Fp16 or fp32 Accumulate?

Hi everyone, Is there a way to know what type of accumulate is being used for a a cublasGemmEx() call when its running on tensor cores. I want to know the exact tflops and compare it with the peak. Because the peaks for both fp16 and fp32 accumulate are different this creates a problem. Does the compute type argument in the cublasgemmex() call specify the accumulate type?