diff --git a/onnxruntime/contrib_ops/cuda/math/gemm_float8.cu b/onnxruntime/contrib_ops/cuda/math/gemm_float8.cu
index 28ab27ee33..07c5de2fe8 100644
--- a/onnxruntime/contrib_ops/cuda/math/gemm_float8.cu
+++ b/onnxruntime/contrib_ops/cuda/math/gemm_float8.cu
@@ -207,7 +207,7 @@ Status GemmFloat8::ComputeGemm(
 #endif
         case CUDA_R_8F_E4M3:
         case CUDA_R_8F_E5M2:
-          compute_type = CUBLAS_COMPUTE_32F_FAST_TF32;
+          compute_type = CUBLAS_COMPUTE_32F;
           break;
 #endif
         default:
@@ -219,7 +219,7 @@ Status GemmFloat8::ComputeGemm(
       compute_type = CUBLAS_COMPUTE_32F_FAST_16BF;
       break;
     case CUDA_R_32F:
-      compute_type = CUBLAS_COMPUTE_32F_FAST_TF32;
+      compute_type = CUBLAS_COMPUTE_32F;
       break;
     default:
       ORT_THROW("Unable to determine computeType in operator GemmFloat8.");