Edge Computing, Explained

Sources:

FCC: This video is sponsored by Brilliant. Links may be affiliate - thanks for your support!

[KenJee_ds]

I'm legit sweating thinking about my driving history being explored

[LiamNebe]

Jordan, I think you might need to double check your volume levels. I've noticed the last couple of vids have had pretty low levels compared to other YouTube videos. Great video otherwise

[JordanHarrod]

And stay tuned for the rest of the hardware videos coming out this month!

[nacoran]

I once did a bandwidth test on my computer when they were working on fixing some problems and the filter, for a moment, wasn't in place. My bandwidth test came back saying I had infinite bandwidth.

I thought it was funny that someone had apparently programmed it so that at a certain speed it said infinite. Somewhere I have a screen shot of it.

[srinitumuluru]

I am not an expert, but there is difference in voice output of this vid compared to other videos of your channel. Please note. Thanks 🙏

[trueriver]

1:40 actually as a boomer i saw cloud computing not as a game changer but as history repeating itself unhelpfully.

I used to use a device that looked rather like an all in one computer but it was only a "terminal". All the programs ran on a remote computer called a "mainframe" and we were connected to it by dedicated wires, so nothing was mobile. We needed a t technician to do wiring to use a terminal in an office that was not wired up.

But the mainframe could be halfway across the country -- even before the internet we had national-scale networks. On a bad day my broadband runs slower than the 1980's connections to the mainframe that i worked on. That was because the terminal multiplexor network was private and therefore not subject to other companies' surges in demand.

The terminal had graphics: still pics on my Honeywell graphics terminal had far better resolution than the first PCs.

The only practical difference between cloud and mainframe computing is the option to go wireless and apart from that I saw it as a big step backwards. It was using a gigaflop computer to do no more than a graphics
-enabled terminal did in the eighties.

Even worse when you emulate a VT100 on unix/linux -- that was a terminal for one specific mainframe that became popular on Unix mainframes, and a huge step down from my Honeywell at the office.

Only more recently with grid computing and now with edge computing is the computing capability of both devices actually used in tandem. In my opinion, that does not so much justify the the cloud as reveal it to have been a blind alley that we needed to back out of on the way to here.

[himanshu]

Hello can you turn on autogenerated captions on your videos? Your voice is really low on my device and I can't hear it lol

[yetinaut]

OK I totally understand that by drawing attention to a deficit I am in some ways complicit in that deficit. But We have to be on balance here, and I would just like to celebrate, however shallow, that I love the fact that this channel demonstrates a very intelligent woman of color, specifically a black woman who is involved in the STEAM fields. Jordan you are an inspiration for countless followers be they black, latino, asian, or whathaveyou, I love not only the fact that aside from all those demarcations are you making truly original and educational content that is universal, but also are you doing the important but also reasonably necessary and hopefully unnecessary work of proving, however unjusty, that you have the background and intelligence of creating this kind of content. I love you and I forever will be grateful for people as powerful as you. Keep shining!

[estebanduran]

Fantastic video! Another application of edge computing is spacecraft. These systems have limited compute, limited bandwidth, and must respond in real-time to the environment. These limitations are greatly exacerbated as the spacecraft go deeper and deeper into space. So having models with a smaller memory footprint and require less data and compute would be huge for spacecraft autonomy.

[benmwaba]

Being from a third world country, latency and bandwidth are a huge problem for real

[cineblazer]

Apple seems to be a clear leader in this space. I think it's super cool that they're including more and more optimized inference-dedicated hardware in their SOCs! I was talking to my dad the other day who just got an iPhone 12 Mini, and we ended up doing a speed test of a style transfer app he'd first downloaded a few years ago on an iPhone 8. The app runs ~4x as fast on the new phone than the old one!

[WDC_OSA]

Definitely going to look a bit further into edge computing. This is something I hadn't heard about before, and it sounds pretty interesting. Thanks!

[dekev]

I'm doing my masters project and thesis in neural networks Accelerators implementation on FPGAs.

[Micetticat]

I like hardware acceleration and I like top secret projects!

[w.maximilliandejohnsonbour]

Nice video on the subject edge computing. To be honest edge computing has been around for thousands if not millions of years in nature. Example look at the Octopus. Most of it's computing takes place at the edge of it's tentacles and not from it's brain.

[Bitcoinmansa]

Its nice learning about computers from a melinated queen !! thanks for the value

[theondono]

If we’re being nitpicky, edge computing and local computing are different concepts.

Edge computing generally refers to something similar to CDN but for ML systems. In the same way that it’s a bad idea to forward all your querys to googles servers in the west coast, it’s a bad idea to have all your users devices pointing at a specific location. Ideally you’d want yo replicate your system everywhere to provide low latency responses. The obvious conundrum is how to do that without exposing your IP, since some of these models are worth a lot of money.

There’s some companies trying to create a model similar to some DSL ISPs, were there’s a provider that essentially rents rack space for your custom built accelerators.

[jamesanthony]

On the autonomous car thing, another fix for latency is mesh networking, where each car acts as a wireless repeater. You still need sufficient bandwidth of course, but at least it deals with the issue of wifi deadspots during rush hour.

[paperlanes]

Just when you think cloud computing was getting boring and BAMM!!😆

Any recommendations on where I can learn more on cloud architecture?

[pavithraes]

Is "on-device AI" the same as edge computing? Or, is it a subset of edge computing...? Or are they totally different things?

I couldn't find many resources on this, and I'm slightly confused because this is the first time I'm hearing about edge computing (feels like I've been living under a rock 🙈)