:00:51.200
ddr5 is newer and shinier doesn't mean

00:00:49.280 --> 00:00:54.960
that you need to buy it unlike our sponsor glasswire lets you

00:00:52.960 --> 00:00:58.800
see past and present network activity detect malware on your pc or Android

00:00:56.640 --> 00:01:02.640
device and block its connections to prevent things from getting worse use

00:01:00.399 --> 00:01:06.120
offer code Linus to get 25 off at the link down below

00:01:12.799 --> 00:01:18.560
i know i know we did a whole video about why

00:01:16.000 --> 00:01:23.200
ddr5 is super cool and it really is but it's got a lot

00:01:21.280 --> 00:01:27.680
working against it right now too you've got first gen RAM chips interfacing with

00:01:25.759 --> 00:01:32.720
first gen memory controllers over imperfectly optimized first gen trace

00:01:29.920 --> 00:01:37.200
layouts and it's perfectly natural to expect all this first january to take a

00:01:35.360 --> 00:01:41.040
while to reach its full potential i mean think about how the first playstation 2

00:01:39.040 --> 00:01:45.360
games ended up looking worse than the last playstation 1 games because

00:01:42.960 --> 00:01:49.680
developers had no idea yet how to unlock the full potential of the emotion engine

00:01:47.360 --> 00:01:55.600
and even in the computer memory space this is nothing new ddr2 ddr3 ddr4 if

00:01:53.600 --> 00:02:00.079
you look at contemporaneous reviews you'll find a lot of reports saying that

00:01:57.600 --> 00:02:05.119
there is no practical benefit to the upgrade and that it is certainly not

00:02:02.640 --> 00:02:10.080
worth the early adopter tax like look at this dork back in 2014

00:02:07.360 --> 00:02:14.959
the synthetic benchmark graphs show ddr4 absolutely wiping the floor with ddr3

00:02:13.040 --> 00:02:19.840
and raw bandwidth but when we look at the numbers in real world applications

00:02:17.360 --> 00:02:25.440
even this relatively low spec kit of ddr3 beats out the fastest ddr4 that was

00:02:23.440 --> 00:02:28.959
available at the time of course

00:02:26.480 --> 00:02:33.280
as we all know ddr4 like its predecessors ended up getting the last

00:02:31.200 --> 00:02:38.319
laugh with mid-generation refinements that led to much higher performance at

00:02:35.680 --> 00:02:44.560
ultimately lower costs i mean the hot item back when that video was shot was 3

00:02:40.959 --> 00:02:46.640
000 mega transfers per second cl 16.

00:02:44.560 --> 00:02:53.680
nowadays we would call that budget at best cheapo at worst and the same thing will

00:02:50.160 --> 00:02:56.239
eventually happen with ddr5 just

00:02:53.680 --> 00:02:59.440
not today unlike our new water bottles those are already proven superior and

00:02:57.920 --> 00:03:04.000
you can get them right now no scalping required ltdstore.com to illustrate our

00:03:02.000 --> 00:03:10.239
point we ran a bevy of benchmarks using our core i9 12900k paired with an rtx

00:03:06.560 --> 00:03:12.080
3090 this is admittedly a very high spec

00:03:10.239 --> 00:03:15.840
machine that not everyone will be using but it allows both of our memory

00:03:14.159 --> 00:03:20.239
technologies to put their best foot forward by removing system bottlenecks

00:03:17.840 --> 00:03:24.239
elsewhere for ddr5 we've included both the highest currently available jdx

00:03:22.000 --> 00:03:28.080
speed and a 5200 mega transfer per second kit which was the fastest that

00:03:26.319 --> 00:03:32.720
our system could handle reliably and then we've included a handful of common

00:03:29.920 --> 00:03:37.680
ddr4 configurations kicking things off our high-end ddr5 pulls off a

00:03:34.560 --> 00:03:39.680
respectable 5 lead over our fastest ddr4

00:03:37.680 --> 00:03:42.799
kit in f1 2021 though it should be noted that we're

00:03:41.519 --> 00:03:46.959
having issues running more than two sticks of the stuff at a time right now

00:03:44.560 --> 00:03:52.560
and the other problem is that our lower end ddr5 kit drops off significantly

00:03:50.239 --> 00:03:58.159
from its bigger brother in fact while the average FPS drop looks fine it

00:03:54.879 --> 00:04:01.040
trails even our top spec ddr4 by as much

00:03:58.159 --> 00:04:04.959
as 10 percent in minimum frame times and that could easily manifest as little

00:04:02.879 --> 00:04:08.959
hitches or stutters in your gameplay forza is a little bit kinder to our ddr5

