[velox_io]

I'm quite interested in what Nvidia is planning. Nvidia purchased Mellanox last year (they make high-end network gear often used in compute clusters). Nvidia is very involved in machine learning with their GPUs. The only missing part of puzzle is CPUs which was either Intel or AMD (a direct competitor). ARM changes things and means they're not dependant on those companies (and the x86 licensing clusterfcuk preventing newcomers), and they have some room to tailor it to fit their needs (like Amazon & Google recently).

These definitely are exciting times. Most Brits born in the 80's and 90's will have used Acorn computers in school, no one predicted that what would grow out of it (they weren't particularly fast).

[russellbeattie]

You seem to have overlooked the Tegra X1, which powers 55+ million Nintendo Switches. Nvidia seems to already be invested in the ARM CPU market and doing relatively well. Their X2 powers the new Shield devices and who knows what else in the future.

Yes, that's a very tiny number compared to the billions of CPUs made by Intel, Apple, AMD, Qualcomm and Samsung, but the point is, they are already doing ARM CPU integration and have carved out a decent niche for themselves.

[hajile]

Also, the X1 uses an ancient A57. To enumerate, since then, we've had A72,A73, A74, A75, A76, A77, and the announced A78 and (interesting named) X1.

That's 7 major generations behind the curve.

[saagarjha]

Video game CPUs aren’t generally where the puck is going, sadly. Cell was interesting but other than that we’ve had a history of picking a semi-popular architecture and usually some old chip (Nintendo) or just a normal x86.

[cbsks]

Nvidia has already embraced ARM, for example the Jetson machine learning platform is based on custom ARM SoCs: https://www.nvidia.com/en-us/autonomous-machines/embedded-sy...

[abainbridge]

> they weren't particularly fast

I remember them being amazingly fast. I was comparing them to Amigas in 1987 though. The Motorola 68000 in the Amiga 500 was less than 1.0 dhrystone MIPS, while the Archemedes A3000 was about 4.0. For reference, the machines cost about £499 and £599 respectively in 1987.

As a result the frame rate, poly count and screen size in Zarch on the Archemedes was higher than Starglider II on the Amiga :-). Look https://youtu.be/MNXypBxNGMo?t=36, https://youtu.be/edDLqlG4quw?t=132

[tokamak-teapot]

I remember switching on and being at the desktop in .. I don’t know, 5 seconds? Consistently as well. No random waits.

Apps opened so fast you couldn’t perceive a delay between double clicking and then appearing. New windows for apps also instant.

The only times I noticed waiting was when doing something that legitimately seemed like I should have to wait. A big operation in a paint program. Starting up a game.

I think this consistent ‘zero delay’ came through firstly the fact that lots of the OS was in ROM, then the tight design and coding of the OS, then perhaps that the ARM CPU was pretty fast.

Running RISC OS on a Raspberry PI natively it’s still just as snappy for normal operations. Things that take significant CPU are also now fast thanks to the several hundred fold increase in MHz.

I’d love to see another ‘personal computer’ OS appear which focuses on making the UX feel like it’s all working hard real time (or as close as it can) so that we can get back to the joy of feeling that the machine is consistently predictable in how it responds. Today’s ‘personal’ OSes feel more like driving a car with automatic transmission, steer-by-wire and several mechanical problems that cause it to randomly fail to respond in the expected time or just do something entirely different.

[ksec]

Nvidia is partnering with Ampere on ARM, and AMD on x86 (!). I would not be surprised if Nvidia made an offer down the line once they figure out whether Ampere is worth the money.

Right now, getting Apps ported to ARM seems a lot more likely than AI / ML porting their work off CUDA. And I am thinking if CUDA will end up like Excel for Microsoft. It is not that we cant move off Excel, but the code and value running on CUDA / Excel will not be worth the effort porting away from it.