In 2003, Apple released a 64-bit dual-core 1.8Ghz system: Power Mac G5.

In 2023, Apple released a 64-bit dual-core 1.8Ghz system: Apple Watch Series 9.

The Watch is faster and has more RAM.

The G5 was too hot to put in a laptop. It'd use up S9's battery in under 2 minutes.

@kornel that's awesome, but I think you're missing the point here.

The Power Mac G5 was very difficult to wear on your wrist! I still have shoulder problems from those days.

@sysop408 @kornel Great workout though! 🤣
@kornel @jimbob tbh i’m surprised a current watch battery has enough electrons in it to run the fans on a G5 for 2 minutes, let alone the rest of it

@rfc6919 @kornel @jimbob Same!

Let's see … I have a cheap NiMH AAA in front of me; it lists its capacity as 900 mAh; at ~1.2 V that's pretty much 1 Wh. Spending that in 2 minutes gives 30 W. Maybe that could sustain the smallest G5 (CPU only) when it's mostly idling? AFAICT the watch in question holds ~1/3 of that.

As 2 minutes was given as a hard upper bound and seems in the right ballpark, I'd say it checks out, but there might be enough wiggle room to reduce that limit to under a minute :)

@rfc6919 @jimbob I calculated it for the best-case scenario of 30W power draw of the CPU alone. The G5 came with a 450W power supply — no match for the 1Wh watch battery.

@kornel @rfc6919 @jimbob My G5 (dual core) runs at 220 watts. You may attach very large displays that will use the same power draw again.

Difference: You probably won‘t be using a Watch in 20 years. You still may run the G5 using a current Debian.

@kornel @stroughtonsmith I really miss having an universal benchmark software, which can run on all old, new and egzotic hardware to be able to compare the speeds. For example there is none available for Apple Watch or tvOS at all

@gklka @kornel @stroughtonsmith Actually! I worked on something like that! And I want to somehow rejuvenate this project and add support for more platforms.

https://github.com/vashpan/isbench

GitHub - vashpan/isbench: A (VERY) simple benchmark aimed on maximum portability, to compare performance of various CPU accross bits and times

A (VERY) simple benchmark aimed on maximum portability, to compare performance of various CPU accross bits and times - vashpan/isbench

GitHub

@gklka @kornel @stroughtonsmith For example: modern CPUs (single core) seems to be around >200 times faster than 486 DX4 100 Mhz (benchmark is tuned against it, 100 pts means this 486 performance)

Other results:

ThinkPad X100e (2009): ~26 times faster
ZOTAC CI320 Nano (2015): ~93 times faster
MacBook Air M1 (2020): ~245 times faster
iMac 5K (2017) i7 7700K: ~243 times faster

@kkolakowski @kornel @stroughtonsmith super interesting! If I will have time, I try to compile it for tvOS/watchOS
@kornel Amazing how tech had evolved
@kornel I'm still sad they never figured out a way to make a G5 laptop.
@WooShell @kornel I’m very curious how dramatic the failures were.
@helianthropy @WooShell @kornel I'm imagining liquid cooling in a laptop form factor and a battery back that looked like a World War 2 infantry radio setup
@WooShell @kornel they did, they swapped the processor out for an intel one 😉
@kornel Good news is that we can put 2GHz system into smartwatch. Bad news is that people are putting 2GHz systems into smartwatch, that 2GHz CPUs are now required for displaying time :-(. You know, battery life still suffers with 2GHz CPUs.You can get month with bangle.js2, or day with Apple...
@pavel @kornel the good news is my Apple Watch does a lot more than tell the time.
@Strwpok @kornel Does Apple watch do anything that should require 2GHz CPU? (Does it even run doom? :-) )

@pavel @kornel

And if all the Apple Watch did was display time, that would be a problem. Good news is, that’s not the reality.

@kornel It’s amazing to think that my current watch is faster than my first Mac.… and weighs about 450 times less (fortunately!).
@kornel is this correct??? It’s crazy!!
@kornel The big difference is that 2003 machine was for creation and 2023 machine is for consumption.
@kornel @stroughtonsmith and 21 years before the G5, the ZX Spectrum came out with a 3.5 MHz Z80, an 8-color display and either 16 or 48 kilobytes of RAM. I'd say Moore's Law dramatically slowed (especially for single-core performance) around the time the G5 came out.
@Alphacheez Moore’s law is about transistor count per cost. Nothing to do with clock speed, and only indirectly with performance. (That said, it _has_ slowed.)
@kornel hold on, a watch should not be faster. It should be precise.

@bonkers @kornel You should consider an atomic clock wristwatch.

http://www.leapsecond.com/pages/atomic-bill/

First Atomic Clock Wristwatch

@kater_s @kornel shut up and take my money!
@kornel fantastic summarization. Well done!

@kornel It's been incredible watching these technological advancements over time.

As a kid, I never imagined Smartwatches would become as advanced as they are in my lifetime. I'm just hitting my 30s, and technology has already surpassed my childhood expectations.

@kornel @becomethewaifu Seems like just doing things better/right also works, one doesn't have to increase the specs.

Of course in this case part of the things being done better probably include the hardware design for the CPU as well.
@kornel These sorts of statistics are why I'm really annoyed at the number of apps that need a constant connection to a distant server in order to do basic functions, meanwhile the my highschool laptop which was *less powerful* could create and render 3D scenes in Blender just fine.