Re:The computers used to do 3D animation for Final Fantasy V...
by zozbot234 on Thursday April 07, 2022 @04:41PM
(#30950084)
Neural networks have non-linearities. There were hacks for doing some sort of general purpose computation using the GPU fixed rendering pipeline, but they're not enough for neural networks especially as understood today. You need general shaders, which were a relatively late development.
Re:The computers used to do 3D animation for Final Fantasy V...
by agumonkey on Thursday April 07, 2022 @03:17PM
(#30949125)
Aight but then why no other company managed to compete on the numerical core array ? still funny, as if the wealthy gaming market funded nvidia venture into serious computing enough to choke the potential competition.
Re:The computers used to do 3D animation for Final Fantasy V...
by jandrese on Thursday April 07, 2022 @03:28PM
(#30949257)
You need huge money to develop the tech, but the HPC industry is tiny. People are always impressed at these million dollar machines, but only a handful are built every year. There is a lot more money in gaming selling millions of $200-$1000 cards every year.
Re:The computers used to do 3D animation for Final Fantasy V...
by ac29 on Thursday April 07, 2022 @07:19PM
(#30951501)
There certainly are some smaller companies attempting to compete on making AI/ML chips. Tenstorrent comes to mind: https://tenstorrent.com/grayskull/
Re:The computers used to do 3D animation for Final Fantasy V...
by Sohcahtoa82 on Thursday April 07, 2022 @05:25PM
(#30950512)
Silicon is expensive to design, and even more expensive to build. The startup costs are crazy high. By the time people realized it could be a thing, nVidia and AMD already owned the market.
Rendering 3D graphics for games and the supercomputing used by AI/ML/research both need the same thing: Embarrassingly parallel math calculations with little or no branching in the code.
For example, a feed-forward neural network is just a whole lot of multiplication and addition. Transforming a 3D vertex in space to a 2D screen coordinate is matrix multiplication, which is just a whole lot of multiplication and addition. If you've already designed silicon to perform those operations in a single clock cycle, making it do HPC/SC rather than gaming isn't that big of a switch.