00:04:07.439 --> 00:04:13.280
but the improvement offered by the high-end kit is negligible and the

00:04:10.720 --> 00:04:17.199
benefit over ddr4 is under 5 in both cases flight simulator's CPU

00:04:15.519 --> 00:04:21.040
intensive nature seems like it should be a good fit for ddr5

00:04:18.959 --> 00:04:26.479
but no both kits are actually consistently between one and four

00:04:22.960 --> 00:04:28.800
percent slower than our fast ddr4 kit

00:04:26.479 --> 00:04:33.680
that's surprising but not as surprising as cs go another game that's CPU heavy

00:04:31.840 --> 00:04:37.680
while the high end kit stays roughly equivalent with a few frames gained or

00:04:35.440 --> 00:04:43.280
lost depending on the metric our lower end kit is on par or worse than ddr4

00:04:41.040 --> 00:04:46.240
running 400 mega transfers per second lower

00:04:44.320 --> 00:04:50.400
save 6 for its part was more or less the same regardless with less than half a

00:04:48.080 --> 00:04:53.759
second separating our ddr5 from our fastest ddr4

00:04:52.160 --> 00:04:58.400
taken together then the gaming tests average out to about a percent win for

00:04:55.759 --> 00:05:03.360
the high-end kit and a two percent loss for the low-end kit compared to high-end

00:05:00.080 --> 00:05:06.400
ddr4 which and i cannot emphasize this

00:05:03.360 --> 00:05:09.600
enough costs less than half as much at

00:05:06.400 --> 00:05:12.000
retail and that is pre-scalper markup

00:05:09.600 --> 00:05:16.400
perhaps productivity will produce ddr5's proper performance potential cinebench

00:05:14.400 --> 00:05:22.080
sure doesn't think so it's a total wash with a measly one percent difference or

00:05:18.960 --> 00:05:22.960
less between our fastest ddr4 and ddr5

00:05:22.080 --> 00:05:27.680
kits compiling firefox though gives ddr5 its

00:05:25.600 --> 00:05:32.800
first major win with even the lower end kit picking up a six percent faster time

00:05:29.919 --> 00:05:37.759
compared to our best ddr4 and the higher end kit being nearly a minute faster

00:05:35.600 --> 00:05:42.240
code compilation does hit system memory pretty hard though so this isn't a huge

00:05:39.280 --> 00:05:45.680
surprise it's just nice to see a w for it somewhere blender meanwhile doesn't

00:05:44.000 --> 00:05:50.479
seem to give two hoots what kind of memory you have as long as it's lots and

00:05:48.400 --> 00:05:54.880
puget bench is an interesting case because every adobe app tested here gets

00:05:52.560 --> 00:06:00.720
the performance uplift but none of them are as dark as photoshop even our low

00:05:57.520 --> 00:06:03.680
end ddr5 kit is 14 faster than our high

00:06:00.720 --> 00:06:07.280
end ddr4 kit and that's saying a lot though this again makes sense because of

00:06:05.280 --> 00:06:11.120
photoshop's memory intensive layer and effects operations where the GPU isn't

00:06:09.440 --> 00:06:15.039
really doing a lot of the work finally we've got spec workstation where some

00:06:12.840 --> 00:06:19.919
ridiculously memory intensive tests come into play giving us substantial 20 and

00:06:17.520 --> 00:06:24.639
even 30 performance uplifts in the product development life sciences and

00:06:22.240 --> 00:06:29.199
energy suites so that's stuff like cad medical imaging and geological surveys

00:06:27.039 --> 00:06:31.440
where there's a lot of data being thrown around

00:06:30.080 --> 00:06:37.520
financial services and general operations meanwhile not so hot which we'd expect considering

00:06:35.600 --> 00:06:40.800
the lighter nature of that work and our previous results overall then

00:06:39.280 --> 00:06:44.639
productivity nets has a rather significant but

00:06:42.560 --> 00:06:48.720
kind of inconsistent five to seven percent bump in performance with ddr5

00:06:46.560 --> 00:06:52.960
compared to the fastest ddr4 so this is one area where depending on

00:06:50.880 --> 00:06:56.639
what you do ddr5 might be worth the investment although i gotta say that

00:06:55.199 --> 00:07:01.759
better be some pretty important work you're doing to be clear

00:06:59.120 --> 00:07:06.560
none of what we're saying is that ddr5 has no advantages for average gamers

00:07:04.319 --> 00:07:10.000
it's more power efficient for one and the nature of its more numerous but

00:07:08.400 --> 00:07:14.639
smaller burst buffers means that architectures that rely on responsive

