Quote:
Originally Posted by Duik
[You must be logged in to view images. Log in or Register.]
The more chips joined with slower bus speeds the more latwncy. An on chip faster solution would allow theoretically faster computaional speeds. This assumes equal quality algorithms. Clearly the chinese algorithms are quicker including the hobbling of their access to chip tech.
Also lmao. You think the chinese don't have the latest intel shit? That's so cute. Anywho.
Lesser chips doing better work = better devs that use resources with superior utility. Sorry guys it's just common sense.
|
the west had zero incentive to improve the cost to compute ratio, they quite fucking literally had a monetary incentive to do the exact opposite from a investment build out prospective, if you look at the actual formula to what they did different in the training phase, it wasn't some whole new fucking approach to this shit, they spent time optimizing something nobody in the west bothered to do because why would they it would be like releasing a paper explaining how to enrich uranium more efficiently then handing over to the Iranians
i'll tap the sign once again pointing to the professor guy who actually knows this shit, he's impressed but reverse engineers it and demystifies and curb stomps the notion this is actually some giant breakthrough
>You think the chinese don't have the latest intel shit?
we've already established Intel and AMD chips are literally useless for AI currently. CUDA is a proprietary nivida thing and the reason their stock is up 30,000% this year or w/e and neither AMD or INTEL is suicidal enough to license CUDA from Nvidia like at that point why are you even still a company lol
AMD made their software free of charge and still nobody gives a single fuck enough to learn it and write all the fucking framework for it from scratch, and Intel is too busy trying to find a CEO that won't tweet out prayers on twitter
[You must be logged in to view images. Log in or Register.]