Rumors before the M2 release said that it used 4nm.
https://www.macrumors.com/2022/03/10/m2-macs-with-tsmc-4nm-process/
Apple says they use "second generation 5nm technology"
TSMC's website says they have 6 different 5nm nodes: N5, N5P, N5A N4, N4P, N4X
https://www.tsmc.com/english/dedicatedFoundry/technology/logic/l_5nm
So the M2 likely uses N5P, N4, or N4P. N4 and N4P are usually called 4nm in marketing material.
There's probably a leaker out there with more knowledge.
Yeah, Apple's vertical integration and volume is enviable.
It's the best option on the market right now and the most compatible one. The drivers are owned by the Linux Foundation, and there are no known hardware bugs with Intel GPUs, unlike with Intel CPUs. You have so much flexibility; Buying an Intel GPU doesn't prevent you from using another CPU, even GPU-less AMD Epyc CPUs that have the cheapest PCIe/$. All you need is a PCIe slot and you get all the benefits of Intel with none of the drawbacks.
I'm a bit of an AMD fanboy sometimes and I own AMD stock, but the A310 can't be beat for Jellyfin transcoding. If you really hate Intel, keep in mind that Intel probably loses money on every GPU they sell 😉
Yes, Graphene developers will not port Graphene to any other devices, because they're insecure. You'll need a new phone to use GrapheneOS.
It's possible that your phone is supported by CalyxOS, which has a lot of the same degoogled feature, but a little less compatibility.
Both chips are 20w class cpus, but the AMD cpu is much faster.
Apple CPUs don't report wattage, so it's a bit tricky to measure actual power consumption, but I can't imagine the AMD cpu uses 50% more power under load.
The Apple CPU might score some wins for idle power consumption though, considering the optimizations in MacOS, and the focus on power consumption across the whole system design.
I believe both M2 and Zen 5 use 4nm. 4nm is just a slightly improved 5nm, though. It's the same process node, not an entirely new process node like 3nm.
AMD's CPUs are faster and more power efficient on the same process node. (i.e. 5nm vs 5nm)
https://www.cpu-monkey.com/en/compare_cpu-amd_ryzen_ai_9_hx_370-vs-apple_m2
Apple just has a big budget to buy out TSMC process nodes a generation early, their designs and architectures aren't actually faster or more power efficient than AMD's x86 cpus.
https://www.macrumors.com/2023/02/22/apple-secures-tsmc-3nm-chips/
I think AMF is still faster/better quality than CPU transcoding, depending on the preset.
I guess that makes sense, but I wonder if it would be hard to get clean data out of the per-token confidence values. The LLM could be hallucinating, or it could just be generating bad grammar. It seems like it's hard enough already to get LLMs to distinguish between "killing processes" and murder, but maybe there could be some novel training and inference techniques that come up.
Options should also be taxed as income from labor for social security and income tax
I thought confidence levels were for image recognition? How do confidence levels work for transformer LLMs?
Sure, I was just explaining it because the whole 5/4 thing is confusing.