1
00:00:00,560 --> 00:00:07,279
he hey everyone it's me the tech tips

2
00:00:03,760 --> 00:00:10,519
man now that ddr5 has reached insane

3
00:00:07,279 --> 00:00:13,120
speeds of over 10,000 megga transfers

4
00:00:10,519 --> 00:00:17,480
per second we've got to know which sticks should you spend your hard-earned

5
00:00:15,200 --> 00:00:22,199
money on a single kit could cost you anywhere from

6
00:00:18,560 --> 00:00:24,560
$160 to over

7
00:00:22,199 --> 00:00:30,080
$400 so making the right choice can be the difference between wallet pain and

8
00:00:27,119 --> 00:00:36,000
wallet death to help you out we put the screws to 10 kits ranging from stock JX

9
00:00:33,160 --> 00:00:42,520
speeds all the way up to 7,800 megat transfers per second on both

10
00:00:39,120 --> 00:00:45,360
Intel and AMD platforms does it matter a

11
00:00:42,520 --> 00:00:50,640
little does it matter a lot has the lab managed to find the ddr5 price to

12
00:00:48,000 --> 00:00:55,320
Performance Sweet Spot there's only one way to find out lenoe cloud computing

13
00:00:53,239 --> 00:00:59,320
from aami Len Noe is a powerful Linux-based cloud computing service

14
00:00:57,399 --> 00:01:02,719
that's affordable and easy to use whether for a small project or a

15
00:01:00,840 --> 00:01:08,880
business critical application see why developers choose lenoe sign up today at

16
00:01:05,040 --> 00:01:11,560
Len no.com to get a free $100 60-day

17
00:01:08,880 --> 00:01:18,680
credit right now Intel has the only platform where we can test both ddr5 and

18
00:01:14,520 --> 00:01:20,680
ddr4 on the exact same CPU so that's

19
00:01:18,680 --> 00:01:24,079
where we started and Intel also had the benefit of being more stable throughout

20
00:01:22,600 --> 00:01:27,479
our testing we'll get into that a little bit more later for now all you need to

21
00:01:26,119 --> 00:01:32,560
know is that we took some of the beefiest hardware commercially available

22
00:01:29,520 --> 00:01:34,479
plugged in some memory kits enabled XMP

23
00:01:32,560 --> 00:01:38,159
and went to town we're going to have all the gear that we used linked down below

24
00:01:36,200 --> 00:01:43,960
if you want to check it out for yourself right out of the gate we can clearly see

25
00:01:40,000 --> 00:01:46,640
that in most games fast ddr4 is still a

26
00:01:43,960 --> 00:01:50,920
valid option but that price to Performance advantage that it enjoyed in

27
00:01:48,439 --> 00:01:55,520
the past seems to have pretty much disappeared this is a 3600 megat

28
00:01:53,520 --> 00:02:01,560
transfer per second kit with pretty tight timings and while it beats our

29
00:01:58,240 --> 00:02:04,079
ddr5 jck kit by anywhere from 1 to 10

30
00:02:01,560 --> 00:02:09,319
FPS on average it does so at a considerably higher price and once we're

31
00:02:06,600 --> 00:02:15,239
spending the same amount on our ddr5 our ddr4 not only loses its Advantage but

32
00:02:12,440 --> 00:02:21,720
falls measurably behind especially in the all important 1% lows the thing is

33
00:02:18,879 --> 00:02:29,480
average frame rates are interesting but the difference between 311 and 3155 FPS

34
00:02:26,720 --> 00:02:35,640
is basically impossible to discern so this bar usually matters a lot more

35
00:02:32,400 --> 00:02:38,400
because a 10 or even 5 FPS difference is

36
00:02:35,640 --> 00:02:42,519
much more likely to be noticeable and it also represents how your system will

37
00:02:40,319 --> 00:02:46,840
perform in the most critical moments when the action is most intense for

38
00:02:44,599 --> 00:02:52,239
gamers then ddr5 is looking like an obvious choice and and more must be

39
00:02:49,040 --> 00:02:55,560
better right not necessarily adding our

40
00:02:52,239 --> 00:02:57,800
7600 and 7800 megat transfer per second

41
00:02:55,560 --> 00:03:03,879
kits to the graphs I mean yeah they performed admirably but just a few more

42
00:03:00,560 --> 00:03:06,440
FPS and sometimes less for a hundred

43
00:03:03,879 --> 00:03:11,040
plus dollars more than our other options what's going on here I mean look at

44
00:03:08,519 --> 00:03:14,840
these numbers in ID to 64 the more we spend the more memory bandwidth we get

45
00:03:13,120 --> 00:03:21,480
so our performance should go up proportionally right well here's the

