Cuda is an Nvidia specific method for using a graphics card to do computation (not just graphics), like physics simulations.
Translation layers would let you use software designed for other graphics cards to work with Cuda, or to let Cuda software work on other graphics cards
CUDA was there first and has established itself as the standard for GPGPU (“general purpose GPU” aka calculating non-graphics stuff on a graphics card). There are many software packages out there that only support CUDA, especially in the lucrative high-performance computing market.
Most software vendors have no intention of supporting more than one API since CUDA works and the market isn’t competitive enough for someone to need to distinguish themselves though better API support.
Thus Nvidia have a lock on a market that regularly needs to buy expensive high-margin hardware and they don’t want to share. So they made up a rule that nobody else is allowed to write out use something that makes CUDA software work with non-Nvidia GPUs.
That’s anticompetitive but it remains to be seen if it’s anticompetitive enough for the EU to step in.
Yeah fuck this.
… What’s a translation layer?
I think it’s about translating cuda to ROCm instructions or something.
Got a Windows app you want to run on Linux? Wine and Proton are well known translation layers.
I guess Graphics Cards are similar. CUDA is basically the NVIDIA equivalent of .exe I think.
Cuda is an Nvidia specific method for using a graphics card to do computation (not just graphics), like physics simulations.
Translation layers would let you use software designed for other graphics cards to work with Cuda, or to let Cuda software work on other graphics cards
So nvidia designed something and they don’t want other companies to use it?
CUDA was there first and has established itself as the standard for GPGPU (“general purpose GPU” aka calculating non-graphics stuff on a graphics card). There are many software packages out there that only support CUDA, especially in the lucrative high-performance computing market.
Most software vendors have no intention of supporting more than one API since CUDA works and the market isn’t competitive enough for someone to need to distinguish themselves though better API support.
Thus Nvidia have a lock on a market that regularly needs to buy expensive high-margin hardware and they don’t want to share. So they made up a rule that nobody else is allowed to write out use something that makes CUDA software work with non-Nvidia GPUs.
That’s anticompetitive but it remains to be seen if it’s anticompetitive enough for the EU to step in.