1
00:00:00,160 --> 00:00:06,799
all other things being equal the more transistors you have on a processor the

2
00:00:04,400 --> 00:00:12,240
more features and performance you get and ever since the 1970s manufacturers

3
00:00:09,519 --> 00:00:18,000
have been finding ways to shove more and more transistors onto a single chip by

4
00:00:15,280 --> 00:00:22,160
making the transistors smaller in fact the widely quoted moore's law that

5
00:00:20,720 --> 00:00:27,199
anticipated that the number of transistors on a CPU would double

6
00:00:24,720 --> 00:00:30,880
roughly every two years held true for decades

7
00:00:28,240 --> 00:00:36,239
but in recent years moore's law has started to look more like moore's pipe

8
00:00:33,200 --> 00:00:38,960
dream as it's becoming harder and harder

9
00:00:36,239 --> 00:00:43,760
to keep shrinking transistors down due to the chemical properties of silicon

10
00:00:41,440 --> 00:00:47,840
so how do we solve this problem and keep making our computers and our gadgets

11
00:00:46,079 --> 00:00:52,960
faster and faster well although the industry has been

12
00:00:50,320 --> 00:00:58,079
looking at making smaller transistors on materials other than silicon as we

13
00:00:55,520 --> 00:01:03,199
briefly discussed in this video another idea that seems to be catching on is

14
00:01:00,320 --> 00:01:09,280
using something called chiplets now traditionally the dye or the part of

15
00:01:06,240 --> 00:01:12,159
a CPU that contains the transistors is

16
00:01:09,280 --> 00:01:17,439
manufactured as a single monolithic piece so there are practical limits to

17
00:01:15,040 --> 00:01:22,640
how big you can make them before the likelihood of a fatal defect landing on

18
00:01:19,840 --> 00:01:27,920
your die becomes too high by contrast

19
00:01:24,080 --> 00:01:31,600
chiplets are modular mini dies that can

20
00:01:27,920 --> 00:01:33,119
be combined into one CPU package to

21
00:01:31,600 --> 00:01:38,159
improve performance but i mean hold on a minute here you guys

22
00:01:36,000 --> 00:01:43,759
might say well what's the point of all this if the transistors are the same

23
00:01:40,880 --> 00:01:49,840
size anyway aren't you just adding an assembly step and making it take up more

24
00:01:46,640 --> 00:01:51,759
room well as it turns out oftentimes

25
00:01:49,840 --> 00:01:57,759
chiplet-based cpus are indeed larger think of AMD's new

26
00:01:55,360 --> 00:02:02,479
epic series of server processors as an example

27
00:01:58,880 --> 00:02:04,880
however they're also hugely advantageous

28
00:02:02,479 --> 00:02:09,200
when you consider yields so if you're not familiar with the

29
00:02:06,320 --> 00:02:15,360
concept of yields here's what it means cpus are manufactured on large silicon

30
00:02:12,319 --> 00:02:17,520
wafers and that manufacturing process

31
00:02:15,360 --> 00:02:22,879
isn't perfect which means that on a given wafer there will be some number of

32
00:02:20,080 --> 00:02:29,120
defects so if we take our wafer and divide it into smaller chiplets it means

33
00:02:25,920 --> 00:02:31,200
less wasted silicon for each defect and

34
00:02:29,120 --> 00:02:36,160
it also makes it easier to take the chiplets that are functional but that

35
00:02:33,519 --> 00:02:40,480
don't perform as well and combine those into lower end products salvaging back

36
00:02:38,879 --> 00:02:45,599
some of the cost on a large scale this leads to faster

37
00:02:42,800 --> 00:02:51,920
manufacturing and lower costs with significant real world benefits in fact

38
00:02:48,400 --> 00:02:55,440
AMD's epic cpus only cost about half as

39
00:02:51,920 --> 00:02:56,879
much thanks to chiplets versus using one

40
00:02:55,440 --> 00:03:01,120
large die now although chiplets are more common in

41
00:02:58,959 --> 00:03:04,959
the high-end desktop and server markets right now the technology will most

42
00:03:03,200 --> 00:03:10,640
likely trickle down to the consumer space such as with AMD's rumored

43
00:03:07,440 --> 00:03:13,120
upcoming 16 core ryzen CPU

44
00:03:10,640 --> 00:03:18,239
that's going to put pressure on Intel to reduce their prices which might force

45
00:03:15,920 --> 00:03:21,440
them to pursue chiplet-based designs as well

46
00:03:19,360 --> 00:03:27,200
another fun benefit is that chiplets also allow manufacturers to separate out

47
00:03:24,319 --> 00:03:31,920
parts of the CPU that don't handle pure number crunching such as input and

48
00:03:29,360 --> 00:03:36,560
output blocks as it's often harder to shrink those parts of a processor down

49
00:03:34,480 --> 00:03:41,280
so then you can just tack those bits on separately and attach

50
00:03:39,040 --> 00:03:46,159
as many chiplets as you want connecting them together with a high speed

51
00:03:43,040 --> 00:03:49,680
interface like AMD's infinity fabric now

52
00:03:46,159 --> 00:03:51,360
of course the obvious drawback of size

53
00:03:49,680 --> 00:03:56,400
still does exist but until we run up against the limits

54
00:03:54,080 --> 00:04:00,879
imposed by signals moving at the speed of light within a microprocessor

55
00:03:58,400 --> 00:04:04,720
chiplets do seem to be one of the best paths forward

56
00:04:02,480 --> 00:04:09,200
speaking of best paths forward the best path for your business might be a

57
00:04:07,040 --> 00:04:13,120
freshbooks account freshbooks is the small business accounting software

58
00:04:10,799 --> 00:04:16,320
that's custom built for how you want to work and it's a simple way whether

59
00:04:14,799 --> 00:04:21,440
you're a small business owner or a freelancer to be more productive more

60
00:04:18,720 --> 00:04:24,800
organized and to get paid faster with freshbooks you can create and send

61
00:04:23,120 --> 00:04:28,720
professional looking invoices in less than 30 seconds you can set up online

62
00:04:26,880 --> 00:04:32,800
payments with just a couple of clicks and get paid up to four days faster you

63
00:04:30,960 --> 00:04:36,720
can see when your client has seen your invoice and put an end to the guessing

64
00:04:34,960 --> 00:04:41,600
games and you can take the whole experience with you on the go with their

65
00:04:38,400 --> 00:04:43,840
fully featured apps for iOS and Android

66
00:04:41,600 --> 00:04:48,800
so for an unrestricted 30 day free trial go to freshbooks.com techwiki we'll have

67
00:04:46,639 --> 00:04:51,440
that linked below and just enter Techquickie in the how did you hear about us

68
00:04:50,479 --> 00:04:55,520
section so thanks for watching guys like dislike

69
00:04:53,919 --> 00:04:59,280
check out our other videos leave a comment if you have a suggestion for a

70
00:04:57,040 --> 00:05:03,919
future fast as possible and make sure that you subscribe so you don't miss any

71
00:05:01,600 --> 00:05:08,800
future tech quickies or even the past ones don't miss those

72
00:05:06,320 --> 00:05:08,800
either