Hacker Newsnew | past | comments | ask | show | jobs | submit | kcb's commentslogin

Because the initial announcement included none of that... it wasn't addressed at all until the harsh sentiment.

It still hasn't been addressed. They walked back half of their wholly unreasonable position in an attempt to legitimize the other half.

Then shouldn't we celebrate the victory, drop it, and move on?

Victory is my device and its OS working the same way it always worked and the way it worked when I bought it.

Just don't install the OS updates then.

Inference isn't really that expensive, its the training of new foundational models that is. With whatever highly optimized setup the big providers are using, they should be able to pack quite a lot of concurrent users onto a deployment of a model. Just think too, it's very possible their use case would be served just fine by a 100B model deployed to a $4,000 DGX Spark.

It's glass...

Another factor, it's not just GPUs it's the full hardware stack. https://static.tweaktown.com/news/1/1/110521_2_nvidia-update...

CUDA has had managed memory that pages between VRAM and system RAM for a decade. Problem is doing so is unusably slow for AI purposes. Seems like an unnecessary layer here.

That slowness is almost useful. It makes the failure mode obvious instead of letting a 'transparent' layer hide it until some sloppy alloc or tensor blowup starts paging through system RAM or NVMe and the whole job turns into a smoke test for your storage stack.

For actual training, explicit sharding and RAM mapping are ugly, but at least you can see where the pressure is and reason about it. 'Transparent' often just means performance falls off a cliff and now debugging it sucks.


But it has chemicals in it...


i wouldn’t last a day as a food chemist


The 100 class Nvidia chips are targeted at training. With Nvidia buying Groq it will further move in that direction.


There's already videos of US/Israeli jets over tehran dropping guided bombs.


Mining is all compute and no IO. Training particularly is heavy compute and insane IO.


it's just HN getting baited for the 100th time by an electrek article.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: