1
00:00:00,720 --> 00:00:02,840
Speaker 1: And you know a question and often here's why would

2
00:00:02,840 --> 00:00:05,600
Speaker 1: anybody need to compute that fast? There's lots of problems

3
00:00:05,640 --> 00:00:07,440
Speaker 1: where you have to compute that fast. If you ever

4
00:00:07,480 --> 00:00:14,000
Speaker 1: want to get an answer from Bloomberg News and iHeartRadio,

5
00:00:14,360 --> 00:00:20,880
Speaker 1: it's the big take. I'm West Cassova today a trip

6
00:00:20,960 --> 00:00:33,880
Speaker 1: to see the world's fastest computer. We did a whole

7
00:00:33,960 --> 00:00:36,120
Speaker 1: lot of COVID work, how the virus makes people are

8
00:00:36,200 --> 00:00:40,400
Speaker 1: really really sick, looking at solving multiphysics problems related to

9
00:00:40,560 --> 00:00:44,640
Speaker 1: the modeling and simulation of nuclear reactors, much more complicated

10
00:00:44,880 --> 00:00:48,280
Speaker 1: mechanistic models of what's going on in opioord's addiction. One

11
00:00:48,400 --> 00:00:50,839
Speaker 1: problem that requires a lot of flops, a lot of

12
00:00:50,840 --> 00:00:53,880
Speaker 1: floating point operations per second. Is climate modeling or weather

13
00:00:53,960 --> 00:00:58,040
Speaker 1: model really trying to crack hard problems? And so we

14
00:00:58,080 --> 00:01:02,800
Speaker 1: need this combination of clever rhythm designed with exoscale computing

15
00:01:03,080 --> 00:01:08,880
Speaker 1: to start to solve the challenges and biology and ecosystems. Hey, Vicky,

16
00:01:09,160 --> 00:01:13,560
Speaker 1: Hey Catherine, Hey, how's it going. Hello Wes, So what

17
00:01:13,600 --> 00:01:16,479
Speaker 1: are we talking about today? Well, I came across these

18
00:01:16,480 --> 00:01:19,880
Speaker 1: incredible photos in BusinessWeek a bit ago, and they were

19
00:01:19,920 --> 00:01:24,000
Speaker 1: reporting on the world's largest supercomputer in fact, it's linked

20
00:01:24,040 --> 00:01:26,399
Speaker 1: to our write up today. So the piece talked about

21
00:01:26,400 --> 00:01:29,560
Speaker 1: the sheer computing power of this mega computer and it

22
00:01:29,680 --> 00:01:33,319
Speaker 1: breaking something called the exoscale barrier. We're going to get

23
00:01:33,319 --> 00:01:35,800
Speaker 1: into that. But my eyes popped and I took it

24
00:01:35,840 --> 00:01:38,200
Speaker 1: to you and Catherine. Remember, oh, yes, I remember this.

25
00:01:38,240 --> 00:01:41,040
Speaker 1: You were very excited, and so where were we and

26
00:01:41,120 --> 00:01:45,160
Speaker 1: so we all decided that you and Catherine needed to

27
00:01:45,200 --> 00:01:48,680
Speaker 1: take a road trip, and we did so. This computer

28
00:01:48,880 --> 00:01:51,520
Speaker 1: is called Frontier. It was built at the Department of

29
00:01:51,640 --> 00:01:56,160
Speaker 1: Energies oak Ridge National Lab, which is just outside Knoxville, Tennessee.

30
00:01:56,560 --> 00:01:59,400
Speaker 1: We reached out to them and they said, come check

31
00:01:59,400 --> 00:02:01,800
Speaker 1: it out. So we packed up our audio gear and

32
00:02:01,880 --> 00:02:04,120
Speaker 1: we went and we got to speak to some of

33
00:02:04,120 --> 00:02:08,160
Speaker 1: the engineers and scientists who are using the supercomputer right

34
00:02:08,200 --> 00:02:12,880
Speaker 1: now for a myriad of projects. Yeah, and we really

35
00:02:12,919 --> 00:02:17,640
Speaker 1: can't stress enough how incredibly powerful this thing is. They're

36
00:02:17,720 --> 00:02:21,440
Speaker 1: looking at questions that they were unable to answer ever before,

37
00:02:21,480 --> 00:02:24,919
Speaker 1: like the origins of the universe, like potential cures for disease,

38
00:02:25,280 --> 00:02:28,920
Speaker 1: and these kinds of questions actually really need a massive

39
00:02:28,919 --> 00:02:31,120
Speaker 1: amount of computing power, and they haven't been able to

40
00:02:31,120 --> 00:02:33,640
Speaker 1: do it up to now. Exoscale is like a next level.

41
00:02:34,240 --> 00:02:36,960
Speaker 1: So scientists are lining up out the door to use

42
00:02:37,000 --> 00:02:38,919
Speaker 1: this thing. So we went down and we took a tour.

43
00:02:39,400 --> 00:02:42,920
Speaker 1: I'm justin Wit and I am the program director for

44
00:02:43,160 --> 00:02:47,320
Speaker 1: the Leadership Computing Facility here at a Christian National Lab.

45
00:02:47,560 --> 00:02:51,360
Speaker 1: We're housing some of the fastest and most powerful supercomputers

46
00:02:51,360 --> 00:02:57,560
Speaker 1: in the world. We currently are deploying the world's fastest computer,

47
00:02:57,720 --> 00:03:01,640
Speaker 1: the Frontier system, and we all have just down the

48
00:03:01,639 --> 00:03:05,440
Speaker 1: hall the Summit Supercomputer, which is at number five currently,

49
00:03:05,440 --> 00:03:09,280
Speaker 1: I believe. All right, so we've established that this thing

50
00:03:09,560 --> 00:03:12,639
Speaker 1: is fast. But what is fast? Like? How fast are

51
00:03:12,639 --> 00:03:15,480
Speaker 1: we talking about? Yeah, we had the same question, and

52
00:03:15,520 --> 00:03:17,920
Speaker 1: so we put it to him. Can you describe it

53
00:03:17,960 --> 00:03:20,600
Speaker 1: a little bit like what has gone into it? How

54
00:03:20,639 --> 00:03:24,000
Speaker 1: big is it? How fast is it? Yeah? Sure, So

55
00:03:24,200 --> 00:03:26,520
Speaker 1: we'll start with fast. You know, that's what Frontier is

56
00:03:26,560 --> 00:03:28,760
Speaker 1: known for as being the fastest computer in the world

57
00:03:28,800 --> 00:03:32,120
Speaker 1: at this point, and it's capable of about one point

58
00:03:32,160 --> 00:03:37,280
Speaker 1: six quintillion calculations a second. That's one point six times

59
00:03:37,320 --> 00:03:40,360
Speaker 1: with eighteen zeros after it. When numbers get that big,

60
00:03:40,400 --> 00:03:43,080
Speaker 1: that they start to lose meaning, right, you can't your

61
00:03:43,120 --> 00:03:45,440
Speaker 1: mind can't grasp that, and one of the things we

62
00:03:45,520 --> 00:03:48,400
Speaker 1: like to compare it to is if every person on

63
00:03:48,440 --> 00:03:52,320
Speaker 1: the planet could do one calculation per second, it would

64
00:03:52,320 --> 00:03:56,080
Speaker 1: take four years for them to do all the calculations

65
00:03:56,120 --> 00:04:00,840
Speaker 1: that Frontier does every second of the day. I think

66
00:04:00,880 --> 00:04:04,440
Speaker 1: the engineering that went into the room itself is pretty astounding.

67
00:04:04,840 --> 00:04:09,720
Speaker 1: For instance, there is so much computing capacity inside each

68
00:04:09,840 --> 00:04:12,840
Speaker 1: cabinet of Frontier. And when I say cabinet, they're about

69
00:04:12,920 --> 00:04:16,520
Speaker 1: seventy four of these refrigerator sized objects, and each of

70
00:04:16,560 --> 00:04:20,240
Speaker 1: those makes up part of Frontier, and they're packed full

71
00:04:20,720 --> 00:04:24,599
Speaker 1: of computing hardware, and each of those is very heavy.

