Thread regarding Oracle Corp. layoffs

@a5 Training and running LLMs, aka Large Language Models, aka "generative AI" require GPU computing power that NVidia produces. NVidia manages to make newer, far, far faster chips every year. After maybe 2 or 3 years they become obsolete as the newer chips are much faster.

It's not unlike the 90s when processing power meant a computer was junk in 3 years.

Hi @a4 - can you clarify "A bunch of NVidia chips running LLM workloads? No way in he-l." - I am trying to figure out the angle. I am not well versed in this and this is an honest question. I do get the old processor world.

Useful life of compute resources depends heavily on the resource.

A server with just regular-ol-processor usage? Sure, I can believe that. You can use a server easily for 5 or 6 years, and new ones don't really give you anything all that more incredible than the old one. I used to work at a place that ran a machine for 10+ years. That was waaaay too long, but we did it.

A bunch of NVidia chips running LLM workloads? No way in he-l.

hyperscalers (both us and Meta as well as others) have extended useful lives (for example for servers its from 4 to 5 years) due to improvements in hardware reliability, software optimizations, and actual usage data. so his math does not add up. he's short orcl and that's reflected in this post. i am more concerned about massive debt we are issuing and the current interest rates and what happens if rates dip back to 1% or so. we'll see, he actually may be partially right, this is a massive gamble either way you slice it.