Putting aside the speed of code that runs, Windows often has a load of things you don't want wasting CPU resources - in a corporate environment, you can often sense when they kick in, as the fans whir and certain things get subtly slower.
They put all that effort into making it fast and nice (and it's lightyears better than it was) yet it still struggles with one of the simplest things: getting out of your way!
Can you explain how Ubuntu is less lean than Debian?
The only difference that people would notice are snaps, which are slow for starting but it shouldn't keep spinning up pointless services like Windows does.
I've identical machines running latest Ubuntu and Debian. The difference in speed & size is very noticeable, for example when you do update or look at memory usage.
Ubuntu has a bunch of apps and services running in background that slow down the system. A lot of stuff related to their way of doing packaging or containers or cloud init.
Is Windows Terminal not fast for the features that they have? I didn't hear anything about PHDs but I am curious about your mention of being proven completely false, PHDs and the story behind that.
Whilst I know this was the M$ PM being deliberately obtuse here, in some cases modern SaaS marketing has elevated previously routine CS topics to "PhD material" to the layperson for a sale - and it's quite a shame when the propaganda drips through to tech-adjacent people.
AMD has a new chip for LLMs, right? Can that also be used for things like Blender rendering? I like the idea of having plenty of general purpose cores, but after around 16 it starts to feel like I need the AI stuff more now.
It seems inefficient to have all of the parallel stuff like graphics/AI in what amounts to basically a second computer that plugs in.
I am guessing that Apple has a lead in this stuff. Are Intel or AMD copying them at all?
> I am guessing that Apple has a lead in this stuff. Are Intel or AMD copying them at all?
Not exactly.
The AMD chip you are thinking of is the MI300, and it was originally a giant server APU with 24 cores. It was marketed as a HPC supercomputer chip.
Intel had something similar, Falcon Shores, a big APU.
...And as it turns out, almost no one cared. The ML industry doesn't want unified memory APUs, an inference chip, a training chip, a bigger training chip, and so on. They want Nvidia GPUs. Everything is already coded to work with this "second computer that plugs in," as you said, so thats what customers want. Even Nvidia themself seems to be spinning their wheels with grace-hopper.
So AMD spun a MI300 into a variant with an even bigger GPU and no CPU. And they emulated CUDA (a work in progress).
Intel canceled Falcon shores and redid it as a pure GPU, with their other AI IP (Gaudi) baked in like tensor cores. With a 2 year delay. And they are moving heaven and earth to support PyTorch and other frameworks out-of-the-box.
I can't emphasize how disruptive this is, especially for Intel. Silicon takes years and 9+ figures to design. They would not have made these moves if there was even modest short-term interest in unified memory chips for AI.
"Select models of [AMD Ryzen™ 7040 series processors] offer the new XDNA architecture with built in AI Engines, the first dedicated AI processing silicon on a Windows x86 processor."
Apple Silicon Macs only have a lead at a substantial markup over computers without modern dGPUs, and if Intel has a lead anywhere in hardware, they're keeping it well hidden. We're not saying that too loudly, though, because latest gen AMD prices could stand to be a little more competitive.
They put all that effort into making it fast and nice (and it's lightyears better than it was) yet it still struggles with one of the simplest things: getting out of your way!