72
00:04:24,800 --> 00:04:28,040
Speaker 1: At this density, each cabinet weighs about as much as

73
00:04:28,080 --> 00:04:31,640
Speaker 1: two f one fifty pickups And we have seventy four

74
00:04:31,680 --> 00:04:34,720
Speaker 1: of those, plus supporting infrastructure, plus the file system and

75
00:04:34,839 --> 00:04:39,039
Speaker 1: all we're supporting about two Bowling seven forty seven's on

76
00:04:39,160 --> 00:04:41,600
Speaker 1: our raised data center floor at this point, so that

77
00:04:42,000 --> 00:04:44,960
Speaker 1: there's a lot of engineering that went into that to

78
00:04:45,000 --> 00:04:48,479
Speaker 1: make that happen. That image of each of those seventy

79
00:04:48,480 --> 00:04:51,919
Speaker 1: four cabinets weighing as much as two f one fifty

80
00:04:51,960 --> 00:04:54,960
Speaker 1: pickup trucks is like sticking in my head. How do

81
00:04:55,000 --> 00:04:59,680
Speaker 1: you power something that big? This is a really complicated system,

82
00:04:59,720 --> 00:05:03,280
Speaker 1: so I'm going to let the experts explain. So this part,

83
00:05:03,279 --> 00:05:05,880
Speaker 1: I think is a little likely. Behind the scenes store

84
00:05:05,880 --> 00:05:08,760
Speaker 1: of Disney, a lot of people think, oh, do you

85
00:05:08,760 --> 00:05:11,120
Speaker 1: you know I like at home, I order a computer,

86
00:05:11,240 --> 00:05:13,919
Speaker 1: it comes, we plug it into the wall. But with

87
00:05:14,040 --> 00:05:17,640
Speaker 1: computers of the size. For instance, we installed about forty

88
00:05:17,680 --> 00:05:22,160
Speaker 1: megawatts of power to run this computers, and it's kind

89
00:05:22,160 --> 00:05:25,240
Speaker 1: of peak operating. It would use about thirty megawatts of

90
00:05:25,320 --> 00:05:28,400
Speaker 1: that and that's enough to power about twenty five thousand homes.

91
00:05:30,360 --> 00:05:33,800
Speaker 1: And what you hear are actually these very large pumps

92
00:05:34,160 --> 00:05:37,479
Speaker 1: pumping that six thousand gallons of water to the system

93
00:05:37,560 --> 00:05:43,400
Speaker 1: every minute. I just got to cut in here and say,

94
00:05:43,480 --> 00:05:47,040
Speaker 1: that's six thousand gallons. He's talking about one. They're in

95
00:05:47,080 --> 00:05:49,800
Speaker 1: a closed loop system, so it's the same water being

96
00:05:49,839 --> 00:05:52,480
Speaker 1: repurposed over and over again. But they are flowing through

97
00:05:52,680 --> 00:05:55,839
Speaker 1: thirty six inch pipes. We're talking like a yard across.

98
00:05:58,160 --> 00:06:01,400
Speaker 1: Tell us about that the six thousand gallons. That is

99
00:06:01,480 --> 00:06:04,120
Speaker 1: just to keep it cool. That is to keep it cool.

100
00:06:04,360 --> 00:06:06,320
Speaker 1: If you put that much power in, you've got to

101
00:06:06,320 --> 00:06:09,120
Speaker 1: get the heat back out, and in those small spaces,

102
00:06:09,200 --> 00:06:11,880
Speaker 1: really the only way to do that is by circulating

103
00:06:11,920 --> 00:06:15,000
Speaker 1: some fluid directly over each of the components on the

104
00:06:15,040 --> 00:06:18,800
Speaker 1: computer to have your own sort of miniature energy grid here,

105
00:06:18,880 --> 00:06:21,760
Speaker 1: or how do you power this? So we're actually power

106
00:06:21,880 --> 00:06:25,440
Speaker 1: from separate substations here, so if one goes down, we

107
00:06:25,520 --> 00:06:29,320
Speaker 1: still get power to the computer and the mechanical plants.

108
00:06:30,200 --> 00:06:33,159
Speaker 1: You'd think that the world's fastest supercomputer would be I

109
00:06:33,160 --> 00:06:35,720
Speaker 1: don't know, immortal, or at least have a long lifespan,

110
00:06:35,839 --> 00:06:37,800
Speaker 1: but we found out that it's actually not all that

111
00:06:37,960 --> 00:06:40,520
Speaker 1: different from the computers we use. So what we like

112
00:06:40,640 --> 00:06:43,920
Speaker 1: to do is run the computer's in production with scientific

113
00:06:44,080 --> 00:06:47,080
Speaker 1: users on the systems doing their work for you know,

114
00:06:47,120 --> 00:06:49,920
Speaker 1: at least five years, and then we like to overlap

115
00:06:49,920 --> 00:06:52,400
Speaker 1: a year with the next computer, so you know, we

116
00:06:52,520 --> 00:06:54,880
Speaker 1: kind of say five to seven years. It's kind of

117
00:06:54,920 --> 00:06:58,120
Speaker 1: the general timeframe we'd like to keep these around. So

118
00:06:58,160 --> 00:07:00,240
Speaker 1: while we were on the tour, there was one point

119
00:07:00,240 --> 00:07:03,560
Speaker 1: where a technician had one of these huge cabinets open

120
00:07:03,600 --> 00:07:05,600
Speaker 1: and he was sort of messing with the guts of it.

121
00:07:05,760 --> 00:07:07,440
Speaker 1: So we got to see inside of it and we

122
00:07:07,480 --> 00:07:11,120
Speaker 1: see all these microtrips and copper and wiring, and it

123
00:07:11,240 --> 00:07:13,920
Speaker 1: sort of got us thinking how do you build this thing.

124
00:07:14,200 --> 00:07:16,960
Speaker 1: We've all experienced the supply chain issues we've had over

125
00:07:17,000 --> 00:07:19,760
Speaker 1: the last couple of years, especially due to COVID and

126
00:07:19,960 --> 00:07:22,360
Speaker 1: building a supercomputer. It doesn't matter if you're the Department

127
00:07:22,360 --> 00:07:24,360
Speaker 1: of Energy, you're going to run into the same supply issues.

128
00:07:24,360 --> 00:07:26,040
Speaker 1: So we asked them how it was for them. We

129
00:07:26,240 --> 00:07:28,880
Speaker 1: jokingly say, one of our biggest lessons learned is don't

130
00:07:28,880 --> 00:07:32,840
Speaker 1: build a supercomputer during a pandemic. So the supply chain

131
00:07:33,000 --> 00:07:38,320
Speaker 1: issues were non trivial and we're hard to overcome. Frontier

132
00:07:38,400 --> 00:07:42,560
Speaker 1: has sixty million individual parts in it, you know that

133
00:07:43,040 --> 00:07:49,080
Speaker 1: balls down to hundreds of individual part numbers, and the

134
00:07:49,120 --> 00:07:52,520
Speaker 1: supply chain issues were across the board. I guess it

135
00:07:52,600 --> 00:07:56,680
Speaker 1: was early twenty twenty we got word from Hewlett Packard

136
00:07:56,800 --> 00:08:00,840
Speaker 1: Enterprises that their suppliers were saying, this could be a

137
00:08:00,880 --> 00:08:03,760
Speaker 1: two year delay. We cannot get chips to do that

138
00:08:04,560 --> 00:08:07,800
Speaker 1: for us. We had some supply chain issues there, but

139
00:08:07,840 --> 00:08:11,440
Speaker 1: it was even the fifty cent parks, the little voltage

140
00:08:11,480 --> 00:08:13,760
Speaker 1: regulators and things that are in here. And at one

141
00:08:13,760 --> 00:08:17,040
Speaker 1: point we had eighteen people that their full time jobs

142
00:08:17,360 --> 00:08:20,000
Speaker 1: were every day going out and finding the parts to

143
00:08:20,040 --> 00:08:23,200
Speaker 1: build Frontier, and so we shrunk that two year delay

144
00:08:23,520 --> 00:08:26,080
Speaker 1: down to less than a three month delay. We're able

145
00:08:26,120 --> 00:08:29,720
Speaker 1: to mostly hold our schedule for this. But again, so

146
00:08:30,120 --> 00:08:33,439
Speaker 1: they got it built and my next question was what's

147
00:08:33,480 --> 00:08:36,840
Speaker 1: it like to use it? I asked Bronson Messer, he

148
00:08:37,000 --> 00:08:40,360
Speaker 1: was another Oakridge scientist who was also along for the tour. Yeah.

