Moore's Law, exponential growth, and extrapolation!
ฝัง
- เผยแพร่เมื่อ 8 มิ.ย. 2024
- Moore's law of exponential growth in computing is important for machine learning, scientific computing, computer graphics, and even for artificial intelligence. In this video we discuss the basis of Moore's law, and projections for the future.
- วิทยาศาสตร์และเทคโนโลยี
I started to get interested in fluid dynamics by watching your videos. Congratulations on the beautiful work.
Very interesting Professor! It is great to have an academic take on this topic. Most coverage I have seen to date fails to mention all of the avenues of possible future development. Thanks for your video.
I would love to see a series on the finite element method (FEM).
Thoroughly fascinating...
Thank you!
Amazing work!
Thank you! It is helpful!
Great video. For quite some time I'm fascinated by the fact that autonomous cars still consume small kW numbers to be autonomous. Yet our brain and body can only handle about 20w, and driving is not the most demanding task we are capable of. In fact, driving a small autonomous EV through the city needs more energy for the autonomous part than it needs for the actual driving.
I would like a video about low-precision computing. Or as I like to call it right-precision computing. IBM has an interesting AIU chip with precisions down to int2. That really is enough if you want to store the result of a hypothesis test and if you trust it or not. If you only classify between view objects, get view frequencies out of your FFT, or your ADC is a certain bit number you can limit yourself to the transistor you actually need. Especially if you do matrix operation or anything where the number of operations goes squared with the number of bits or entries you have this is a huge deal. ( 32x32 = 1024, 8x8 = 64, 1024/64 = 16 : if you can fit your problem into 8bit you only need 1/16th of the transistors)
I also like the idea of parallel computing, we do this in automotive engineering all the time by adding more and more chips into basically everything, to hock it on a bus later on.
Both ideas combined get much closer to my intuition about how humans "compute" driving.
You' re the best thank you
Love the video
love your videos
The other thing to consider, related to exponential growth (or broadly like it), are "network effects".
Magnificent video presentation of complexity. My question is whether we can extract distance information between a and within objects in a flat photograph perhaps using a version of projective geometry?
Hi Dr. Brunton. Do you know of a well written paper that compares the tradeoffs (rigorously or just implications) of the different computing methods (photonic, neuromorphic, etc) you mentioned?
Dear Steve, is there a way to get de data you used to make the plottings?, of course you will be referenced.
a side note on the "supercomputer" on your pocket... yeah, it might be, but they make it so difficult to program that you cannot use it as such (unless you belong the "elite").. specially the company on the example...
There is a point where the background radiation of natural silicon (or whatever semiconductor) will cause too many single event upsets. Changes in the general architecture of the devices, such as vertical devices, extend when, but it will happen eventually.
I’m only 13 minutes in, but I was hoping for a breakdown of Moore’s law in relation to simulation problems. If you’ve got the time may you add in time stamps?
Edit: He gets there! My bad.
RIP Gordan Moore.
K - knowledge
t - time
dK/dt=K(t) →K(t)=exp(t)
🤔
people has said that moore's law is dead
Exponential growth does not go on for ever? hmmm tell that to our Universe.