17 comments

[ 3.1 ms ] story [ 50.1 ms ] thread
Would you say that Nvidia's role in AI is based off software or hardware?
Just like a car needs both an engine and tires, software and hardware are equally critical to success. Something that AMD just does not get.
Software, there are 100s of ai hardware startups. It’s easier relatively to build a domain specific chip, but building a compiler stack that integrates with a major framework and gets good perf from your chip is incredibly difficult. Getting people to switch from nvidia? Even harder
Is there any reality to crypto-mining gpu operations converting to AI model services (assuming they have the correct hardware)? Where can more GPUs be found and aggregated?
> Is there any reality to crypto-mining gpu operations converting to AI model services (assuming they have the correct hardware)?

Yes. Coreweave was an Ethereum mining operation that reconverted into this.

Coreweave intially loaded up on GPUs to mine Ethereum, made money off that during the boom, then pivoted to being a GPU cloud, and found themselves with all these GPUs in the ML/AI bubble. Amazing run.
Mining used consumer GPUs which are not allowed to be used for datacenters. So I doubt they managed to reuse the hardware
No mining GPUs are in their fleet anymore, that's long gone. It's all datacenter GPUs as per their website.
> demonstrating that AI can take human input and produce sophisticated responses

sophisticated looking broken plagiarism

FTFY