149
00:08:40,360 --> 00:08:42,840
Speaker 1: For the most part, you set at your own workstation

150
00:08:43,320 --> 00:08:47,640
Speaker 1: and you log into Frontier remotely. Almost all of our users,

151
00:08:47,440 --> 00:08:51,480
Speaker 1: with some notable exceptions, are not local. In fact, we

152
00:08:51,520 --> 00:08:54,360
Speaker 1: have a lot of international users. Users all over the

153
00:08:54,400 --> 00:08:58,120
Speaker 1: country everywhere. I have students that, for example, don't interact

154
00:08:58,120 --> 00:09:00,679
Speaker 1: with the computer any other way except something called a

155
00:09:00,760 --> 00:09:04,840
Speaker 1: Jupiter notebook, which is actually a web interface to the computer,

156
00:09:05,880 --> 00:09:09,520
Speaker 1: and it's sort of everything in between. More from oak

157
00:09:09,640 --> 00:09:19,800
Speaker 1: Ridge National Lab when we come back, Vicky, you mentioned

158
00:09:19,800 --> 00:09:23,160
Speaker 1: earlier that scientists are lining up to use the saying

159
00:09:23,160 --> 00:09:26,480
Speaker 1: you can see why. How do they decide though, which

160
00:09:26,480 --> 00:09:30,480
Speaker 1: scientists get time on this computer? And then once they

161
00:09:30,520 --> 00:09:32,000
Speaker 1: get the time, what do they actually use it for.

162
00:09:32,440 --> 00:09:36,120
Speaker 1: Remember Bronson Messer from earlier, he is actually the director

163
00:09:36,160 --> 00:09:39,320
Speaker 1: of science for the oak Ridge Leadership Computing Facility. That's

164
00:09:39,320 --> 00:09:42,760
Speaker 1: where we were within this massive complex, and we had

165
00:09:42,760 --> 00:09:45,960
Speaker 1: a chance to sit down and ask him. You are

166
00:09:46,200 --> 00:09:48,560
Speaker 1: the go guy, right, You're the guy who says yes

167
00:09:48,640 --> 00:09:50,480
Speaker 1: or no to the project. So how do you vet these?

168
00:09:51,600 --> 00:09:54,840
Speaker 1: Like all peer reviewed science, we use peer review pounds, right,

169
00:09:54,880 --> 00:10:01,679
Speaker 1: So we have several different competitive allocation programs for the machine.

170
00:10:01,559 --> 00:10:04,400
Speaker 1: You know, some of them are open to everybody, some

171
00:10:04,480 --> 00:10:08,280
Speaker 1: are open mostly to DOE researchers. We also reserve a

172
00:10:08,320 --> 00:10:10,480
Speaker 1: small amount of time for what we call our director's

173
00:10:10,520 --> 00:10:15,080
Speaker 1: discretionary program. But for all these programs, we impanel other

174
00:10:15,160 --> 00:10:18,520
Speaker 1: scientists to sort of decide, yeah, this is an important

175
00:10:18,520 --> 00:10:21,200
Speaker 1: scientific problem and we need to use the resources to

176
00:10:21,200 --> 00:10:24,719
Speaker 1: do this first, or you know, maybe this is important too,

177
00:10:24,720 --> 00:10:26,880
Speaker 1: but maybe we should wait just a minute. So those

178
00:10:26,920 --> 00:10:30,839
Speaker 1: same proposals get checked out. And basically the question is

179
00:10:31,080 --> 00:10:33,480
Speaker 1: can they use the big computer and do they need

180
00:10:33,520 --> 00:10:35,880
Speaker 1: to use the big computer? And if you can't answer

181
00:10:35,920 --> 00:10:39,640
Speaker 1: both of those without resounding yes, you probably, frankly, you

182
00:10:39,640 --> 00:10:41,439
Speaker 1: probably don't want to try to compute on our big

183
00:10:41,480 --> 00:10:45,040
Speaker 1: computer because it's it's hard for a variety of reasons.

184
00:10:45,440 --> 00:10:47,600
Speaker 1: Inventing all of these there has to be a certain

185
00:10:47,600 --> 00:10:50,600
Speaker 1: amount of national security implications. We don't do any classified

186
00:10:50,640 --> 00:10:53,440
Speaker 1: work here, so we're strictly an open science shop that

187
00:10:53,480 --> 00:10:55,760
Speaker 1: it will see if so, tell me what that means.

188
00:10:55,880 --> 00:10:58,240
Speaker 1: It means that all the all the science that we do,

189
00:10:58,640 --> 00:11:00,640
Speaker 1: basically all the inputs and all the outputs can be

190
00:11:00,640 --> 00:11:03,680
Speaker 1: published in the open literature and be promulgated to anybody.

191
00:11:03,800 --> 00:11:07,199
Speaker 1: So in a journal publication, we do have the ability

192
00:11:07,240 --> 00:11:09,120
Speaker 1: to do things that are a little bit more sensitive

193
00:11:09,120 --> 00:11:13,000
Speaker 1: than just strictly open science, things like protected health information

194
00:11:13,120 --> 00:11:16,280
Speaker 1: or other things like that. But we take that security

195
00:11:16,320 --> 00:11:18,439
Speaker 1: measures really seriously. But what it boils down to is

196
00:11:18,480 --> 00:11:20,720
Speaker 1: we do what's called an exploit control review on every

197
00:11:20,720 --> 00:11:23,480
Speaker 1: single project that gets on the machine. We make sure

198
00:11:23,880 --> 00:11:27,360
Speaker 1: that it's possible that these the results can be can

199
00:11:27,400 --> 00:11:32,640
Speaker 1: be promulgated to everybody. How concerned are you all about

200
00:11:33,440 --> 00:11:36,839
Speaker 1: breach of security or hackers and what kinds of safeguards

201
00:11:36,840 --> 00:11:39,600
Speaker 1: do you have against that. We have a huge cybersecurity

202
00:11:39,600 --> 00:11:41,800
Speaker 1: cred So if you've met, if you fielded the world's

203
00:11:41,880 --> 00:11:45,240
Speaker 1: largest computer for at this point more than a decade,

204
00:11:45,840 --> 00:11:48,960
Speaker 1: we have script kitties knocking at the door every five

205
00:11:49,080 --> 00:11:52,520
Speaker 1: or six seconds. What's a script kitty? Somebody who wants

206
00:11:52,559 --> 00:11:55,720
Speaker 1: to break into the computer for nefarious purposes, Either either

207
00:11:56,000 --> 00:12:00,240
Speaker 1: organized nefarious purposes or just hey, I has hacked into

208
00:12:00,320 --> 00:12:04,400
Speaker 1: the world's largest supercomputer kind of right, Yeah, so we

209
00:12:04,040 --> 00:12:07,760
Speaker 1: have we have huge effort and intrusion detection making sure

210
00:12:07,800 --> 00:12:12,080
Speaker 1: that the edge of our computer is protected from that

211
00:12:12,160 --> 00:12:14,480
Speaker 1: kind of thing. Um. And just to back up a

212
00:12:14,480 --> 00:12:17,200
Speaker 1: little bit, why is it called the Frontier? Yeah, I

213
00:12:17,240 --> 00:12:19,920
Speaker 1: think it really I think Frontiers an optimist named for

214
00:12:19,960 --> 00:12:23,840
Speaker 1: the machine. It's it's the culmination of a more than

215
00:12:23,960 --> 00:12:27,440
Speaker 1: decade long project to get us to the exo scale.

