Comment by DaveCharlieLen | Hacker News Reader

Comment by DaveCharlieLen | Hacker News Reader

Parent Full thread

> I need an ARM64 machine. Period.

>After spending time on Apple’s M1/M2 Macs (coming from a large x86_64 desktop), going back to x86_64 feels like a regression, both in performance and battery life. Unfortunately, there’s currently no FreeBSD-supported (or even Linux, as far as I can tell) ARM64 laptop that truly rivals Apple Silicon. I really hope Framework or someone else changes that in the coming years.

I thought the parent maybe was off base, but that is literally all the reasoning.

For guidance the arch doesn't impact battery life or performance as much as a specific process node and model. There are slow as shit x86 chips from VIA(maybe? I just retired a node that ran on one.) and fast ARM64 chips, but any high end Ryzen will blow an ARM64 chip out of the water on most real benchmarks I run(HPC stuff, lotsa matrix multiplications and such(CUDA and double float performance is too much premium and yes I have done alot of profiling and benchmarking.))

amd64 definitely beats any aarch64 when it comes to performance, but the trick Apple (and if proper Linux driver support ever lands, Qualcomm) pull is that they get 80% of amd64 performance at 25% of the battery cost.

Not a lot of people are running intensive calculations all day. My day-to-day usage is 90% thinking/planning/writing/reviewing code and maybe 10% of time spent on running that software.

There are two factors that blow up Apple's aarch64 chips above the competition: false comparisons ($1000 cobbled-togerher PCs versus $3000 Macbooks) and Enterprise Hardware vendors like Dell and Lenovo raising their prices to match Apple's without the hardware or software to match (i.e. $2000 Thinkpad workstation laptops that are slow, overheat, and draw huge amounts of power, but are priced as "workstations" because they have a GPU with fp64 support).

For the people who spend a lot of money on a computer for the first time, Apple's aarch64 chips are more likely to be a good deal than the treacherous landscape of high-end laptops and prebuilts. Even in Apple's price, competing range vendors still dare sell 1080p60 displays. Buying desktop/workstation units, 10gbps ports seem to be made of solid gold and actual, full-speed Thunderbolt/USB4 support is restricted to one specific port, often placed at an inconvenient location.

You can exceed Apple's performance in general and their performance-per-dollar as well, but it requires time and attention to look through marketing bullshits and hidden-away spec sheets. That's especially the case now that every store page has four popups about "AI" and "NPU" and "Copilot" for some absurd reason.

> I really hope Framework or someone else changes that in the coming years.

Building a custom CPU like Apple Silicon is plain impossible for a company like Framework. The best they can do is have support for ARM motherboards and sell those with the best CPUs on the open market.

I’m hearing good things about Snapdragon nowadays. I’ll certainly give it a try in a bit.

Specifically the M series from Apple have a very wide, very fast interface to DRAM, which is connected to DRAM chips soldered basically next to the CPU. That makes it possible to use the entire unified RAM as the GPU RAM, and reasonably run decent ML models (for code, text, audio, pictures) locally. No CUDA, no kilowatt power supplies. This is the real differentiator.

Notably, this is also Qualcomm's X2 architecture.

> That makes it possible to use the entire unified RAM as the GPU RAM, and reasonably run decent ML models (for code, text, audio, pictures) locally. No CUDA, no kilowatt power supplies. This is the real differentiator.

That might be relevant and a differentiator in your circles; it is entirely irrelevant in mine. Plain basic integer performance wins here.

>> performance and battery life

> any high end Ryzen will blow an ARM64 chip out of the water

I'm very skeptical about this. I've read that many benchmarks show ~40% better performance per watt on ARM than the best x64 machines. Do you have any sources that say differently?

Performance per watt isn't the same as performance. On the high end (think Threadripper), amd64 still wins most performance tests by having many high-performance cores working all at once (at the cost of single-core performance).

I disagree with "any high-end Ryzen" blowing an ARM64 chip out of the water, though, it takes quite a beefy CPU to beat an M4 Max.

Reply on news.ycombinator.com