Framing It: 3.3.2020

Someone does something the wrong way and I get mad about it, 5G is coming, and so is 5nm, Vatican responds to AI, and Efficiency at Scale

Efficiency goes up at scale, who knew?

Apparently data centers all look like 90s raves with the blacklights turned waaaay up. Image: Josh Edelson / Agence France-Presse / Getty Images

• Like it or not, the last decade of computing can be defined by The Cloud. Need a server? Amazon’s got it. Need to store multiple Terabytes of data? Throw it in the cloud.
• At the beginning of the decade, some wondered if this increase in computation would mean greater energy usage. This would, in turn, mean greater emissions.
• A new study revealed while computing output jumped sixfold from 2010 to 2018, energy consumption rose only six percent. Thus, the concerns that large data centers would put more stress on the grid is overstated.
• Is this part of a concerted effort of cloud providers? No, not really; computers of today are capable of performing more calculations per unit of energy than computers from a decade ago. It says as much in the paper. The only takeaway, really, is that data center’s aren’t more efficient, it’s just that they’re the same size, and using newer hardware.

Let’s Write All The Music

One of the biggest problems for content creators is music licensing, and copyright notices on videos. The music publishing industry has a lot of lawyers, and if you dare use something that even sounds like a song you don’t have the rights to, prepare for the takedown notice. A few programmers had an idea though; why not write every possible melody, and release into the public domain?

• Here’s what this actually is: It’s multiple Terabytes of MIDI files, each of them eight notes, twelve beats, with all the notes in a single octave. If you want to write a melody with those parameters, it’s already on a hard disk somewhere.
• Writing ‘all the melodies’ is a problem of combimetrics; There are twelve notes in an octave (13 if you count a rest as a ‘note’, which is how you would program this). If you want a melody that covers two octaves (the Star Spangled Banner is a hard song to sing and is only an octave and a half) that’s 16 bars long, that’s an immense amount of data. Too much for one hard drive, but that’s why cloud storage exists, I guess.
• Wait, holy hell, these guys are doing it wrong. You don’t have to fix the melodies in a medium for copyright to apply; the output of a computer program can be copyrighted, you only need to print out the first and last fifty pages or something and send them into the copyright office. You could do this without a hard drive.
• This is an academic exercise, because someone will need to defend this in a court of law. Let’s go over the weird music rights cases like My Sweet Lord, the time John Fogarty was sued for sounding like John Fogarty, and the time Neil Young was sued for not sounding like Neil Young.
• Of course there’s a TED talk:

5G Problems Ahead!

The biggest thing in technology now is 5G, and not just because lobbyists are bribing your congress critters for 5G regulations, or because 5G press releases are bombarding the inboxes of your favorite tech journalists. There’s a new problem for 5G — the silicon required for 5G is hitting its limits, eetimes reports.

• Companies developing 5G are hitting their theoretical limits in designing 5G front-end modules
• The solution to this is weirder substrates. gallium nitride (GaN), gallium arsenide (GaAs), and silicon carbide (SiC) are all being touted as replacements.
• Right now, there’s a small market for making GaN wafers for semiconductors, as GaN MOSFETS are making their way into very small and very efficient switching power supplies. But if you’re not using straight silicon, the price of a ‘weird’ wafer will always be higher.

This is actually the definition of Moore’s Law, and nothing else can be inferred from Moore’s observation.

Over the last forty years, the feature size on semiconductors has been shrinking. We’ve gone from multiple-micron sized transistors in really old CPUs to 0.8 um (800 nm) transistors in the original Pentium processor to single digit nanometer features on modern CPUs. No, the smallest feature on modern CPUs isn’t 3nm or 5nm or 7nm; that’s just marketing. But stuff is getting smaller.

• Now the battle for 5nm and 3nm is heating up, several foundries are producing 5nm chips, and companies must face a choice of designing in 5nm or waiting until the 3nm process comes online.
• Samsung and TSMC are currently producing 7nm chips with their ‘finFET’ process, with 5nm chips coming ‘later this year’.
• Around 2022, both companies will be producing 3nm chips, using different technologies. Intel, though? Uhhhh….

The Holy See is on the forefront of technology. They have payphones that take credit cards.

I gave up blockchain for Lent.

• Lent began last Wednesday, which means last Tuesday was Faschnaut day. Did you eat a donut? This is your excuse to eat a donut.
• But nevermind the donuts! The Vatican has published a guideline for ethical AI! And it’s being endorsed by Microsoft and IBM.
• The Vatican’s “Rome Call for AI Ethics” is a pledge to develop artificial intelligence in a way that protects all people and the planet. IBM and Microsoft are among the first companies to sign the pledge.
• The pledge also asks for regulations on facial recognition and, “ AI that safeguards the rights of all humans, especially the underprivileged.”
• I don’t know about you, but getting theologians into the AI discussion is going to be fascinating. These are the guys who came up with the idea of a homunculus, and you don’t even want to know what happens to a victim of cannibalism after the Last Judgement.