216
00:12:27,920 --> 00:12:30,880
Speaker 1: More than a decade ago, the federal government basically made

217
00:12:30,880 --> 00:12:33,199
Speaker 1: a commitment to getting us to the exo scale as

218
00:12:33,240 --> 00:12:37,160
Speaker 1: as a concrete sort of flag that we could When

219
00:12:37,200 --> 00:12:39,040
Speaker 1: you say the federal government, is that Congress was it

220
00:12:39,240 --> 00:12:42,360
Speaker 1: legislated everybody. It was a whole of government effort, right,

221
00:12:42,480 --> 00:12:44,920
Speaker 1: And of course it took getting Congress behind it because

222
00:12:44,960 --> 00:12:47,600
Speaker 1: they controlled the first strings and it costs money. And

223
00:12:47,640 --> 00:12:50,240
Speaker 1: so there's really two major components of getting us here.

224
00:12:50,320 --> 00:12:53,320
Speaker 1: There's this machine and other machines like it, but this

225
00:12:53,400 --> 00:12:56,640
Speaker 1: machine first, and then alongside of that there's there was

226
00:12:56,640 --> 00:12:59,480
Speaker 1: something called the Exo Scale Computing Project, which was really

227
00:12:59,480 --> 00:13:03,240
Speaker 1: concerned with making sure that we had software and codes

228
00:13:03,240 --> 00:13:06,160
Speaker 1: that could actually take advantage of the machines once they

229
00:13:06,160 --> 00:13:08,800
Speaker 1: got here. That was a big worry, say seven years ago,

230
00:13:09,000 --> 00:13:11,520
Speaker 1: that we would build it and nobody would come and

231
00:13:11,600 --> 00:13:14,800
Speaker 1: nobody could come right because they wouldn't be ready. One

232
00:13:15,280 --> 00:13:17,719
Speaker 1: problem that requires a lot of flops, a lot of

233
00:13:17,760 --> 00:13:21,920
Speaker 1: floating point operations per seconds. Climate modeling or weather modeling. Now,

234
00:13:21,920 --> 00:13:24,440
Speaker 1: the thing about weather modeling is you kind of have

235
00:13:24,520 --> 00:13:27,559
Speaker 1: to get an answer before the weather actually happens, or

236
00:13:27,600 --> 00:13:31,560
Speaker 1: it's not super useful, right, and so weather codes typically

237
00:13:31,559 --> 00:13:33,440
Speaker 1: want to be able to get answers within just a

238
00:13:33,440 --> 00:13:35,959
Speaker 1: few hours or maybe even sub hour in some cases.

239
00:13:36,760 --> 00:13:38,560
Speaker 1: To be able to do that and to have the

240
00:13:38,720 --> 00:13:41,000
Speaker 1: same the physical fidelity, it's going to tell you it's

241
00:13:41,040 --> 00:13:42,880
Speaker 1: going to rain here and it's not going to rain there,

242
00:13:43,120 --> 00:13:44,800
Speaker 1: or you're going to have a windstorm here, you're not

243
00:13:44,840 --> 00:13:46,720
Speaker 1: going to have a windstorm there, when you're talking about

244
00:13:47,000 --> 00:13:51,280
Speaker 1: a few miles apart. Only at the excess scale, we're

245
00:13:51,320 --> 00:13:55,200
Speaker 1: going to be able to do that with regularity. Climate

246
00:13:55,240 --> 00:13:57,520
Speaker 1: models are going to have to be at that sort

247
00:13:57,559 --> 00:14:01,200
Speaker 1: of that level. AI and machine learning my help, because

248
00:14:01,640 --> 00:14:03,200
Speaker 1: you know, we all have a feel for what the

249
00:14:03,240 --> 00:14:05,960
Speaker 1: weather is like, right, So guess what you can train

250
00:14:06,040 --> 00:14:08,160
Speaker 1: machines to sort of have a feel for what it's

251
00:14:08,200 --> 00:14:10,760
Speaker 1: like as well. But you're having to stop right there

252
00:14:10,760 --> 00:14:13,880
Speaker 1: because you said the machine will have a feel And

253
00:14:14,160 --> 00:14:16,240
Speaker 1: this is something that I think we are trying to

254
00:14:16,280 --> 00:14:19,640
Speaker 1: wrap our heads around. These are human words, they are

255
00:14:19,920 --> 00:14:21,920
Speaker 1: so how do we get a feel when when when

256
00:14:21,960 --> 00:14:23,880
Speaker 1: people say I have a feel for what that is,

257
00:14:23,960 --> 00:14:26,720
Speaker 1: how do you get a feel? Well, it's through experience.

258
00:14:26,880 --> 00:14:29,920
Speaker 1: It's through seeing the same thing over and over with

259
00:14:30,120 --> 00:14:34,960
Speaker 1: slight variations and being able to predict with some amount

260
00:14:34,960 --> 00:14:37,680
Speaker 1: of fidelity, what's what's going to happen again? Is it

261
00:14:37,680 --> 00:14:39,680
Speaker 1: gonna happen again exactly? Like that? If if I right,

262
00:14:39,840 --> 00:14:43,080
Speaker 1: this guy is gray, it might rain today exactly, or

263
00:14:43,200 --> 00:14:45,200
Speaker 1: or if it's it's clear blue sky, it's not going

264
00:14:45,240 --> 00:14:47,120
Speaker 1: to rain today. I have a feel for that. Even

265
00:14:47,280 --> 00:14:50,040
Speaker 1: something as simple as shooting a basketball right if I

266
00:14:50,160 --> 00:14:52,440
Speaker 1: sort of if I sort of cork it this way,

267
00:14:52,480 --> 00:14:54,320
Speaker 1: I'm gonna miss it. If I don't, If I feel

268
00:14:54,360 --> 00:14:55,920
Speaker 1: it go out of my hand the right way, it's

269
00:14:55,920 --> 00:14:58,720
Speaker 1: probably going to make it. In feeling like that, that's

270
00:14:58,920 --> 00:15:01,320
Speaker 1: very much the kind the thing that happens with AI

271
00:15:01,360 --> 00:15:06,160
Speaker 1: and mL. You basically train mL and AI tools mL,

272
00:15:06,240 --> 00:15:09,520
Speaker 1: machine learning, machine learning on lots and lots of data,

273
00:15:09,720 --> 00:15:14,640
Speaker 1: so lots and lots of experiences, and then you use

274
00:15:14,720 --> 00:15:18,560
Speaker 1: that to anticipate if you see a similar data set

275
00:15:18,560 --> 00:15:22,920
Speaker 1: in the future, what might happen. Does Frontier understand emotion? No,

276
00:15:23,400 --> 00:15:28,080
Speaker 1: Frontier does not understand emotion. It doesn't understand tone and

277
00:15:28,480 --> 00:15:30,400
Speaker 1: is all it knows is sort of what it's been fed.

278
00:15:30,680 --> 00:15:35,320
Speaker 1: It can fake understanding emotion. Right. So you know, a

279
00:15:35,320 --> 00:15:39,240
Speaker 1: lot of the AI and mL tools feel like human right.

280
00:15:39,520 --> 00:15:42,240
Speaker 1: Chat GPT is a prime example of this ring as

281
00:15:42,760 --> 00:15:46,160
Speaker 1: they feel rather human, but they're not right. They're they're

282
00:15:46,200 --> 00:15:50,280
Speaker 1: not formulating original fault. They're using a backlog. They're making

283
00:15:50,360 --> 00:15:53,920
Speaker 1: new things, right, but out of a limited set of things.

