Interesting in their choice of TFLOPS announcement. They could’ve simply claimed 33 and put an asterisk for FP16 performance on the precision and called it a day. They’re listing AMD’s FP32 spec, which is divergent from Ampere/Ada which has the same output regardless of precision.
Interesting in their choice of TFLOPS announcement. They could’ve simply claimed 33 and put an asterisk for FP16 performance on the precision and called it a day. They’re listing AMD’s FP32 spec, which is divergent from Ampere/Ada which has the same output regardless of precision.