WEBVTT

00:00:00.439 --> 00:00:06.319
ddr4 memory it's fast it's power

00:00:03.840 --> 00:00:12.000
efficient it's a requirement if you want to pick up a fancy new CPU that needs it

00:00:09.360 --> 00:00:16.000
and you can learn all about it here so with ddr4

00:00:14.000 --> 00:00:20.000
available why would you want ddr3 anymore on top of all the other stuff

00:00:17.680 --> 00:00:26.000
that i just said ddr4 has four in it that's like one more than

00:00:22.560 --> 00:00:29.599
ddr3 the answer is cost at the time of

00:00:26.000 --> 00:00:32.800
making this video ddr4 is more expensive

00:00:29.599 --> 00:00:35.120
for a given capacity than ddr3 so for

00:00:32.800 --> 00:00:42.000
that reason many potential customers will want to see it justify its higher

00:00:38.160 --> 00:00:44.399
cost so welcome to our apples to apples

00:00:42.000 --> 00:00:47.559
comparison of ddr3 and ddr4

00:00:52.160 --> 00:00:58.079
cooler master v-series semi-modular power supplies feature 80 plus gold

00:00:56.000 --> 00:01:02.399
efficiency and their gold guarantee five year warranty click now to learn more so

00:01:00.399 --> 00:01:08.159
let's introduce our test platform shall we there are no cpus on the market that

00:01:04.960 --> 00:01:10.400
support both ddr3 and ddr4 at least

00:01:08.159 --> 00:01:16.240
consumer ones so we have to be pretty careful to reduce variables and make

00:01:12.880 --> 00:01:18.400
this comparison as fair as possible it

00:01:16.240 --> 00:01:24.000
starts with the processor so we chose two cpus based on the same haswell micro

00:01:21.600 --> 00:01:28.400
architecture from Intel the core i7 4790k

00:01:25.680 --> 00:01:34.880
and the core i7 5820k then we proceeded to disable two of the

00:01:31.439 --> 00:01:37.840
cores on the 5820k and clocked them both

00:01:34.880 --> 00:01:42.640
at straight 4.4 gigahertz with no turbo boost so neither of them would have any

00:01:40.159 --> 00:01:47.200
raw clock speed advantage one thing we couldn't do anything about is the

00:01:44.400 --> 00:01:51.680
difference in cache size but this is as close as we could get to a level playing

00:01:49.200 --> 00:01:57.119
field for the CPU next up is motherboards we've got two Gigabyte

00:01:53.759 --> 00:02:00.320
gaming 5 motherboards the z97x gaming 5

00:01:57.119 --> 00:02:01.119
and the x99 gaming 5. we were looking

00:02:00.320 --> 00:02:06.159
for feature equivalent boards to use for

00:02:03.360 --> 00:02:12.400
both chips and both of these have a very similar loadout with SATA express

00:02:09.280 --> 00:02:15.680
10 gigabit m.2 drive support amplified

00:02:12.400 --> 00:02:18.160
onboard audio qualcomm killer networking

00:02:15.680 --> 00:02:23.200
similar CPU power designs with gigabytes all digital implementation and of course

00:02:20.879 --> 00:02:28.080
a sexy matte black and red color scheme that i find myself caring about probably

00:02:25.120 --> 00:02:32.560
far too much the x99 version does have some extras to go along with its higher

00:02:29.840 --> 00:02:37.840
price a more advanced sound processor three and four-way sli support and some

00:02:35.840 --> 00:02:42.319
led lighting accents including on the heatsink and the i o shield to make it

00:02:40.000 --> 00:02:46.080
easier to plug things in in the dark but enough about that you guys want to know

00:02:43.840 --> 00:02:52.560
about the memory we're using right representing ddr3 at up to the highest

00:02:49.360 --> 00:02:55.280
jdex speed 2133 megahertz with enhanced

00:02:52.560 --> 00:03:02.000
timings we've got a 16 gig dual channel kit of xpg v1 memory resplendent in an

00:02:58.959 --> 00:03:05.599
anodized red finish and representing

00:03:02.000 --> 00:03:08.239
ddr4 at up to the highest jdec ddr4

00:03:05.599 --> 00:03:15.760
speed 2400 megahertz although higher speeds will be coming a 16 gig quad

00:03:11.440 --> 00:03:17.599
channel kit of adata's new ddr4 xpg z1

00:03:15.760 --> 00:03:22.480
series with an updated aggressive looking heat spreader also featuring an

00:03:19.920 --> 00:03:26.000
anodized red finish so huge thanks to adata for providing us with these

00:03:24.080 --> 00:03:30.799
modules for testing even the brand new ddr4 modules operated flawlessly

00:03:28.800 --> 00:03:34.560
throughout the benchmarking process and for that matter so did the motherboards

00:03:32.480 --> 00:03:41.040
i'm really impressed especially with the x99 gaming 5. this one has made a lot of

00:03:38.239 --> 00:03:47.200
progress gigabytes dropping BIOS updates like that and the stability of it is

00:03:43.840 --> 00:03:48.400
much much improved over x99 at launch so

00:03:47.200 --> 00:03:52.959
yeah everything just kind of went smoothly for a change anyway speaking of

00:03:50.959 --> 00:03:57.920
the benchmarking process i guess uh that's what's next so i'm gonna show you