284
00:15:54,640 --> 00:15:57,040
Speaker 1: And if you steer them right, they'll just copy what

285
00:15:57,120 --> 00:15:59,440
Speaker 1: they've learned before. I understand, And so does that all

286
00:15:59,600 --> 00:16:04,320
Speaker 1: get stored? That memory learning gets stored, and then it

287
00:16:04,400 --> 00:16:08,040
Speaker 1: can be basically resurfaced as needed, like right, and so

288
00:16:08,080 --> 00:16:12,040
Speaker 1: the basically there's two sides to the artificial intelligence and

289
00:16:12,160 --> 00:16:15,960
Speaker 1: machine learning coin. One is training. It turns out that

290
00:16:16,000 --> 00:16:19,560
Speaker 1: machine big, huge machines like Frontier really good at training, right,

291
00:16:19,600 --> 00:16:21,600
Speaker 1: because you need a lot of memory to store all

292
00:16:21,600 --> 00:16:25,560
Speaker 1: those data right to be able to train an AI

293
00:16:25,680 --> 00:16:28,080
Speaker 1: model on, and you need to be able to do

294
00:16:28,080 --> 00:16:31,440
Speaker 1: it fast, so huge machine like Frontier is really good

295
00:16:31,440 --> 00:16:33,760
Speaker 1: at that. Then you get a model, you get a

296
00:16:33,800 --> 00:16:36,400
Speaker 1: reduced sort of set of things that you can use,

297
00:16:37,040 --> 00:16:39,440
Speaker 1: and you want to do inferns, so you want to

298
00:16:39,440 --> 00:16:41,920
Speaker 1: feed in some new inputs and get the machine learning

299
00:16:41,920 --> 00:16:44,000
Speaker 1: model to tell you what the output ought to be.

300
00:16:44,440 --> 00:16:48,160
Speaker 1: Those are much smaller. They typically fit like on one

301
00:16:48,160 --> 00:16:50,960
Speaker 1: of those nodes you saw earlier inside the data center,

302
00:16:51,080 --> 00:16:53,880
Speaker 1: fairly effectively. But the key is you probably want to

303
00:16:54,000 --> 00:16:57,400
Speaker 1: use them everywhere all over the computer. You probably want

304
00:16:57,400 --> 00:17:01,840
Speaker 1: to use it and lots and lots of place. This

305
00:17:02,000 --> 00:17:04,159
Speaker 1: was another question for my eleven year old. How do

306
00:17:04,160 --> 00:17:06,720
Speaker 1: you ask it? A question? The question is usually of

307
00:17:06,760 --> 00:17:11,399
Speaker 1: the form if I do a what will happen? So

308
00:17:11,480 --> 00:17:15,399
Speaker 1: like an equation almost almost, but you want to know if, basically,

309
00:17:15,400 --> 00:17:19,040
Speaker 1: if I change something, what's going to happen, because that's

310
00:17:19,040 --> 00:17:21,560
Speaker 1: the way I understand how something works. So, to go

311
00:17:21,560 --> 00:17:24,880
Speaker 1: back to my basketball analogy, if I talk my old bow,

312
00:17:24,920 --> 00:17:27,200
Speaker 1: I out or keep it in, what is I actually

313
00:17:27,240 --> 00:17:28,880
Speaker 1: going to do to the fly to the ball. That's

314
00:17:28,920 --> 00:17:31,639
Speaker 1: typically very much how you ask the computer a question.

315
00:17:31,720 --> 00:17:35,600
Speaker 1: You do several simulations where you've tweaked one or two

316
00:17:35,640 --> 00:17:37,800
Speaker 1: little things as you go along, or try to see

317
00:17:37,840 --> 00:17:43,040
Speaker 1: what the answer actually is, what this big computer is

318
00:17:43,280 --> 00:17:53,560
Speaker 1: actually used for. After the break, Vicky, now we know

319
00:17:53,640 --> 00:17:56,359
Speaker 1: how you sit down and use this computer. So I

320
00:17:56,359 --> 00:17:59,159
Speaker 1: wanted to ask you one more question. Well wait, before

321
00:17:59,160 --> 00:18:02,359
Speaker 1: you do, us have to say they really buried the

322
00:18:02,440 --> 00:18:04,879
Speaker 1: lead on us. At the end of the tour, we

323
00:18:04,920 --> 00:18:06,960
Speaker 1: stopped by an old building in an area that seemed

324
00:18:07,000 --> 00:18:09,240
Speaker 1: really decrepit, and it was on the verge of demolition.

325
00:18:09,280 --> 00:18:12,359
Speaker 1: Almost We walked into that building and we came to

326
00:18:12,440 --> 00:18:15,760
Speaker 1: understand it was actually the site of the Manhattan Project.

327
00:18:16,680 --> 00:18:19,840
Speaker 1: The reason they built this here was because in around

328
00:18:19,880 --> 00:18:24,600
Speaker 1: nineteen thirty eight in Germany, vision was discovered, and so

329
00:18:24,680 --> 00:18:28,480
Speaker 1: they convinced Albert Einstein to write a letter at President

330
00:18:28,560 --> 00:18:31,720
Speaker 1: Roosevelt saying, we need a program because we don't want

331
00:18:31,720 --> 00:18:35,520
Speaker 1: the Nazi ste it there before we did, and so

332
00:18:35,920 --> 00:18:38,520
Speaker 1: they began a lot of work on what became known

333
00:18:38,560 --> 00:18:41,960
Speaker 1: as the Manhattan Project. This part of it was to

334
00:18:42,560 --> 00:18:47,159
Speaker 1: pilot the demonstration of plutonium made in a reactor. That

335
00:18:47,280 --> 00:18:50,720
Speaker 1: was Bill Cabbage. He's a public information officer at oak

336
00:18:50,800 --> 00:18:54,320
Speaker 1: Ridge and he was the tour guide for the Manhattan

337
00:18:54,359 --> 00:18:57,960
Speaker 1: Project site. It was incredible. We had spent all day

338
00:18:58,400 --> 00:19:00,720
Speaker 1: looking at what's next in the future, and all of

339
00:19:00,720 --> 00:19:03,399
Speaker 1: the sudden we walked back into the past. It was

340
00:19:03,480 --> 00:19:07,600
Speaker 1: filled with vintage instruments and photographs. Yeah. I really thought

341
00:19:07,840 --> 00:19:10,359
Speaker 1: the wowl factor of the tour had ended when we

342
00:19:10,480 --> 00:19:13,280
Speaker 1: arrived at this sort of rundown site, but I have

343
00:19:13,320 --> 00:19:15,240
Speaker 1: to say it was about as cool as seeing the

344
00:19:15,240 --> 00:19:19,479
Speaker 1: supercomputer up close. Why though, did they choose oak Ridge

345
00:19:19,640 --> 00:19:22,119
Speaker 1: as one of the sites with the Manhattan Project. Well,

346
00:19:22,160 --> 00:19:24,600
Speaker 1: we asked Bill dot and here's what he said. This

347
00:19:24,720 --> 00:19:29,920
Speaker 1: area recognize it had a lot of water resources here,

348
00:19:30,000 --> 00:19:33,119
Speaker 1: it was a rail hub center Knoxville was so you

349
00:19:33,280 --> 00:19:37,080
Speaker 1: had electricity. He would also essentially in the middle of nowhere.

350
00:19:38,720 --> 00:19:40,240
Speaker 1: That might have been the sense back in the day,

351
00:19:40,280 --> 00:19:42,240
Speaker 1: but nobody would think that this was in the middle

352
00:19:42,280 --> 00:19:46,680
Speaker 1: of nowhere anymore. This was an incredible complex with a

353
00:19:46,720 --> 00:19:48,679
Speaker 1: bunch of buildings we didn't even have a chance to

354
00:19:48,720 --> 00:19:51,000
Speaker 1: go to, but we did have a chance to sit

355
00:19:51,040 --> 00:19:54,120
Speaker 1: down and speak with a number of engineers and scientists

356
00:19:54,160 --> 00:19:57,280
Speaker 1: who are actually using the Frontier system. One of them

357
00:19:57,480 --> 00:20:02,440
Speaker 1: was Steve Hamilton. Hamilton work in the Fission Infusion Energy

358
00:20:02,480 --> 00:20:06,320
Speaker 1: Sciences Directorate in a group called the High Performance Computing

359
00:20:06,359 --> 00:20:10,720
Speaker 1: for Nuclear Applications, and I am the primary investigator for

360
00:20:10,840 --> 00:20:16,800
Speaker 1: a project called XSMR, which is looking at solving multiphysics problems,

361
00:20:17,560 --> 00:20:22,160
Speaker 1: so problems that involve multiple different domain sciences related to

362
00:20:22,280 --> 00:20:25,320
Speaker 1: the modeling and simulation of nuclear reactors. And when you

363
00:20:25,359 --> 00:20:28,040
Speaker 1: say reactor, are we talking a nuclear reactor, So we're

364
00:20:28,040 --> 00:20:33,040
Speaker 1: talking nuclear fission reactors. So these would be designs that

365
00:20:33,080 --> 00:20:39,240
Speaker 1: would be targeting production of in most cases commercial electricity production.

