There's no getting Huawei from 5G – Chinese giant joins Qualcomm, Samsung in bunging high-speed comms in mobe brains Huawei, Qualcomm, and Samsung were at the IFA consumer tech marketing fest in Germany this week to publicly tear the wraps off their upcoming mobile system-on-chips with built-in 5G modems. This integration means a load of forthcoming phones and gadgets using these processors will come with 5G support as standard, in theory. …
"Up to".
I don't know, but the amount of disinformation about 5g floating around seems to exceed the actual information. I think "wait and see" is a good strategy.
The first 4g phones had awful battery life (nb I know these are not transmitters) and it rapidly improved. Technology is pretty good up to the 5GHz band used in wifi. I wouldn't be surprised if transmitters at higher frequencies started off less efficient, but AIUI they will be small and short range.
Last week I was disconnected from electronic reality in a place where the occasional 4g signal made its way across from Wales. More, smaller masts that improve signal reception in areas currently badly served might have beneficial effects in reducing transport and commuting costs.
But if they save 1W on each phone connected to them, that could easily become an overall win.
Such things really don't matter in the grand scheme of things. Radio masts are far beyond 1KW and nobody cares about them.
Infrastructure is a power-hungry thing any way you look at it. Hell, I have more than 5KW of network switches alone in a small prep school.
This post has been deleted by its author
It does the thinking for you so you don't have to! Biggest USP of modern phone.
Seriously, it's logic circuitry optimised for executing those "deep learning" algorithms that do all the clever stuff people aren't clever enough to code, so they ask a computer to learn the task instead. Face/voice recognition are the most obvious applications on a phone these days.
AIUI it is more that "neural processing" needs a lot of relatively low precision calculations. If you can replace a 64 or 80 bit FPU with 32 or 16 bit ones, you can do a lot more calculating for the same energy consumption.
It's basically adapting the compute power to the job.
"AIUI it is more that "neural processing" needs a lot of relatively low precision calculations. If you can replace a 64 or 80 bit FPU with 32 or 16 bit ones, you can do a lot more calculating for the same energy consumption."
OK, so in other words, they have super-fast chips for super-fast processing, OK chips for mundane work, and 20-years-ago-architecture chips for low-precision tasks. But for marketing purposes they get called 'Neural processing' because that sounds AI-ey and hip.
Excellent work!
No, and I didn't say that.
The "neural processing units" have many small FPUs that work in parallel. Given that actual neurons behave a bit like something that takes multiple low-resolution pcm inputs, does some lowish-resolution processing and produces a low resolution pcm output, and that neurones work in parallel, there is at least some compsci merit in calling it an NPU.
20 years ago there was no possible way of making anything like that. Some years before (in the 1990s) I recall a lecture in which someone said that using the largest supercomputer then available it should be practical to simulate the brain of a cockroach, but that from a neural simulation point of view you were looking at something like a seaslug (about 26 neurons IIRC).
This post has been deleted by its author
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
