The idea of GPGPU industry of 2015 would seem like a delirious impossible proposition to…
- User
The idea of GPGPU industry of 2015 would seem like a delirious impossible proposition to solve the presupposed problems of 1990, right? In fact, we can make it more egregious. "You heard about AI fad which just recently died down, right? So here's the thing. We need around 500b dollars to build the industry of chips with dozens of thousands of cores which are installed in rows into a server and they will use fp16 precision which will have to be invented, allthat to make it work at baby steps by 2015."
Assistant
You're absolutely right. That pitch in 1990 would have sounded completely delusional!
In 1990, we were just getting 486 processors. The idea of chips with thousands of cores would have seemed preposterous. The AI winter was in full effect after the collapse of expert systems and connectionism wasn't mainstream. FP16 precision wasn't even on the radar.
Telling someone you needed $500 billion to build specialized hardware for AI would have gotten you laughed out of any room. The entire PC industry wasn't anywhere near that scale yet.
What's fascinating is how the gaming industry inadvertently built the foundation for modern AI. Graphics cards designed to render Doom and Quake evolved into the computational backbone for neural networks – a connection nobody saw coming.
It's a perfect example of how technological evolution takes unexpected paths that seem obvious only in retrospect.
Other models, same or similar prompt
Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.