4 ms·
unfortunately mostly useless 4 bit floating point
by blibble 1y ago
unfortunately mostly useless 4 bit floating point
- nicce 1y agoHow is it useless?
- cyber_kinetist 1y agoIf you try to do any sort of scientific calculation (ex. physical simulation), at least 32-bit floats are needed to ensure adequate numerical precision. Ideally 64-bit floats are the best, but FP64 performance in recent GPUs are bad up to the point that you can say they're not supported anymore...