__simt__

Once in a while something happens that reminds me how much growing in engineering involves apprenticeship.

I’ve gotten good at demanding time from my seniors, so don’t worry about me, but I can forget to carve out time for my juniors.

I think (hope?) I just had a key interaction for someone.

It’s a Swedish company incorporated in Luxembourg, though. I can’t believe it’s not American… Capitalism.

Maybe that's all moot, at least for now, since 5090 is pretty svelte.

It doesn't mean the chassis has to be huge, necessarily.

What needs to change is the design ethos needs to change to wrap the big GPU instead of the CPU mainboard.

Mini mainboard -> mega GPU -> case wrapping that.

IMO anything else is a anachronistic design dinosaur.

I see it a different way. PC desktops don't have much a reason to keep existing *except* for access to this kind of hardware.

Why make a chassis that can't hold the hardware that justifies the perpetuation of the concept of a chassis? Else, you'd be better off with an integrated design.

Best trivia. 🤯

I wouldn’t expect to see high end cards ever be bus-powered again. That looks like the opposite of the trend.

But a 1080’s worth of performance could probably be bus-powered now or soon. The 5050 should be around there even at the 75W edge connector limit. Roughly.

GeForce 5000 is the first product NV has released that I haven't worked on, in more than 22 years.

It's been fun to read the various bits of writing they put out and infer what's actually been done in this one.

Hopefully we get a CUDA toolkit soon. It's been like 2 years since the last major drop.