[10] Mike Davies, Director of Intel's Neuromorphic Lab

Ever wondered why a CPU doesn't do math like a brain? What if we had technology that acted like a brain? It turns out we do, and it's all called Neuromorphic computing. Intel has just launched its second-generation Neuromorphic processor, so we sat down with Mike Davies, Director of Intel's Neuromorphic Computing Lab, to find out all the juicy details about how research is progressing.
0:00 Intro
1:05 Q1: What is Neuromorphic Computing?
4:21 Q2: Roadmaps
6:55 Q3: Easy Upgrade?
8:03 Q4: Good for Intel 4?
9:51 Q5: Pre-production Intel 4?
11:40 Q6: Advances in Loihi 2
15:37 Q7: Examples of Scale
18:16 Q8: Loihi with Ethernet
19:28 Q9: Power and TDP
21:07 Q10: Neuromorphic Sensors
24:11 Q11: Parallels to Biology
25:40 Q12: Competitive Training
27:54 Q13: Equivalent to IPC?
30:37 Q14: Software and Lava
32:45 Q15: Homebrew Use?
34:10 Q16: Main Takeaway
35:13 Cat Tax
Welcome to the TechTechPotato (c) Dr. Ian Cutress
Ramblings about things related to Technology from an analyst
Patreon helps support this channel: / techtechpotato
Join the TechTechPotato Discord! All we ask is you sign up for Patreon for as little as $1.50 a month and your account will be linked - it only has to happen once, and then you can cancel if you wish. Come chat and hang out!
Support the channel by hitting that subscribe button, or using our Amazon affiliate link. Please note that TTP may receive a commission if you purchase anything through these links
Intel Core i7-11700K (8C/16T): geni.us/SreZM
Intel Core i5-11600K (6C/12T): geni.us/T4WG
Intel Core i5-11400F (6C/12T): geni.us/6UISblx
AMD Ryzen 9 5950X (16C/32T): geni.us/5950X
AMD Ryzen 9 5900X (12C/24T): geni.us/5900X
AMD Ryzen 7 5800X (8C/16T): geni.us/5800X
AMD Ryzen 5 5600X (6C/12T): geni.us/5600X
Intel Core i9-10900K (10C/20T): geni.us/Bench2020-10900k
Intel Core i7-10700K (8C/16T): geni.us/Bench2020-10700K
AMD Ryzen 9 3950X (16C/32T): geni.us/Bench2020-3950x
AMD Ryzen 7 3700X (8C/16T): geni.us/Bench2020-3700x
AMD Ryzen 5 3600 (6C/12T): geni.us/d08ThJ4
Follow Ian on Twitter at / iancutress
Follow TechTechPotato on Twitter at / techtechpotato
#techtechpotato #neuromorphic #loihi

[@__aceofspades]

8x neurons while at HALF the die size? Impressive. Really looking forward to seeing what Intel can do Loihi. Also excited to see Intel 4 node in the wild!

[@crustybread9286]

Mike is great to work with ! Worked with him as temp as fresh grad with him at Fulcrum pre intel acquisition . Good memories great team from caltech.

[@crustybread9286]

Wonder if Mike still runs professional marathon :)

[@julienheinzel]

Is it possible to have an interview with the company Brainchip Holdings? They have developed a commercial neuromorphic chip with even more synapses and neurons. True neural network and easy programming. Even one-shot learning on the Edge without cloud works! I can't wait to see it as it is a real game changer for developers!

[@salokinsoistok1304]

BRAINCHIP AKIDA can do so much more right now and is ready for commercialization!

[@paulkirkland9860]

As a Neuromorphic Engineering researcher, this is great to see being covered along with other general tech news. Great interview always great to hear from Mike he describes the benefits of Neuromorphic computing very eloquently. First time watching techtechpotato but now you have my interest and a sub for the great interview.

[@auseire8656]

I'd like to know how Loihi 2 compares to Brainchips Akida 1000 Neuromorphic processor? Akida doesn't require any connection to the internet so it not cloud based and requires ultra low power to work.

[@TheCgOrion]

This was fascinating to listen to. Thank you very much, and thanks to Mike Davies for taking the time to do this interview.

[@josephkelly4893]

Love the interview style Ian. Thanks for providing these

[@RohitSharma-mi8gt]

These interviews with research scientists and engineers are really cool. Great to learn about how people think about problems.

