I think 32-bit floats. I don't know of development environments like CUDA to use the VideoCore VII for accelerated computing. Nvidia reportedly puts 80 percent of their time into software. 80 GFLOPS from the Raspberry Pi is not much motivation for anyone to spend similar effort.You say this are single precision FLOPs. Is that 16 bits floats or 32 bit float? For CPUs single precision would mean 32 bits and double precision 64 bits. But I'm not sure about GPU.
In the past I have suggested a GPU programming environment for the Pi could be motivated by the importance of teaching people how to write code for heterogeneous accelerated computing.
My impression is that writing the documentation that would allow any third party to develop an accelerated computing platform is more difficult than actually creating the environment. In other words, one reason Linus could make Linux is because Intel had already provided the necessary documentation for Microsoft to make Windows.
Statistics: Posted by ejolson — Sat Dec 14, 2024 4:15 pm