366
00:20:40,480 --> 00:20:43,440
Speaker 1: So Steve's work on a very basic level is modeling

367
00:20:43,440 --> 00:20:46,840
Speaker 1: and simulating how new designs for smaller scale nuclear reactors

368
00:20:46,880 --> 00:20:50,080
Speaker 1: would function. He and his team are basically using Frontier

369
00:20:50,160 --> 00:20:52,720
Speaker 1: to a stress test these by building a replica and

370
00:20:52,760 --> 00:20:55,080
Speaker 1: then they test it out before trying to actually make

371
00:20:55,080 --> 00:20:58,520
Speaker 1: them at scale. There's multiple reasons that we want to

372
00:20:58,560 --> 00:21:03,680
Speaker 1: do these simulations. The largest factor really is cost for us.

373
00:21:04,000 --> 00:21:06,880
Speaker 1: So there are some safety considerations, but in large part,

374
00:21:07,359 --> 00:21:10,000
Speaker 1: if somebody proposes a reactor design, they say, I think

375
00:21:10,040 --> 00:21:12,840
Speaker 1: if we had this geometry and we put this fuel

376
00:21:13,400 --> 00:21:16,600
Speaker 1: and we use this coolant, I think it would be

377
00:21:16,880 --> 00:21:19,320
Speaker 1: much more efficient than anything we have out there. And

378
00:21:19,680 --> 00:21:23,680
Speaker 1: somebody puts this idea out there, it could take billions

379
00:21:23,680 --> 00:21:25,720
Speaker 1: of dollars for them to be able to take that

380
00:21:25,880 --> 00:21:28,600
Speaker 1: idea and push it forward to an actual prototype that

381
00:21:28,680 --> 00:21:32,399
Speaker 1: they could demonstrate and confirm that, yes, this reactor is

382
00:21:32,400 --> 00:21:35,879
Speaker 1: going to operate exactly the way that I thought it would.

383
00:21:36,840 --> 00:21:40,480
Speaker 1: If we use a computer like frontier, if we build

384
00:21:40,520 --> 00:21:44,359
Speaker 1: up a computational model, and we don't like to use

385
00:21:44,400 --> 00:21:46,720
Speaker 1: the term too much, but one of the terms that

386
00:21:46,880 --> 00:21:49,600
Speaker 1: is used a lot as digital twins. And so you're

387
00:21:49,600 --> 00:21:55,880
Speaker 1: building up this digital computational representation of the system. And

388
00:21:56,400 --> 00:22:00,600
Speaker 1: if you have enough confidence in your computational tools that

389
00:22:00,680 --> 00:22:06,000
Speaker 1: they're accurately predicting the physics, so they're predicting what would

390
00:22:06,000 --> 00:22:10,880
Speaker 1: happen inside of that reactor, then you can perform computational experiments.

391
00:22:11,359 --> 00:22:14,520
Speaker 1: You don't have to have all of the physical experiments,

392
00:22:14,600 --> 00:22:17,800
Speaker 1: and so you can build up confidence in these different

393
00:22:17,840 --> 00:22:22,200
Speaker 1: reactor designs purely using a computer like file. So you're

394
00:22:22,240 --> 00:22:24,480
Speaker 1: building a replica so that you can test it out

395
00:22:24,520 --> 00:22:28,760
Speaker 1: before bringing it to scale. Essentially, what are some real

396
00:22:28,800 --> 00:22:31,320
Speaker 1: life use cases like would we eventually see them powering

397
00:22:31,359 --> 00:22:35,560
Speaker 1: small cities? So that is definitely in the cards. There

398
00:22:35,680 --> 00:22:40,080
Speaker 1: are people who are targeting reactors. For instance, remote villages

399
00:22:40,720 --> 00:22:44,280
Speaker 1: that right now are relying on diesel generators for their

400
00:22:44,280 --> 00:22:49,240
Speaker 1: electricity production. If they could deploy a relatively small nuclear reactor,

401
00:22:49,280 --> 00:22:55,240
Speaker 1: they could have long term stable energy production for that community,

402
00:22:55,560 --> 00:22:59,360
Speaker 1: and it would ultimately be at a substantially lower cost

403
00:22:59,440 --> 00:23:03,240
Speaker 1: thanesel generator would cost to operate over long periods of time.

404
00:23:03,880 --> 00:23:08,960
Speaker 1: But there even within the smaller reactors doesn't necessarily imply

405
00:23:09,160 --> 00:23:12,480
Speaker 1: that it's these niche markets. So the small modulear reactors

406
00:23:12,480 --> 00:23:16,239
Speaker 1: can also be deployed for large scale electricity production. We

407
00:23:16,280 --> 00:23:19,199
Speaker 1: also spoke with Dan Jacobson. He's a chief scientist for

408
00:23:19,280 --> 00:23:24,919
Speaker 1: something called Computational systems Biology, So I'm a computational systems biologist.

409
00:23:25,000 --> 00:23:27,840
Speaker 1: We're trying to utilize data from all of those different

410
00:23:27,840 --> 00:23:31,920
Speaker 1: types of molecules together with environmental information and then using

411
00:23:31,960 --> 00:23:36,720
Speaker 1: whole ensembles of algorithms on the exoscale computing systems here

412
00:23:36,880 --> 00:23:40,240
Speaker 1: to solve what are really complex problems. This stuff was

413
00:23:40,280 --> 00:23:42,960
Speaker 1: pretty mind blowing. Dan talked to us about a bunch

414
00:23:42,960 --> 00:23:47,280
Speaker 1: of ways he's been using exoscale computing to layer information

415
00:23:47,600 --> 00:23:52,439
Speaker 1: to tackle issues like climate health or future pandemics. But

416
00:23:52,480 --> 00:23:55,240
Speaker 1: there's one thing he's using Frontier for that we did

417
00:23:55,359 --> 00:24:00,000
Speaker 1: not expect. One of the big concerns in the US population,

418
00:24:00,880 --> 00:24:04,440
Speaker 1: even perhaps more so in the veterans population is suicide,

419
00:24:04,480 --> 00:24:08,840
Speaker 1: suicide ideations, suicide attempts, and one of the initial questions

420
00:24:09,000 --> 00:24:12,680
Speaker 1: in that is are there genetic architectures that predispose people

421
00:24:12,920 --> 00:24:17,040
Speaker 1: to suicide attempts. We have this wonderful collaboration between the

422
00:24:17,160 --> 00:24:21,200
Speaker 1: Veterans Administration and the DUE where actually all the electronic

423
00:24:21,280 --> 00:24:23,520
Speaker 1: health records for the entire VIE systems. We have a

424
00:24:23,560 --> 00:24:26,800
Speaker 1: copy of herod O creation gets updated nightly, and we

425
00:24:26,920 --> 00:24:31,000
Speaker 1: have genotype information now for over eight hundred thousand patients

426
00:24:31,000 --> 00:24:32,760
Speaker 1: and the goal is they get that to two million.

427
00:24:33,240 --> 00:24:36,280
Speaker 1: This is out of the MVP million Veterans program from

428
00:24:36,320 --> 00:24:39,960
Speaker 1: the VA, and this is a tremendous resource for human

429
00:24:40,040 --> 00:24:42,879
Speaker 1: systems biology and human health trying to understand again the

430
00:24:42,960 --> 00:24:46,520
Speaker 1: mechanisms of disease, both at the clinical level as well

431
00:24:46,520 --> 00:24:50,280
Speaker 1: as at the molecular level. Having this larger population that

432
00:24:50,400 --> 00:24:54,199
Speaker 1: we have genomics information and clinical information for allows us

433
00:24:54,240 --> 00:24:58,680
Speaker 1: to start to tackle those sorts of questions, questions about

434
00:24:58,760 --> 00:25:01,840
Speaker 1: data and privacy with all of these frontier projects kept

435
00:25:01,840 --> 00:25:04,640
Speaker 1: coming up throughout our visit, so we wanted to know

436
00:25:05,000 --> 00:25:07,840
Speaker 1: how are they protecting privacy of these veterans health records.