[@rougenaxela]

6:48 This talk of "hidden away" uses for the neuromorphic computing in a processor, makes me immediately think "neuromorphic computing based branch prediction engine"

[@Altirix_]

what i was thinking. this could be a major part of intels IPU plans against Apple

[@kelownatechkid]

Yup, no doubt this is a key reason this is being worked on

[@MarcosScheeren]

Awesome! Been following Loihi progress for a while now. Let's listen!!

[@metallurgico]

16:56 he basically said that in the future there will be drones with neural CPUs that scan the environment and look for matches in a database. Nice.

[@Patrick73787]

28:20 "The number of synaptic operations processed per unit of time" I like that new metric :)

[@quantuminfinity4260]

Fantastic interview, thank you!

[@kozmizm]

I'm only 7 minutes in, and I've seen Mike Davies recently in a youtube video about Loihi, and I feel like in this video, right away, he doesn't convey clearly enough the energy advantages about this style of computing. It is so extremely efficient for certain tasks that nothing else even comes close. This kind of architecture could solve so much about our energy demands and heat dissipation problems of the current model of chip design. Already some companies are trying to integrate memory with computation, but nature has been iterating for millions of years on cell biology, possibly longer, and the neuronal model is an incredibly efficient model where each bit of memory(neurons) are one bit of a computation. Essentially neuromorphism is copying the efficiency of nature and furthering our limited understanding of the animal brain. It's truly remarkable. I wish he would have expressed that clearly in the beginning of this video, because I think that is the greatest advantage of this architecture and why it will dominate the future in many areas of computing

[@TechTechPotato]

He's doing more and more videos these days, finding the best way to express things

[@kelownatechkid]

These are amazing, thank you so - so much!

[@user-ih9ey5lu3m]

Please do a video about Brainchip Holdings and their Akida1000. Like many mentioned in the comments before, is the chip superior in performance compared to the Loihi, entering the commercialization stage, and I'm certain that a lot of people would love to see a high-quality video such as this one about Brainchip Holdings, and it's Akida chip.

[@vladimirholecka7462]

Greetings Ian, It’s Vladimir from Youmaker, one of the developing video platforms based in the US. Youmaker is international destination where creators’ freedom of expression is honored and it does not censor for political reasons. We have seen your KZbin channel and it’s really incredible. We absolutely think that our fans would love your videos. We would like to invite you to join us. Also, it can serve for you as a free back up storage for your videos. Can I send you details?

[@GNARGNARHEAD]

that's amazing I had no idea that human brain scale was so close, sounds like incredible work

[@killerarchery]

I can't wait to see your interview with Brainchip about the already commercially available Akida. It seems that Intel is having problems with their development. Maybe because they keep running into Brainchip's patents.

[@444haluk]

♫♫ "I see friends shaking hands saying ' It's a fully memory and compute integrated model, computing elements sitting very close to the storage state' ♫♫ They're really saying, "I love you"♫♫

[@sungk3962]

This is nice and all, but does it integrate with popular optimization software like CPLEX, Gurobi and Mosel?

[@codycast]

No

[@Veptis]

So we get back from an all digital approach again? Even something that is more analog again?

[@Omnis2]

He looks like Anders Hellman the dude in Cyberpunk that made the Relic chip. 😂

[@henrycobb]

What's the average number of elements lost per chip due to errors with the new process? (At least not a total writeoff like an X86 core with a single defect though.) Also will there be any market share left when Intel finally joins the desktop optronics neuromorphic crowd?

[@erichrybinczak8894]

I remember IBM made one as well a few years ago, not sure if they still producing them. Think be awesome for ai and ml. Lots of the work loads that being g said seems be like the ml enginees being aced I to processors now.

[@bhuvaneshs.k638]

Is there any online courses on neuromorphic

[@gerryjamesedwards1227]

I wonder if the development of these chips will ultimately help us to understand the Human brain better?

[@gerryjamesedwards1227]

Isn't instinct really neuron programming in action?

[@Cineenvenordquist]

Instinct is seeing some great research (murine) where we see how attuned to catching crickets can reflect such programming, but of course this CNN/DNN programming reflects the Ca+ and Ca2+ channel affect on dendrite and axon synapse growth first, and GAN or Allen Brain Center revelations on cell progeny, development and longer experience in manners to be published yet. So good news, not everybody wants to work in reproduction, and ABC (supra) can show you some pretty ace microscopy of entire fruit fly and mouse brains, plus a bunchload of visual cortex (e'vbody) and hippocampus to chase your notions of what takes more than a hypothalamus to be a gifted actor by. So, it's a network of weights and affinities n neurons deep and m wide, but with a plethora of algebras to emulate those MACs by; whereas instinct reflects metagenomics, developmental cell phylogeny, and dev. behavior. But maybe you can find DBS baseballnomicians who aren't jumping the gun, too?

