Nvidia is still the only game in town for training at scale, and the fastest for inference (plus, at scale, the cheapest for inference, as hard as that sounds to believe)

▲

impulser_ 6 hours ago | parent | next [-]

I think both Google and AMD both prove this is not true anymore.

▲

cowlevel 6 hours ago | parent [-]

To use Google chips you have to be Google, and AMD continues snatching defeat from the jaws of victory with almost every executive decision.

	▲	impulser_ 6 hours ago \| parent \| next [-]
		No. This is also not true anymore. In fact they are even letting other cloud providers buy them like Oracle. https://www.oracle.com/news/announcement/ai-world-oracle-bri...
	▲	bryanlarsen 6 hours ago \| parent \| prev [-]
		Anthropic and OpenAI both use large numbers of Google Tensor chips.

▲

LeFantome 6 hours ago | parent | prev [-]

Unless your town is in China