You're correct, the base (non-Pro, non-Max) chips are still not great for local LLMs. M5 and M6 both improved prefill speed/time-to-first-token over prior gens, but bw is still a bottleneck.

There's rumored to be no M6 Pro or M6 Max coming, i.e. the higher bandwidth chips will skip a gen. M7 is also rumored to have better memory bandwidth across the range.

To the best of my recollection, the GeForce 256 was the first one that was marketed as a “GPU” and began to popularize that term in the industry.

Specifically because it did transformation and lighting calcs in (non-programmable until GeForce 3) hardware, instead of in software on the main CPU.

Yes, NVIDIA made 3D accelerators before the GeForce line. I have many of them in my collection. But (IIRC) it did not market them as “GPUs”. That term wasn’t really used yet.

"The GeForce 256, which nVidia calls the world's first graphics processing unit, will be the successor to its TNT 2 line of graphics chips." https://web.archive.org/web/20241023155821/https://money.cnn...

> (non-programmable until GeForce 3) hardware

I think the term you're looking for is fixed-function hardware.

Funnily enough, I already “found” that term in my brain and then decided to write “non-programmable” instead, in the hopes of being a bit clearer to readers who don’t know this space and its specialized terms already.

But thank you for condescending anyway.

Got it. Well, to be honest the marketing distinction isn't terribly interesting. But it would still be cool to see how the early GPUs (even if not called that) evolved.

Doing all those calculations in hardware (the distinction the marketing was based on) seemed like a pretty big deal at the time, at least within the context of consumer PC graphics cards. (It had long existed in the high-end workstation space, SGI etc. but the workstation and consumer PC ecosystems were more separated than they are these days.)

Which is kind of funny because the 1996 N64 RSP was a programmable GPU with Hardware T&L.

Yep, developed by/in partnership with SGI!

There were many periods in the 80s and 90s where console graphics hardware was kinda ahead of (even high-end/enthusiast) PC graphics hardware in certain ways that mattered for games of the time. Hasn't really been the case for a couple decades now, I guess.

I was gonna say hey that isn't true, the PS3 had hardware that was ahead of PC graphics hardware, and then I realized that was a couple decades ago. Oof.

How? Once they realized Cell isn't going to cut it PS3 just included Nvidia graphics chip.

Yeah, Cell was a disaster, and it showed Sony leadership at the time was actually clueless about big reasons the PSX and PS2 were so successful (it's the developers dummies!)

Really the PS3's only saving grace was it was a great deal for a Blu-ray player.

xbox 360 would have been better example, xenos with its unified shaders was in some ways more advanced than its contemporary desktop gpus at launch.

The Elon analogy is good.

That said, some of the people posting slop projects as their own probably idolize him and would embrace the comparison!

(Code)named after https://en.wikipedia.org/wiki/Coppermine_River , apparently.

POD was pretty rad :) Thanks for the reminder of it (and all the MMX marketing around it.)

I read those tutorials as well! Around the same age, in '96 or '97, when I first got online, in the UK. And it might have been someone's "translation" of them into C, although I do recall reading the Pascal originals as well.

Namedropping another contemporary reference, I probably downloaded them from the Hornet Archive or a mirror.

They were a major step along the short but satisfying road to my first 3D starfield using Borland C++ (acquired via a magazine coverdisc). I felt so l33t.