If someone can round up 51% of Bitcoin's miners and hold them at gunpoint, then they have full control over Bitcoin's rules.

Well over 51% of bitcoin mining happens in China, so Xi Jinping can probably change Bitcoin's rules tomorrow if he desires.

> if someone can round up 51% of Bitcoin's miners and hold them at gunpoint, then they have full control over Bitcoin's rules.

This is not true. Most of the rules are maintained using cryptography and game theory.

For someone to break bitcoin, they need to be irrational. i.e spend more than they will benefit from an attack.

And even if such an attack happens, it's very trivial to roll back & undo the damages.

It's not a democracy, it's a game.

Technically FLP impossibility only applies to deterministic algorithms, not randomized ones, otherwise no consensus protocol would work. So you could imagine a complicated technical solution that could work (eg. treating everything as a big Raft cluster and using Raft's membership changes every time a device is connected / disconnected).

In practice the solution is to realize that failures are rare (especially compared to events like the soldier simply misplacing the card), and to have a trusted human available to resolve any issues that do occur.

These instructions are not for the Neural Engine. That is a separate hardware block outside of the CPUs. Apple refers to this feature as "AMX" in their marketing documentation.

While AMX could be used for deep learning in a pinch despite the lack of support for common formats like fp16 and int8, I suspect Apple had some other use cases in mind as well. For example, 64-bit float support is expensive and generally useless for ML. However, they are useful (though not necessarily required) in problems such as bundle adjustment that may appear in the context of frameworks like ARKit.

Most recent tape standards have multiple "bands" and "wraps" placed in parallel. The head reads only one wrap at a time, so it takes many passes to read the whole tape. For example an LTO-8 tape has 52 wraps each within 4 bands, requiring 208 passes to read completely.

Why not just have multiple heads then?

There are. The tradeoff of speed vs. complexity is currently sitting at 32 heads.

It would be like downloading a file using GetRight back in the day, where you'd have one thread downloading the 0-25% chunk, one downloading the 25-50% chunk, and so on.

You could hypothetically do that, for sure, but the software basically is derived from the tape era where you'd just have one logical stream coming out of the tape.

Positioning the tape should become harder, I suppose.

Also, I suppose they limit the read bandwidth with something like what a single Infiniband connection would support; few disk arrays support much higher speeds.

It may be interesting to look at a more completely labeled die image (eg. a recent AMD laptop chip [1]).

I/O controllers, various PHYs, the display pipeline and the media engine account for a significant fraction of chip.

We also know the M1 has an integrated SSD controller, ISP and Thunderbolt which probably take a good chunk of room too.

[1] https://potatopc.net/amd-renoir-die-annotation-raises-hopes-...