WWDC 2024

  • Thread starter Thread starter Joelist
  • Anyone can edit the first post of this thread WikiPost WikiPost
  • Start date Start date
Status
The first post of this thread is a WikiPost and can be edited by anyone with the appropiate permissions. Your edits will be public.
I’m sure there’s already a fairly high percentage of germanium in those substrates. Good for putting a hitch in the energy band so that you improve electron current by reducing electron-hole recombination.

What you want to do next is GaAs (gallium arsenide) or InP (indium phosphide).

I’m a lot of fun at parties.
I was thinking gallium nitride, high electron mobility transistors, and the like 🎉 (party thusly)
 
I was thinking gallium nitride, high electron mobility transistors, and the like 🎉 (party thusly)

GaN bandgap is 3 or 4 times higher than GaAs or Si, which makes it great for high voltage electronics, but not so great for microprocessors.

I can’t think of anyone who’s tried to make a GaN CPU, but I’ve actually made a GaAs CPU. (It was a dumb idea, but it worked. 1GHz at 100W in 1996, which was pretty good for the time. IPC was attrocious - something like 0.6, though lots of CPUs in those days were in that ballpark)
 
I’m sure there’s already a fairly high percentage of germanium in those substrates. Good for putting a hitch in the energy band so that you improve electron current by reducing electron-hole recombination.

What you want to do next is GaAs (gallium arsenide) or InP (indium phosphide).

I’m a lot of fun at parties.
I actually expected this response; not the specific content, but the qualitative nature ;).
 
I actually expected this response; not the specific content, but the qualitative nature ;).
when i used to interview engineers who were looking for jobs at AMD, I would ask them about what happens to the Fermi levels and bandgaps in semiconductors as you lower the temperature from 8k down to 4k. I didn’t ever expect them to get it right, but it was useful to see how they responded to the pressure.
 
I’m just hoping for the ANE being added to Metal, and Metal support for fast (2x or 4x) low-bit vector math; say 4 and 8-bit minifloat and integer (int2 at 8x if it’s techxmas). One of those “and the best part is that you can try it out on hardware you already have, the M4 iPad Pro” kind of announcements.

Also hope that Awni and mlx crew get some of the spotlight.

Oh, also, I’m praying the AI deployment isn’t a half-baked shitstorm like some of the recent other big players’.
 
Last edited:
when i used to interview engineers who were looking for jobs at AMD, I would ask them about what happens to the Fermi levels and bandgaps in semiconductors as you lower the temperature from 8k down to 4k. I didn’t ever expect them to get it right, but it was useful to see how they responded to the pressure.
What is the right way to respond to pressure in that situations in your opinion? I don't think I've ever come across any such questions when applying for a job (even though I know they're common) but I don't think I'd be able to say anything better than "I'd need to look that up" (which, I guess, it's better than making stuff up 😂).

Oh, also, I’m praying the AI deployment isn’t a half-baked shitstorm like some of the recent other big players’.
My concern as well. I hope they keep up the high-quality approach to AI they've been using on iOS so far and don't add a crappy useless LLM just because it's in fashion.
 
What is the right way to respond to pressure in that situations in your opinion? I don't think I've ever come across any such questions when applying for a job (even though I know they're common) but I don't think I'd be able to say anything better than "I'd need to look that up" (which, I guess, it's better than making stuff up 😂).

I think there are only wrong ways to answer. About a third of people I interviewed pulled the LLM technique of very confidently giving me the wrong answer. The reason it’s a good question for this is that a certain type of person hears it and thinks “oh, I am so smart! i see how to get the answer!” and they tell me the “the fermi level goes down because fewer electrons are excited into the valence band“ or whatever.

Some people who got the wrong answer were smart enough to add “but, of course, weird stuff happens near absolute zero, and there could be some sort of quantum effect or something I’m missing.” I‘d hire those people.

And some people just panicked and refused to think about it.

As long as you gave me some sort of logical thought process, and admit uncertainty, you were good. There were people who didn’t know what a Fermi level was, and I wouldn’t hire those people unless they had years of proven experience doing the job I was hiring them for. If you’re fresh out of school and looking for a job designing high-end CPUs, most of us thought you should know basic semiconductor physics.

I only ever had one person answer it correctly. It’s not something you’d ever need to know, and you have to have taken some grad-level quantum physics to really figure it out. So mostly it was to weed out people who think they know everything (but don’t), and to weed out people who would panic under pressure (and we were always under a ton of pressure).
 
Why not just use the supposed Einstein technique? Just invite a promising candidate to dinner, and if they salt their food before tasting it, look for someone else.
 
As long as you gave me some sort of logical thought process, and admit uncertainty, you were good.

That was key, even when a wrong or less than ideal answer was given. Being able to articulate assumptions made in answering questions was also important. Because many times the questions we'd ask were intentionally vague, or where not enough information is provided, or there could be multiple answers based on assumptions made, especially in a systems engineering realm.
 
I am oddly not attracted to that calculator. It is cute, but not particlarly useful.
 
What an underwhelming WWDC. When Math Notes is the most compelling thing they show it is pretty dull - Apple Intelligence was okay with the highlight being that it is not cloud based.

Announcing M4Pro and M4Max would have been more interesting.
 
What an underwhelming WWDC. When Math Notes is the most compelling thing they show it is pretty dull - Apple Intelligence was okay with the highlight being that it is not cloud based.

Announcing M4Pro and M4Max would have been more interesting.
i was just going to say this. Math Notes was the main thing that excited me.
 
Really disliked the crappy AI generated images shown in the Keynote. For a company like Apple that values great design so much, it's a shame that they seemed to be happy with the output of the AI models.
 
Beyond the whiz bang AI announcements there were a couple quality of life improvements - eg delayed sending of messages and yeah math notes and note taking. The AI stuff which … well … well just have to see - some of the improvements to Siri sound nice and overdue (again quality of life stuff). There was GPTK 2.0 but sadly no nuts and bolts Metal improvements announced in the keynote. Maybe the developer videos will have something. And of course no major iPad/Mac overhaul/deeper conjoining which wasn’t really expected but would’ve been nice. iPhone on the Mac is an interesting way to get around an app that refuses to bring their app the Mac. 🙃
 
Last edited:
It may have gone under the radar for many people, but I thought one of the most welcome announcements was the ability to use the Messages app over satellite when there's no other connection on iOS 18/iPhone 14 or higher. This feature was previously limited to contacting emergency services or roadside assistance, but now it's open to all messaging. Apple didn't say if there will be limitations on the new service or whether it would cost users anything, but I think it could come in handy in places with no connectivity, which isn't uncommon for me.
 
Back
Top