1
00:00:00,000 --> 00:00:05,200
Whatever your feelings about Intel, they haven't put out any CPUs lately that have been absolute

2
00:00:05,200 --> 00:00:11,120
duds. Now, I'm talking like completely bad, but this wasn't always the case.

3
00:00:11,120 --> 00:00:16,160
As Intel has had plenty of costly mistakes in its history, so let's take a look back at

4
00:00:16,160 --> 00:00:22,000
Team Blue's Hall of Shame. First, let's go all the way back to 1981, when Intel wasn't even old

5
00:00:22,000 --> 00:00:28,320
enough to drive. They came out with a line of CPUs that used an architecture called IAPX432,

6
00:00:28,320 --> 00:00:33,280
which, aside from being annoying to say, was actually supposed to be the long-term replacement

7
00:00:33,280 --> 00:00:40,560
for x86, which had been around by then for only about three years. You see, IAPX432 was meant to

8
00:00:40,560 --> 00:00:46,400
be used with very high-level programming languages. Now, high-level here means that the language is

9
00:00:46,400 --> 00:00:51,840
far removed from the raw zeros and ones that the physical hardware the processor uses. Instead,

10
00:00:51,840 --> 00:00:56,240
a high-level language is very user-friendly, relying on lots of natural words that have to

11
00:00:56,400 --> 00:01:00,800
be translated so that the CPU can make sense of them. High-level language support would make it

12
00:01:00,800 --> 00:01:07,200
easier for developers to code more complex advanced programs, and IAPX432 was specifically

13
00:01:07,200 --> 00:01:14,640
optimized for one such language called ADA. Yes, that ADA. Intel thought ADA would end up becoming a

14
00:01:14,640 --> 00:01:19,040
much more important and popular language for the more powerful programs of the future,

15
00:01:19,040 --> 00:01:23,600
especially as it got attention from the U.S. Department of Defense for its own computer systems.

16
00:01:23,680 --> 00:01:28,800
However, ADA didn't quite take off as expected in the consumer space, and the processor itself

17
00:01:28,800 --> 00:01:33,600
simply wasn't very high performance. The reasons for this are complex, but it boils down to the

18
00:01:33,600 --> 00:01:38,640
fact that the physical processor designs of the day weren't advanced enough to run the complicated

19
00:01:38,640 --> 00:01:44,000
instructions of languages like ADA. It was a product that was just too ambitious and tried to

20
00:01:44,000 --> 00:01:48,560
include too many features at the expense of performance. So after roughly five years of

21
00:01:48,560 --> 00:01:54,640
disappointing sales, the IAPX432 project was axed while x86 continued to dominate.

22
00:01:54,640 --> 00:01:59,040
But Intel had a much higher profile chip embarrass them that you might actually remember.

23
00:01:59,040 --> 00:02:05,200
The Pentium 4 was released back in 2000 to great fanfare, and unlike the IAPX432, computers were

24
00:02:05,200 --> 00:02:09,760
mainstream enough for the general public to notice this time, especially considering there was a

25
00:02:09,760 --> 00:02:16,080
$300 million ad campaign headlined by the Blue Man Group. Intel tried to achieve never-before-seen

26
00:02:16,080 --> 00:02:20,640
levels of performance by pushing clock speeds higher and higher. In fact, they planned to scale

27
00:02:20,640 --> 00:02:26,720
speeds up to a whopping 10 gigahertz as process nodes shrunk. That's roughly double the speed of

28
00:02:26,720 --> 00:02:33,120
today's best processors. But if you know anything about CPUs, you know that clock speed isn't

29
00:02:33,120 --> 00:02:38,320
everything, and anyone who bought an Intel PC in the early 2000s learned that the hard way.

30
00:02:38,320 --> 00:02:43,680
True, these chips weren't just souped up Pentium 3s. Instead, featuring an all-new architecture

31
00:02:43,760 --> 00:02:50,720
called Netburst, which sounds hilariously close to nut bust? Oh boy. But that meant programs had to

32
00:02:50,720 --> 00:02:55,520
be specifically optimized for the new architecture to see any real performance schemes, even though

33
00:02:55,520 --> 00:03:00,800
later revisions introduced hyperthreading for the first time. A key area in which the Pentium 4

34
00:03:00,800 --> 00:03:05,920
struggled was branch prediction, which is simply the ability of the CPU to predict what the next

35
00:03:05,920 --> 00:03:10,800
instruction is going to be. Branch prediction is a crucial feature in all modern CPUs, and because

36
00:03:10,800 --> 00:03:16,000
the Pentium 4 was so bad at it, it kept having to go back and correct its own mistakes. The Pentium

