Microsoft CTO says he wants to swap most AMD and Nvidia GPUs for homemade chips Microsoft buys a lot of GPUs from both Nvidia and AMD. But moving forward, Redmond's leaders want to shift the majority of its AI workloads from GPUs to its own homegrown accelerators. The software titan is rather late to the custom silicon party. While Amazon and Google have been building custom CPUs and AI accelerators for …
Every vertically structured giant company developing their own chips in house, not even bothering to directly compete with NVidia because their own advertising and SaS serfdoms are so huge they have no need to. Its like the days of DEC before Silicon Valley surpassed that closed model with decentralized, flat hierarchy, collaborative style of business, which proved vastly more competitive and productive.
Only this time, the usurper is more likely to be in <redacted> than in the US. It not too late, but the clock is ticking.
Read or heard something recently comparing Google's AI stuff to Meta's AI stuff, saying something like Meta had deployed something like 1000x more Nvidia GPUs vs Google but their AI stuff wasn't as good as Google's, failing to mention that Google has their own AI hardware too.
Doing a quick web search seems to indicate that Meta does have their own AI chips, and may start testing the first systems in production later this year, apparently ordering up to 6,000 racks of them.
Nvidia's going to need all the (other) customers they can get, no wonder they are itching so hard to sell to China, once these hyperscalers get going, for sure they will dramatically scale down purchases of Nvidia hardware.
>Redmond also has its own CPU called Cobalt
That immediately made me think of; https://en.wikipedia.org/wiki/Cobalt_bomb?useskin=monobook#Radiation_levels_and_time
(While those wouldn't destroy the Earth, those would make the bombed area uninhabitable for at least 100 years without decontamination).
We want flexibility….
Except it takes a few years between design to manufacture, and then you need a fab slot, and assembly. Which means you need at LEAST 3 years between ‘I want to do this’ and ‘this’ being a real physical option.
At LEAST 3 years without changing your mind or ballsing up the Project Management…
But it’s an easy thing for the CEO of a software licensing company to say.
I think it’s pretty cool you can just ask the A.I. something and it spontaneously writes you a wikipedia page about it. Then you press it for details and it spews out lies and exaggerations. So “why” is it lying and exaggerating, it can’t be a part of its programming, could it? The training material itself must be full of lies and exaggeration.
Heyyyy what if A.I. is a gigantic swindle perpetrated by the computer itself? Where does the bombastic ignorance originate?
