AMD Confirms Different FSR 4.1 Model For RDNA 3 GPUs; Delayed Support On RDNA 2 Is Due To Significant Optimization Needs
… The reason for using a different model of FSR 4.1 on RDNA 3 is the absence of FP8 8-bit floating-point support. RDNA 3 only supports INT8 8-bit Integer data type , and it's necessary to requantize the model so it runs efficiently with the INT8 data type instead of FP8. …