[@2dozen22s]

At these really low power consumption levels, die stacking looks super appealing

[@olo398]

whoosh!

[@Stadtpark90]

Neuro-Morphic-Computing with electro-optic superconducting “Loop-Neurons”: watch Lex Fridman #225 interviewing Jeffrey Shainline from NIST.

[@Stadtpark90]

kzbin.info/www/bejne/e6jYlqSanK6pp5Y

[@TheRogbar]

Intel does great work in computing development but would do its community a great service and save 3 year research by giving credit to the 20 years development of AKIDA Neuromorphic SOC by Brainchip Inc. Credit where Credit is Due !!!

[@souravmukherjee4987]

Why aren't they using memristors?

[@Irthex]

I wonder how much dark silicon there is on the Loihi chips. With this kind of spikey behaviour they should be able to keep dark silicon to a minimum.

[@codycast]

7%

[@utp216]

So basically Mike is the Miles Bennett Dyson of the real world… I hope the processors him and his team design don’t enable Skynet!

[@apefu]

Haven't listened yet, but I DEFINITELY thought this was an episode from Lex Friedmans podcast at first.

[@TheParadoxy]

lol, same!

[@adriankelly3338]

Ian is a much better interviewer than Lex Fridman! Much more knowledgeable.

[@zorbalight3933]

Interesting you made that conclusion before you listened to it?

[@apefu]

@@zorbalight3933 *sigh* Its the thumbnail.

[@TheParadoxy]

@@adriankelly3338 ftr, they feel incomparable to me. They're both great imo. You're welcome to disagree and I do find myself watching more of Ian's interviews. Just wanted to clarify since I wasn't expressing regret or anything that it was a ttp interview

[@Ironclad17]

Ian you were so happy about Intel's node names, but even you can't get them right. It's Intel 4, not 4nm!

[@banana-vo6bx]

There is nothing in intel 22/14/10/7nm,tsmc N10/7/5/4 actually that size. Its just marketing.

[@TechTechPotato]

TSMC N7 is called 7nm Intel 4 will be called 4nm It's actually a writing nightmare

[@tringuyen7519]

Neuromorphic cell= 1/2 precision floating point core with a memory cache. Because why compute anything when you can just remember it, right? I am on the fence about how effective this architecture will be. I think Apple and AMD will come up with different architectures from Intel. In a human brain, an idea/thought is not a single neuron but a set of neurons.

[@Cineenvenordquist]

Ian is ready to start ordering Lambda Neuromorphic sessions, whereas the researcher is trying to convince him that a) the field isn't dead yet, and b) this is part of why an experimenter and experiment both still matter (premature optimizations et al!) Even if having a manic session with neuromorphic fabric will always still mean not -cooking- the fabric with stimulation, I might have expected to hear it asked whether physical but comparatively innumerate MXene DNNs might occupy fabrics with Intel Neuromorphic Accelerators and trade duties to purpose (or as many purposes fit, if we need those bullet points filled out.) Maybe a Cargill or Komatsu or Grumman follow-on interview would be to Dr. Ian's taste; suppose a farmer wants a field free of weed seeds and an intensive start. Can she order a HALO strike to use cyclical atmospheric energy to condense a particle, drop it just so, and clear the field but not the wetlands? What about capturing necessary biota and transferring the will to control pest insects and fungi, or even shape fruiting bodies later on to get awesome shaped crops? Some canny crowd work to get it picked and shipped without extra steps (and before triggering rush pricing?)

[@Speak_Out_and_Remove_All_Doubt]

'Neuromorphic computing', we want AI to think and mimic like a human brain, future projects are called Skynet.

[@Capeau]

combine this with mobileye = winning

[@SebastianSzkoda]

Voice is out of sync with with vision, so little mistake, so annoying. Uniq interview with great guest, super easy to fix, will live its internet life broken, making viewers headaches looking at the guest.

[@thatguy7595]

annoying yes but you can just listen to this video, there's nothing to see

[@orlovskyconsultinggbr2849]

I think this Intel guy looks kind of desperate.....

[@orlovskyconsultinggbr2849]

@TROY DONOVAN wow so i was right about my guess , cool!