437
00:25:08,440 --> 00:25:12,720
Speaker 1: The VA data sits here a very secure enclave. It's

438
00:25:12,720 --> 00:25:17,440
Speaker 1: a PHI certified enclave, and they are very strict access controls.

439
00:25:18,359 --> 00:25:21,280
Speaker 1: Everybody goes through lots of training to be able to

440
00:25:21,440 --> 00:25:24,199
Speaker 1: use this data. But the data that we see is anonymized.

441
00:25:24,240 --> 00:25:28,639
Speaker 1: We don't see names or addresses. That's all been scrubbed

442
00:25:28,640 --> 00:25:31,720
Speaker 1: by the time we see it. And so we're interested

443
00:25:31,720 --> 00:25:36,080
Speaker 1: in understanding the clinical variables, and we're looking at thousands

444
00:25:36,160 --> 00:25:37,960
Speaker 1: or millions of them, so we're never going to drill

445
00:25:38,000 --> 00:25:42,280
Speaker 1: down to an individual patient. Genomic tracking is already pretty impressive,

446
00:25:42,600 --> 00:25:45,359
Speaker 1: but Dan explained that it's just the tip of the

447
00:25:45,359 --> 00:25:49,560
Speaker 1: iceberg in understanding and treating disease. We're doing a lot

448
00:25:49,560 --> 00:25:51,800
Speaker 1: of this in addiction work. We study a lot of

449
00:25:51,840 --> 00:25:56,800
Speaker 1: neuropsychological conditions, cardiovascular disease, cancer, infectious diseases, and we're actually

450
00:25:56,880 --> 00:25:59,200
Speaker 1: scaling up to do this on about two thousand to

451
00:25:59,280 --> 00:26:02,760
Speaker 1: three thousand human diseases, so we'll have this complete network

452
00:26:02,800 --> 00:26:07,359
Speaker 1: model of the interconnections between human disease and mechanistic understanding

453
00:26:07,359 --> 00:26:10,760
Speaker 1: and how you have a lot of comorbidities people with

454
00:26:10,840 --> 00:26:13,359
Speaker 1: multiple diseases that tend to co occur. This will help

455
00:26:13,480 --> 00:26:17,399
Speaker 1: us understand why why are these things tied together. Another

456
00:26:17,440 --> 00:26:19,879
Speaker 1: area he and his team built was a climate model

457
00:26:19,920 --> 00:26:24,600
Speaker 1: of basically the entire planet. We've taken about the past

458
00:26:24,680 --> 00:26:28,040
Speaker 1: sixty years of climate information for every point of land

459
00:26:28,080 --> 00:26:32,760
Speaker 1: on the planet around the world, and we built vectors tensors,

460
00:26:32,760 --> 00:26:36,600
Speaker 1: strings of numbers representing all the different components in the

461
00:26:36,720 --> 00:26:40,440
Speaker 1: environment at every position. Well, those vectors, now we can

462
00:26:40,600 --> 00:26:44,119
Speaker 1: use the Comet codebase we develop on Summit and Frontier

463
00:26:44,240 --> 00:26:46,520
Speaker 1: to compare all of those vectors to each other. So

464
00:26:46,560 --> 00:26:48,840
Speaker 1: we're trying to say what are the similar environments of

465
00:26:48,880 --> 00:26:51,359
Speaker 1: all the possible environments on Earth. So now we have

466
00:26:51,400 --> 00:26:55,840
Speaker 1: a really fascinating color representation and this is the highest

467
00:26:55,840 --> 00:27:00,360
Speaker 1: resolution climate type map ever done in human history. It

468
00:27:00,400 --> 00:27:04,200
Speaker 1: looks really cool, but the truly amazing thing was learning

469
00:27:04,320 --> 00:27:06,479
Speaker 1: what you can actually do with this model. We can

470
00:27:06,560 --> 00:27:10,680
Speaker 1: zoom into different areas of interest in this case for pandemics,

471
00:27:10,720 --> 00:27:16,840
Speaker 1: things like Eastern Australia, Indonesia, Bangladesh from Central Africa, all

472
00:27:16,920 --> 00:27:19,199
Speaker 1: hot spots for different types of viruses that we know

473
00:27:19,240 --> 00:27:21,960
Speaker 1: are being affected by climate change, and this is allowing

474
00:27:22,040 --> 00:27:24,119
Speaker 1: us to shine the flashlight where do we need to

475
00:27:24,200 --> 00:27:28,520
Speaker 1: be concerned. Where's the biggest relativistic climate change affecting bat

476
00:27:28,520 --> 00:27:31,560
Speaker 1: populations and viruses and their food sources. Where do we

477
00:27:31,600 --> 00:27:34,000
Speaker 1: need to be focusing. So to pull this off, we

478
00:27:34,200 --> 00:27:37,920
Speaker 1: did nine point three extra flop calculation. That's the fastest

479
00:27:37,920 --> 00:27:41,000
Speaker 1: calculation done in human history. Overall, it was one hundred

480
00:27:41,000 --> 00:27:44,040
Speaker 1: and sixty eight zeta ops, so that's ten to the

481
00:27:44,080 --> 00:27:48,000
Speaker 1: twenty three zeros mathematical operations that had to be calculated

482
00:27:48,000 --> 00:27:51,040
Speaker 1: to pull this off. It's one of the largest calculations

483
00:27:51,040 --> 00:27:54,560
Speaker 1: done and about ten to the seventeenth network edges calculated.

484
00:27:54,600 --> 00:27:57,240
Speaker 1: This one of the largest networks ever created. Again, how

485
00:27:57,280 --> 00:28:00,320
Speaker 1: we're using these sorts of exoscale resources to tackle problems

486
00:28:00,359 --> 00:28:03,440
Speaker 1: we simply couldn't do before. So at this point we're

487
00:28:03,440 --> 00:28:05,760
Speaker 1: really starting to get a sense of how much is

488
00:28:05,800 --> 00:28:08,919
Speaker 1: possible when you have computing power with like eighteen zeros

489
00:28:08,960 --> 00:28:11,440
Speaker 1: after it per second. But I think for us both

490
00:28:11,800 --> 00:28:15,359
Speaker 1: this next example was absolutely the most fascinating is something

491
00:28:15,440 --> 00:28:18,680
Speaker 1: called zoonosis, and it made us really glad we were

492
00:28:18,720 --> 00:28:20,800
Speaker 1: there with some really large brains that are working on

493
00:28:20,840 --> 00:28:23,280
Speaker 1: these problems. A lot of our focus now is on

494
00:28:23,520 --> 00:28:29,679
Speaker 1: trying to understand zonosis. How pathogens, viruses and other microbes

495
00:28:29,920 --> 00:28:34,080
Speaker 1: often sit nascent in animal reservoirs. When those viruses get

496
00:28:34,119 --> 00:28:37,040
Speaker 1: into other species that are not ready for them, like us,

497
00:28:37,400 --> 00:28:40,200
Speaker 1: that leads to disease, and COVID nineteen is one example

498
00:28:40,240 --> 00:28:43,560
Speaker 1: of that. But the rules of zonosis, how and why

499
00:28:43,680 --> 00:28:47,240
Speaker 1: this happens and what's driving it are actually fairly poorly understood.

500
00:28:47,640 --> 00:28:51,880
Speaker 1: And then he hit us with flying foxes. So the

501
00:28:51,960 --> 00:28:55,040
Speaker 1: system are studying are in flying foxes. So these are

502
00:28:55,080 --> 00:28:57,640
Speaker 1: these giant bats that have like a six foot wingspan.