46
00:03:17,799 --> 00:03:23,519
thing Ida 64's read write and copy

47
00:03:21,480 --> 00:03:28,720
benchmarks are super useful for uncovering system bottlenecks and

48
00:03:25,599 --> 00:03:30,640
diagnostic purposes but they are purely

49
00:03:28,720 --> 00:03:35,799
synthetic that that is to say that they are not representative of any real world

50
00:03:33,519 --> 00:03:41,519
workload other than copying data directly into or out of memory and not

51
00:03:39,000 --> 00:03:47,040
every application is actually hungry for more bandwidth for some applications

52
00:03:44,239 --> 00:03:52,599
latency is actually much more important and that's where our ddr4 kit shines

53
00:03:49,680 --> 00:03:57,920
with only the fastest ddr5 kits closing the Gap most of the ddr5 kits that we

54
00:03:55,720 --> 00:04:05,799
tested today hit first word latency figures of 10 NS or faster but our ddr4

55
00:04:01,879 --> 00:04:07,280
kit managed less than 8 NCS you can

56
00:04:05,799 --> 00:04:11,720
roughly calculate these values for your own kit by the way just take your

57
00:04:09,200 --> 00:04:16,120
module's cast latency or how long it takes to complete a clock cycle then

58
00:04:13,720 --> 00:04:21,280
multiply it by 2,000 and divide by the transfer speed the lower the first word

59
00:04:18,519 --> 00:04:25,919
latency generally the snappier your memory will be but as evidenced by the

60
00:04:24,280 --> 00:04:31,680
performance improvements we see with our faster kits it is both this latency and

61
00:04:29,720 --> 00:04:35,160
transfer speeds that are crucial when it comes to getting the most out of your

62
00:04:33,080 --> 00:04:40,000
system's memory for applications and gains today look at these two 6,000

63
00:04:37,840 --> 00:04:44,880
megat transfer per second kits there's a pretty Stark difference in cast timings

64
00:04:42,000 --> 00:04:52,479
here resulting in a 3.3 nond difference in latency the faster one is a little

65
00:04:47,680 --> 00:04:55,280
over $20 more for about a 1 to 5 FPS

66
00:04:52,479 --> 00:05:02,039
Improvement but if we bump up the speed to this 6400 kit with timings that keep

67
00:04:57,960 --> 00:05:04,080
us at 10 NCS we get more performance but

68
00:05:02,039 --> 00:05:08,639
start to hit our point of diminishing value returns and then anything past

69
00:05:06,600 --> 00:05:13,120
this point sees such a stark drop in performance Improvement that it doesn't

70
00:05:10,600 --> 00:05:17,800
really warrant spending tens or hundreds of extra dollars to try to make the line

71
00:05:15,039 --> 00:05:23,120
go up at least at current pricing but what about AMD well for team red our

72
00:05:20,960 --> 00:05:28,199
biggest hurdle was just getting kits faster than 6,400 megat transfers per

73
00:05:25,120 --> 00:05:30,560
second to even run long story short they

74
00:05:28,199 --> 00:05:36,199
didn't and that's likely to change anytime soon our testing started mid

75
00:05:33,440 --> 00:05:41,759
last month so that's December of 2022 and we were running on a GSA version

76
00:05:38,039 --> 00:05:43,759
1.0.0 do3 patch d right before we filmed

77
00:05:41,759 --> 00:05:48,800
this Gigabyte dropped a new BIOS for our motherboard that included an update to

78
00:05:46,400 --> 00:05:53,680
1.0.4 supposedly improving memory support and it does but the difference

79
00:05:51,960 --> 00:05:58,400
wasn't enough to boot any of our previously incompatible kits I'm certain

80
00:05:56,639 --> 00:06:03,160
that professional overclockers will be able to use this extra heav room to push

81
00:06:00,639 --> 00:06:08,039
the platform to new heights but our goal was to measure the outof Box experience

82
00:06:05,880 --> 00:06:13,800
so we didn't do any tuning Beyond plugging in the CPU plugging in the RAM

83
00:06:10,160 --> 00:06:15,360
and enabling XMP or Expo as it were the

84
00:06:13,800 --> 00:06:21,199
good news is that even though it relatively slow our 6400 memory

85
00:06:18,199 --> 00:06:24,120
absolutely slayed on AMD going from a

86
00:06:21,199 --> 00:06:30,919
basic ddr5 kit to a premium but still not outlandishly priced kit saw almost

87
00:06:26,720 --> 00:06:32,880
every game gain 10% average FPS or more

88
00:06:30,919 --> 00:06:37,680
this is especially notable because these are similar gains to what we saw in

