HARDWARE

Ministry of Energy for Artificial Intelligence Chips


12:55 PM EDT – Welcome to Hot Chips! This is the annual conference on the latest, largest and upcoming big silicon that excites us all. Follow our regular AnandTech blogs live on Mondays and Tuesdays.

12:55 PM EDT – It starts at five o’clock

1:03 PM EDT – The third introductory word of the American government

1:05 PM EDT – ‘I do not claim to speak on behalf of the US government’

1:05 PM EDT – ‘These are my views’

1:06 PM EDT – I worry about that

1:06 PM EDT – Science and technology are developing differently, but they are intertwined

1:07 PM EDT – Technology leads science, and science leads technology

1:09 PM EDT – The Ministry of Energy is really the energy of everything

1:09 PM EDT – I have to report to the president every year in March

1:11 PM EDT – he talks a lot about DoE holistically

1:11 PM EDT – DOE ‘solves problems that no one else can’

1:12 PM EDT – 17 national laboratories, workforce of 100 thousand

1:13 PM EDT – It’s not about being faster, it’s about being smarter

1:13 PM EDT – make decisions on the basis of which actions can be carried out

1:14 PM EDT – ‘I can’t talk about everything we do’

1:14 PM EDT – They do not necessarily have to be experts, but they know how to solve complex problems.

1:15 PM EDT – the questions that are asked are often unanswered

1:16 PM EDT – moving towards the exascale

1:18 PM EDT – DoE was hit by Congress because of Fugaku

1:18 PM EDT -Make a top-down examination

1:18 PM EDT – DoE does things that are considered crazy, but have proven fruitful

1:19 PM EDT – 3 years ahead of the Darpa program to move to fifth scale with the PS3

1:19 PM EDT – Defining things with a purpose – to solve problems, and these are instruments

1:22 PM EDT – Learn to solve complex problems

1:23 PM EDT – Learning modeling

1:24 PM EDT – I quote from Neumann

1:26 PM EDT – Our largest computer simulations emphasize our largest systems – architectures have been historically developed to solve equations, not to deal with data on tomorrow’s scale

1:29 PM EDT – Working on Fukushima, we had to answer in an hour whether to evacuate American citizens from Tokyo

1:30 PM EDT – There is no trivial problem if you decide to do so

1:32 PM EDT – There is no metric as to how close you could be to some of these high-level decisions

1:34 PM EDT – This is what DoE does

1:35 PM EDT – More on “just simulate faster won’t help”

1:37 PM EDT – DoE is a big customer of SambaNove

1:38 PM EDT – There are Cerebras, Groq, Graphcore

1:38 PM EDT – Don’t just think about the immediate chip market

1:39 PM EDT – I have to do much more than this

1:42 PM EDT – Each slide has a long descriptive treatment, but there is little to bite

1:44 PM EDT – Creating petabytes of data using five-scale systems

1:45 PM EDT -One person can no longer be a top-to-bottom solution for epiphanies

1:47 PM EDT – health care due to rare outcomes

1:50 PM EDT – I’m still waiting to talk about the chip

1:51 PM EDT – waiting to see if the reconciliation will take place in the house for money in accordance with the CHIPS Act

1:55 PM EDT – Time for questions and answers

1:57 PM EDT – Q: What do future chips look like? A: PIM is good, you need to visit the architectures again. The old systems were locked into the world we grew up in. These used to be the only ways to get things done. The forties were the last time we cleaned the sheet from top to bottom as we changed things

2:02 PM EDT – Q: Should DoE develop its own chips or buy commercial silicon? A: You have done this in the past. But you have to have an ecosystem. Over time, the perspective adjusted the maps of the company, the RNE worked, harmonizing it with the requirements of DoE. It must be sustainable. The idea is to build a design that others build, but it’s important to work with the private sector. DoE understands the route plan and gets involved, whether there are points of intersection with DoE and these private companies. It’s not always the point, but conversations start early

2:04 PM EDT – Q: Investing in startups A: Investing in startups unlike VC – our investments include successful workers that include data modalities – well-organized scientific data and scarce data. They invest in data. Have a place to practice the technologies that work for the problems they face. We are not organized enough to have a large enough opening in the private market to do that – we need more test beds to give real data to people who offer real problem solving

2:05 PM EDT – ROI for DoE is different from VC

2:06 PM EDT – It’s a wrapper



Source link

Naveen Kumar

Friendly communicator. Music maven. Explorer. Pop culture trailblazer. Social media practitioner.

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *

Back to top button