503
00:28:57,960 --> 00:29:01,040
Speaker 1: They traditionally live in roosts of two hundred thousand bats

504
00:29:01,040 --> 00:29:04,960
Speaker 1: in the jungle. They carry these viruses called hendra and hinipoviruses,

505
00:29:05,080 --> 00:29:07,840
Speaker 1: which in the Australian system, if they get into horses

506
00:29:07,880 --> 00:29:10,360
Speaker 1: they kill about seventy percent of the horses. When they

507
00:29:10,360 --> 00:29:12,840
Speaker 1: get from horses into people, it's also a seventy percent

508
00:29:12,920 --> 00:29:16,040
Speaker 1: mortality rate. So, I mean, COVID nineteen has been horrific,

509
00:29:16,120 --> 00:29:19,560
Speaker 1: but it's a sub one percent mortality rate. Seventy percent

510
00:29:19,720 --> 00:29:23,680
Speaker 1: mortality rate is game over right, So we want to

511
00:29:23,720 --> 00:29:26,520
Speaker 1: really understand these systems and learn how to prevent that

512
00:29:26,560 --> 00:29:29,120
Speaker 1: sort of spread before they become efficient at human to

513
00:29:29,200 --> 00:29:32,640
Speaker 1: human spread. All right, Wess, let's take a moment here

514
00:29:32,680 --> 00:29:34,840
Speaker 1: for you to digest this. It took us a minute

515
00:29:35,360 --> 00:29:38,600
Speaker 1: the idea that this virus could go from flying foxes

516
00:29:38,640 --> 00:29:42,760
Speaker 1: to horses and then horses to humans. I mean, this

517
00:29:42,880 --> 00:29:47,920
Speaker 1: could potentially wipe out populations. So we're actually looking at

518
00:29:48,040 --> 00:29:51,040
Speaker 1: We know that when there are food shortages, that puts

519
00:29:51,080 --> 00:29:54,480
Speaker 1: the bats under stress, messing with their ecosystem. So we

520
00:29:54,560 --> 00:29:57,800
Speaker 1: want to understand what causes these food shortages. So we

521
00:29:57,920 --> 00:30:00,640
Speaker 1: took all these environmental variables and we threw them into

522
00:30:00,720 --> 00:30:04,000
Speaker 1: one of our explainable AI approaches over time and space,

523
00:30:04,440 --> 00:30:06,920
Speaker 1: and we're actually we thought this was a ten year

524
00:30:07,000 --> 00:30:09,360
Speaker 1: goal that we could someday learn some of these rules

525
00:30:09,400 --> 00:30:12,600
Speaker 1: about an environment and Baigali, we're about to submit the

526
00:30:12,600 --> 00:30:16,440
Speaker 1: paper showing this predictive model that is very good at

527
00:30:16,440 --> 00:30:18,680
Speaker 1: picking up these little black dots at the top are

528
00:30:18,760 --> 00:30:21,640
Speaker 1: when food shortages are occurring, so we can start to

529
00:30:21,720 --> 00:30:24,160
Speaker 1: get up to a year warning when one of these

530
00:30:24,280 --> 00:30:26,840
Speaker 1: pandemic breakoats could be occurring that we need to start

531
00:30:26,880 --> 00:30:29,200
Speaker 1: watching for and then zooming on where and when to look.

532
00:30:30,000 --> 00:30:33,600
Speaker 1: So you saw all of that in just one day,

533
00:30:34,040 --> 00:30:37,120
Speaker 1: Ed oak Ridge, Yeah, we did. And you know, it's

534
00:30:37,200 --> 00:30:39,840
Speaker 1: crazy because these are just some of the first projects

535
00:30:39,880 --> 00:30:44,240
Speaker 1: getting started because it just got built and it's got,

536
00:30:44,280 --> 00:30:46,880
Speaker 1: as we heard, five to seven years of lifespan left,

537
00:30:46,920 --> 00:30:50,120
Speaker 1: so there's a lot of room to grow, and they're

538
00:30:50,160 --> 00:30:53,960
Speaker 1: just getting started, and as we also learned, they're already

539
00:30:54,000 --> 00:30:58,240
Speaker 1: building the next one. So I mean, for us, one

540
00:30:58,240 --> 00:31:01,360
Speaker 1: of the big challenges of on a machine like Frontier,

541
00:31:01,360 --> 00:31:04,600
Speaker 1: which has also been one of the fun aspects of it.

542
00:31:04,720 --> 00:31:08,040
Speaker 1: So it's all a big challenge, but it's just the

543
00:31:08,120 --> 00:31:11,400
Speaker 1: sheer number of people that you have to bring together

544
00:31:11,880 --> 00:31:14,240
Speaker 1: to be able to run one of these simulations, the

545
00:31:14,360 --> 00:31:18,400
Speaker 1: number of interactions and managing that as you're working all

546
00:31:18,440 --> 00:31:21,880
Speaker 1: working towards getting to this final capability that you have

547
00:31:22,000 --> 00:31:24,280
Speaker 1: to be able to use a machine like Frontier. It's

548
00:31:24,320 --> 00:31:28,040
Speaker 1: been very, very rewarding. Vic Katherine, next time you're go

549
00:31:28,080 --> 00:31:30,320
Speaker 1: on a field trip, I'm going with you. We'd love

550
00:31:30,360 --> 00:31:33,880
Speaker 1: to have you us. You got it. Thanks for listening

551
00:31:33,880 --> 00:31:35,880
Speaker 1: to us here at the Big Take. It's a daily

552
00:31:35,920 --> 00:31:39,040
Speaker 1: podcast from Bloomberg and iHeartRadio. For more shows from My

553
00:31:39,200 --> 00:31:43,160
Speaker 1: Heart Radio, visit the iHeartRadio app, Apple Podcasts, or wherever

554
00:31:43,240 --> 00:31:46,160
Speaker 1: you listen, and we'd love to hear from you. Email

555
00:31:46,240 --> 00:31:50,000
Speaker 1: us questions our comments to Big Take at Bloomberg dot net.

556
00:31:50,760 --> 00:31:54,440
Speaker 1: The supervising producer of The Big Take is Vicky Bergolina.

557
00:31:54,480 --> 00:31:58,360
Speaker 1: Our senior producer is Katherine Fink, and they both produced

558
00:31:58,400 --> 00:32:03,560
Speaker 1: this episode. Raphael M Seely is our engineer. Our original

559
00:32:03,640 --> 00:32:07,920
Speaker 1: music is composed by Leo Sidrin. I'm West Casova. Will

560
00:32:08,040 --> 00:32:11,080
Speaker 1: be back on Monday with another Big Take. Have a

561
00:32:11,120 --> 00:32:24,800
Speaker 1: great weekend. My kids are at an age where we've

562
00:32:24,840 --> 00:32:28,520
Speaker 1: been watching The Terminator. Terminator to cyberdine, will it become

563
00:32:28,560 --> 00:32:30,600
Speaker 1: self aware? Do you do you believe that it has

564
00:32:30,640 --> 00:32:34,040
Speaker 1: a capability to become self aware? I don't. I will

565
00:32:34,080 --> 00:32:36,800
Speaker 1: say that colleagues of mine who are interested in this

566
00:32:37,080 --> 00:32:38,840
Speaker 1: question and I talked about it that I just talked

567
00:32:38,880 --> 00:32:41,760
Speaker 1: about about where do the heavy elements come from? Um.

568
00:32:42,240 --> 00:32:45,520
Speaker 1: They built a code module to sort of do that,

569
00:32:45,600 --> 00:32:48,239
Speaker 1: to sort of trace the transmutation of elements as they

570
00:32:48,240 --> 00:32:53,320
Speaker 1: get heavier and heavier. They called it Skynet, which I

571
00:32:53,360 --> 00:32:57,240
Speaker 1: thought was like dangerously arrogant of their part. Yeah, I

572
00:32:57,360 --> 00:32:59,640
Speaker 1: thought I thought, really, you're going to tempt fail. It's

573
00:32:59,680 --> 00:33:01,840
Speaker 1: a little too it's a little too real. We're getting

574
00:33:01,880 --> 00:33:03,960
Speaker 1: a little late in the game, if you will, to

575
00:33:04,000 --> 00:33:06,200
Speaker 1: call something that. But yeah, they just plowed ahead. And

576
00:33:06,240 --> 00:33:08,520
Speaker 1: the in fact, they even use and you know on

577
00:33:08,600 --> 00:33:11,240
Speaker 1: their website, Yeah they use the symbol from the movie. Yeah,