00:03:55.360 --> 00:04:05.519
three different scenarios here first is ddr3 dual channel next is ddr4 dual

00:04:02.640 --> 00:04:10.480
channel so we have a very direct apples to apples comparison between the two

00:04:07.920 --> 00:04:17.280
memory technologies at a variety of speeds and timings and finally to show a

00:04:13.599 --> 00:04:19.519
very best case scenario for modern ddr4

00:04:17.280 --> 00:04:25.040
i threw in a set of results with the RAM overclocked to three gigahertz in quad

00:04:22.720 --> 00:04:29.840
channel mode the configuration that most folks running ddr4 capable processors

00:04:27.360 --> 00:04:34.320
will be running until broadwell comes to the mainstream desktop in dual channel

00:04:32.000 --> 00:04:38.720
only sometime next year now synthetic benchmarks are where ddr4

00:04:37.040 --> 00:04:44.160
should be able to really flex its muscles these ida 64 benches are

00:04:41.360 --> 00:04:48.639
designed to show us the uppermost limits of the theoretical performance of the

00:04:45.919 --> 00:04:53.759
memory subsystem of a pc in our apples to apples test ddr4 delivers better

00:04:51.199 --> 00:04:59.280
performance in reads than ddr3 but actually worse right and latency results

00:04:56.639 --> 00:05:03.919
making this a bit of a wash but in its full quad channel

00:05:00.960 --> 00:05:09.199
configuration our high speed ddr4 is untouchable here next up is real world

00:05:06.880 --> 00:05:13.360
non-gaming results multi-threaded rendering and our 7-zip compression and

00:05:11.039 --> 00:05:18.560
decompression benchmarks are a good way to show off the performance of modern

00:05:15.600 --> 00:05:23.120
cpus with lots of processing cores and high-speed memory is needed to keep all

00:05:20.560 --> 00:05:26.720
of those cores fed so something to note here is that while the performance

00:05:24.800 --> 00:05:32.240
scaling with better memory might not look very impressive the lga 2011 3

00:05:29.600 --> 00:05:38.880
socket is likely to be with us for a few years and has cpus available for it

00:05:34.960 --> 00:05:41.840
already with up to 18 cores so while the

00:05:38.880 --> 00:05:47.280
extra bandwidth even in quad channel isn't needed for this artificial

00:05:44.639 --> 00:05:51.360
quad core that we created here there are ways to kit out a board like this so

00:05:49.440 --> 00:05:55.360
that it's much more likely that that extra memory bandwidth would be used

00:05:53.759 --> 00:06:00.240
which leads to our final set of benchmarks games i only ended up running

00:05:58.240 --> 00:06:04.080
a couple of games the canned benchmark in tomb raider at these settings and my

00:06:02.400 --> 00:06:08.560
custom run at the end of bioshock infinite at these settings because the

00:06:06.160 --> 00:06:14.400
results here don't leave much for me to interpret even going all the way down to

00:06:11.039 --> 00:06:17.360
ddr3 13 33 megahertz so nearly the

00:06:14.400 --> 00:06:21.039
launch speed of ddr3 in dual channel i couldn't create a situation where game

00:06:19.600 --> 00:06:27.120
performance was limited in any meaningful way by the memory subsystem

00:06:24.000 --> 00:06:28.400
on a haswell-based CPU with only four

00:06:27.120 --> 00:06:33.199
cores so in conclusion while it might be

00:06:30.639 --> 00:06:37.919
worthwhile from a platform point of view to invest in ddr4 for your six core or

00:06:35.919 --> 00:06:42.160
eight core processor since you don't really have a choice if you want the

00:06:39.280 --> 00:06:47.280
latest cpus if you're a normal consumer who doesn't really need the extra course

00:06:44.240 --> 00:06:50.240
it looks like lga 1150 with its ancient

00:06:47.280 --> 00:06:56.000
ddr3 memory still has a lot of life left in it and given how unconstrained our

00:06:52.720 --> 00:06:58.560
haswell cpus are even by very low speed

00:06:56.000 --> 00:07:02.080
ddr3 i think it's safe to say that the performance improvements we can expect

00:07:00.240 --> 00:07:06.160
to see from Intel's upcoming broadwell desktop processors will be thanks to

00:07:04.319 --> 00:07:11.199
architectural improvements not the use of ddr4

00:07:09.360 --> 00:07:14.319
guys like this video if you liked it dislike it if you disliked it leave a

00:07:12.720 --> 00:07:19.120
comment letting me know what you thought were you expecting ddr4 in an apples to

00:07:17.039 --> 00:07:23.199
apples comparison to run away with it or were you expecting exactly the results

00:07:21.360 --> 00:07:26.800
you saw today check out the link in the video description if you want to support

00:07:24.720 --> 00:07:29.680
us it says support us you click that link you can buy a cool t-shirt like

00:07:28.160 --> 00:07:34.000
this one you can give us a monthly contribution or you can change your amazon bookmark to one with our

00:07:32.319 --> 00:07:37.680
affiliate code so we get a small kickback whenever you buy toothbrushes

00:07:35.680 --> 00:07:41.520
or whatever it is you buy on amazon thanks for watching again guys and as

00:07:39.840 --> 00:08:00.800
always don't forget to subscribe for more videos just like this one

00:08:00.800 --> 00:08:02.879
you