News for hackers. Stuff that matters.
Re:Nvidia deliberately kept GPU prices high through circular... by liendolucas on Saturday November 14, 2026 @10:36PM (#50503750)
Can someone explain why is so hard to break away from CUDA and why companies like Intel or AMD can't release a competing product? What it really can't fit in my mind is the reliance of just Nvidia for this boom... With all the expertise and knowledge that both Intel and AMD have I can't see why is so difficult for them to have at least a second option to Nvidia.
Re:Nvidia deliberately kept GPU prices high through circular... by kitchi on Saturday November 14, 2026 @10:49PM (#50503809)
Nvidia has been building CUDA and the software stack on top of cuda since the mid 2000s iirc. They've open sourced almost all of it, and have been actively hiring experts in different domains to build out different cuda products.

AMD and Intel may have similar computing expertise, but they almost certainly don't have experts in atmospheric modeling, fluid dynamics, cosmology etc etc working for them. Or not at the same level of maturity anyway.

I'm an academic in a compute heavy STEM field and attended my first GTX last year and I was blown away by the breadth of domain experts that were already well embedded in the ecosystem.

I think that was a brilliant move on nvidias part, to make the software stack free and open source, and therefore heavily inspectable and auditable. But to tie it to their hardware so they drive sales from multiple domains.

Re:Nvidia deliberately kept GPU prices high through circular... by hdhdhsjsbdh on Saturday November 14, 2026 @10:49PM (#50503810)
Probably a mix of tech debt and organizational inertia. CUDA didn’t happen overnight, it was a long-term bet; it’s mature enough that its place in the org and the stack are well established. Hardware companies are notoriously bad at software engineering management and decision making – it’s an open secret that Intel in particular is a cluster fuck – so pulling off something well-designed and strongly supported like CUDA is much easier said than done