89
00:06:34,759 --> 00:06:42,120
Intel but we aren't spending as much on our top performing memory since we

90
00:06:39,560 --> 00:06:45,759
topped out at a lower speed and it's safe to assume that if future BIOS

91
00:06:43,840 --> 00:06:50,120
updates can get the jello to stop shaking AMD's performance should only

92
00:06:47,960 --> 00:06:56,240
improve as we get stable access to significantly faster memory

93
00:06:52,160 --> 00:06:57,960
right well not exactly we are still

94
00:06:56,240 --> 00:07:03,360
hitting the point of diminishing returns here especially when you consider to the

95
00:06:59,840 --> 00:07:05,840
price AMD's chips really like tighter

96
00:07:03,360 --> 00:07:12,879
timings and that seems to be the biggest Factor when it comes to games I mean

97
00:07:08,360 --> 00:07:15,039
sure the JX spec 4800 cl40 kit is way

98
00:07:12,879 --> 00:07:22,280
slower than our top end but look at this 5600 cl28 result it's on par or just

99
00:07:19,840 --> 00:07:28,560
behind its more expensive counterpart and for about $20 less and the reason

100
00:07:25,680 --> 00:07:33,840
lies here while the transfer speed might be higher on that 6400 kit it actually

101
00:07:31,360 --> 00:07:38,479
has the same first word latency as the 5600 configuration I mean if you were to

102
00:07:36,520 --> 00:07:41,800
take that faster kit and sit around for hours tuning the timings chances are

103
00:07:40,639 --> 00:07:45,639
that you would end up with a better result during our quick sanity check

104
00:07:43,919 --> 00:07:52,599
with the latest BIOS we found that a jisa 1.0.4 does improve stability at 6,400

105
00:07:50,120 --> 00:07:57,039
with manually tightened timings but the reality is that many of you don't even

106
00:07:54,520 --> 00:08:01,120
overclock your CPU these days let alone your memory and realistically your time

107
00:07:59,159 --> 00:08:04,680
is probably better spent buying our new underwear from LTT Store.com we've got

108
00:08:03,199 --> 00:08:09,520
new colors available check them out today at least that's true for gaming

109
00:08:07,599 --> 00:08:14,039
for productivity where we'll have a lot more CPU cores to feed with delicious

110
00:08:11,759 --> 00:08:18,039
data it's a whole different Beast similar to what we saw with Intel faster

111
00:08:15,919 --> 00:08:22,400
modules seem to make a huge difference in some workloads and then basically

112
00:08:20,360 --> 00:08:28,720
none in others handbrake for example sees an impressive boost of

113
00:08:24,759 --> 00:08:30,400
22% over jde wow but then just about

114
00:08:28,720 --> 00:08:34,599
everything else see either no improvement a clearly linear progression

115
00:08:32,959 --> 00:08:39,200
that is so small you might as well save your money or what could best be

116
00:08:36,399 --> 00:08:44,560
described as run to run variants one major issue to note is that we have more

117
00:08:41,599 --> 00:08:50,040
productivity results but our 6400 kit didn't manage to finish all of them so

118
00:08:47,080 --> 00:08:54,279
hence the did not finish results we do expect this to continue to improve over

119
00:08:51,880 --> 00:08:58,079
time as am5 matures as a platform and new BIOS updates roll out but it was a

120
00:08:56,200 --> 00:09:01,720
major problem for us at the time of testing and the improvements that were

121
00:09:00,360 --> 00:09:06,560
available by the time we're actually filming it have been pretty marginal so

122
00:09:04,760 --> 00:09:10,959
nobody knows exactly what the future will hold and it's still not 100% clear

123
00:09:09,040 --> 00:09:15,200
where the ddr5 bang for your buck sweet spot is going to ultimately end up I

124
00:09:13,079 --> 00:09:20,160
mean considering that ddr4 launched in 2014 and higher speed kits took years to

125
00:09:18,079 --> 00:09:24,040
become the norm it's still early days for this new generation but what is

126
00:09:22,120 --> 00:09:28,640
clear is that you don't want to buy something with significantly slower

127
00:09:26,200 --> 00:09:34,160
firstword latency if you can avoid it particularly on the lower transfer speed

128
00:09:31,079 --> 00:09:36,880
modules oh and also that there is no

129
00:09:34,160 --> 00:09:42,600
point buying faster than 6,000 or at most 6400 speed memory for AMD at least

130
00:09:40,519 --> 00:09:47,160
for this generation if you enjoyed this video we haven't tried it with ddr5 yet

131
00:09:44,880 --> 00:09:52,760
but you might want to check out our just how bad is mixing memory video it's not

132
00:09:50,360 --> 00:09:55,399
as bad as you might think or at least wasn't