00:07:11.840 --> 00:07:19.599
memory like ryzen may find more benefit even at these early speeds despite the

00:07:16.639 --> 00:07:24.800
looser timings compared to ddr4 but since a ddr5 ryzen platform doesn't

00:07:22.319 --> 00:07:28.800
exist yet and by the time it does both motherboard manufacturers and memory

00:07:26.960 --> 00:07:32.720
manufacturers might have a better idea of how to optimize their implementations

00:07:31.199 --> 00:07:36.639
i think it's safe to say that you shouldn't be paying a premium for it today

00:07:35.120 --> 00:07:41.199
leading us to the big question we need to answer here why are scalpers scalping

00:07:39.039 --> 00:07:46.400
it then well i think there's a big emotional component people like having

00:07:43.680 --> 00:07:49.280
the latest and greatest thing and i think that we in the media need to

00:07:47.680 --> 00:07:54.080
accept at least part of the blame for that but there's another angle to it and

00:07:51.759 --> 00:08:00.080
that's motherboard manufacturers all of the highest and shiniest most appealing

00:07:56.400 --> 00:08:02.000
motherboards are ddr5 only and now that

00:08:00.080 --> 00:08:06.160
there's a shortage of it it's not just as simple for them as

00:08:03.759 --> 00:08:09.440
plonking down some ddr4 slots in place of the ddr5 ones to improve consumer

00:08:08.400 --> 00:08:13.120
options trace layouts and lengths topology and

00:08:11.840 --> 00:08:18.639
of course the fact that the power management ic is on each ddr5 module

00:08:17.039 --> 00:08:22.720
rather than built into the board means that there is a non-trivial amount of

00:08:20.720 --> 00:08:27.360
work involved with retooling a board for ddr4

00:08:24.240 --> 00:08:29.840
the good news for you today is that most

00:08:27.360 --> 00:08:33.120
of the highest end motherboards don't really do a lot to justify the extra

00:08:32.240 --> 00:08:37.519
spend at least if you're not overclocking with liquid nitrogen

00:08:36.240 --> 00:08:41.919
as we've seen in our previous motherboard investigation down to earth

00:08:39.599 --> 00:08:45.920
gaming motherboards like ASUS rg strix or MSI's tomahawk series are going to

00:08:44.000 --> 00:08:51.279
give you performance and features very close to the bleeding edge and if

00:08:48.240 --> 00:08:53.040
you really must have high end Gigabyte

00:08:51.279 --> 00:08:57.839
actually has gone to the effort of remaking nearly their entire lineup in

00:08:55.440 --> 00:09:00.880
ddr4 flavors leaving only the cream of the crop out of reach we're gonna have

00:08:59.519 --> 00:09:03.920
some links to these down below if you're looking to build a 12th gen rig for

00:09:02.800 --> 00:09:08.880
today though the solution is to hoist a great big

00:09:05.920 --> 00:09:12.399
middle finger to scalpers by buying ddr4 while it's still relevant hopefully by

00:09:11.040 --> 00:09:16.720
the time you don't want it anymore the scalpers will have moved on to greener

00:09:14.640 --> 00:09:20.560
pastures maybe some game console they can buy or something

00:09:19.279 --> 00:09:26.000
actually please don't do that game consoles are one of the only ways people can get a like reasonably priced gaming

00:09:24.080 --> 00:09:29.680
rig right now thanks brilliant for sponsoring today's

00:09:27.519 --> 00:09:33.120
video brilliant teaches you by solving puzzles and problems on their website

00:09:31.200 --> 00:09:37.519
and app they've got over 60 interactive courses in math science and computer

00:09:35.279 --> 00:09:40.800
science and brilliant is just a great website to help you achieve your goals

00:09:38.959 --> 00:09:44.160
in stem their courses are designed to puzzle and surprise you and one that

00:09:42.720 --> 00:09:48.399
probably all of you will like is mathematical fundamentals is it possible

00:09:46.640 --> 00:09:51.760
to drag the number tiles so that every row and column adds up to the target sum

00:09:50.000 --> 00:09:56.240
beside it find out by giving it a try today just go to brilliant.org Linus

00:09:54.160 --> 00:10:00.240
tech tips to find out the first 200 of you to click on the link will get 20 off

00:09:58.320 --> 00:10:05.200
today thanks for watching guys go check out our previous video on ddr5 for more

00:10:02.720 --> 00:10:09.920
info on the inner workings and just how it differs from ddr4 i promise it is

00:10:07.920 --> 00:10:14.240
super interesting even if you can't or at least shouldn't buy it right now