This case study is obviously also an advertisement for a specific brand and type of 3d printing filament. I still found it interesting enough to share.
This is the core of why I've enjoyed my Garmin watch. It tells time, displays texts, has a quick and easy flashlight, tracks workouts, even acts as a dive computer. I get minimum two weeks of usage between charges. It's not something I have to think about on the daily. Now, on the rare occasion I'm not wearing it, I notice its absence.
Device trees are certainly less portable across hardware, and more difficult to create than just booting a generic kernel on a machine that supports ACPI. However, the device tree is shipped as part of the kernel, and as long as the bootloader correctly loads the DTB and kernel, it's no more difficult to swap out the userspace portion/rootfs than any other Linux system, generally.
Interesting to see the adjective "small" used to describe a single binary in the same amount of space used by an entire embedded Linux-based operating system.
Probably to interface with existing cars by replacing the engine, retaining everything after, including the existing transmission, differential(s), and suspension.
HN is overloaded with AI stuff, its hard to break through all the noise. I say this as someone very interested in AI. Even I skip some links because its just too much.
I see it making claims about 10x efficiency, but how is tokens / second / watt? The only machines I've seen with the memory bandwidth to effectively do local inference are Mx arm chips on mac.
because it's not faster than the Ryzen 395's GPU. power efficiency doesn't matter as much as TTFT for desktop users, especially when they're tasking their AMD box as a dedicated inference machine.
some older pre-395 AMD articles suggested it'd be possible to use the NPU for prefill and the GPU for decoding and this would be faster than using either alone, but we have yet to see that (even on Windows) for any usefully sized models, just toys like LLaMA-8B.
On average according to Geekbench, the M5 compared to the 9950X is ~17% faster in single thread performance and ~30% slower in multithread performance.
Individual benchmarks tell the bigger picture. These two are optimized for different use cases, with Apple heavily leaning towards low latency single thread throughput with low sustained power usage.
I think the point of the line of questioning is to illustrate that "tools" like a code interpreter act as scratch space for models to do work in, because the reasoning/thinking process has limitations much like our own.