37
00:03:16,000 --> 00:03:21,280
4 also had a very long pipeline, which is just what it sounds like, an electronic pipe where

38
00:03:21,280 --> 00:03:25,760
instructions are loaded one after the other so that multiple commands can be kept moving at the

39
00:03:25,760 --> 00:03:31,280
same time. The P4's lengthy pipeline would often stall out because of poor branch prediction, and

40
00:03:31,280 --> 00:03:37,200
to top it all off, the chip was very expensive and ran very hot as a result of increased power

41
00:03:37,200 --> 00:03:42,320
consumption from the high clock speeds and power leakage from the transistors. That said, Intel

42
00:03:42,320 --> 00:03:48,560
still sold a boatload of these, mostly in pre-built machines built by OEMs, so it was more of a flop

43
00:03:48,560 --> 00:03:54,560
in terms of performance and customer satisfaction than with sales. Ultimately, P4 ended up being

44
00:03:54,560 --> 00:03:59,200
the final flagship Pentium chip before the core series of CPUs took over, giving us multiple

45
00:03:59,200 --> 00:04:06,080
physical cores and a whole new architecture by 2006. But not all of Intel's big flops were CPUs.

46
00:04:06,080 --> 00:04:11,120
Although Intel's recent Arc lineup marked Team Blue's first foray into the modern discrete GPU

47
00:04:11,120 --> 00:04:16,560
scene, this actually isn't the first time they tried to make a graphics card. This is the Intel

48
00:04:16,560 --> 00:04:21,680
i7-40, and it was the first real consumer gaming graphics card that they ever made. It also had

49
00:04:21,680 --> 00:04:27,760
some incredible DNA, as the i7-40 was developed from a Lockheed Martin project that originally

50
00:04:27,760 --> 00:04:32,560
provided a visual flight simulator for astronauts in the Apollo moon landing program. Lockheed Martin

51
00:04:32,560 --> 00:04:37,360
later spun off this division as the company Real3D to try and tap into the consumer market,

52
00:04:37,360 --> 00:04:43,440
and Intel partnered with Real3D in developing the i7-40. By the time the i7-40 came out, it was

53
00:04:43,440 --> 00:04:47,760
hugely anticipated, with some in the industry saying that it would lead Intel to dominate the

54
00:04:47,760 --> 00:04:54,880
discrete GPU space. But unfortunately, Intel only sold the card for about 18 months. What really?

55
00:04:55,760 --> 00:05:01,280
The main issue is how the i7-40 made use of memory. The card connected to the motherboard

56
00:05:01,280 --> 00:05:07,040
through an AGP slot, and remember this was before PCI Express was invented, and the AGP slot provided

57
00:05:07,040 --> 00:05:12,080
a more direct connection to System RAM than the then standard conventional PCI. The idea was to

58
00:05:12,080 --> 00:05:17,040
have the game store textures in System RAM instead of the cards built in VRAM, with the AGP interface

59
00:05:17,040 --> 00:05:21,520
allowing the card to access that texture data more quickly, so you could build a card that didn't

60
00:05:21,520 --> 00:05:28,160
need as much VRAM, and was therefore cheaper. But despite this advantage, the i7-40's reliance

61
00:05:28,160 --> 00:05:33,440
on system memory made it slower than other cards that had sufficient VRAM to store textures of their

62
00:05:33,440 --> 00:05:39,360
own. And this was a time when game textures were becoming far more detailed, meaning that

63
00:05:39,360 --> 00:05:43,840
even though the i7-40 GPU itself could have actually delivered good performance,

64
00:05:43,840 --> 00:05:48,480
it just couldn't load all that texture data quickly enough. In fact, some i7-40 models came

65
00:05:48,480 --> 00:05:54,720
with as little as two megabytes of VRAM, which was mainly used for buffering, not storing textures.

66
00:05:54,720 --> 00:06:00,240
Unsurprisingly, NVIDIA's Riva TNT and 3DFX's Voodoo 2 quickly knocked the i7-40 out of the

67
00:06:00,240 --> 00:06:05,760
market, and Intel wouldn't release another discrete graphics product for 24 years. I mean,

68
00:06:05,760 --> 00:06:10,640
they tried with Larabee. I know rejection can be hard to stomach, but man, can't wait that long

69
00:06:10,640 --> 00:06:14,640
to put yourself back on the horse, you know? Thanks for watching, guys. Like, dislike, check out

70
00:06:14,640 --> 00:06:18,960
some of our other videos, comment with video suggestions down below, and don't forget to subscribe and follow.