diff --git a/ggml/src/ggml-cuda/ggml-cuda.cu b/ggml/src/ggml-cuda/ggml-cuda.cu
index ef56e944d01ab..dd94ab03d5b6c 100644
--- a/ggml/src/ggml-cuda/ggml-cuda.cu
+++ b/ggml/src/ggml-cuda/ggml-cuda.cu
@@ -1760,11 +1760,13 @@ static void ggml_cuda_mul_mat(ggml_backend_cuda_context & ctx, const ggml_tensor
     //printf("src0 is contiguous %d, transposed %d, type = %s, name = %s\n", ggml_is_contiguous(src0), ggml_is_transposed(src0), ggml_type_name(src0->type), src0->name);
     //printf("src1 is contiguous %d, transposed %d, type = %s, name = %s\n", ggml_is_contiguous(src1), ggml_is_transposed(src1), ggml_type_name(src1->type), src1->name);
 
-    if (!split && src0->type == GGML_TYPE_F16 && src1->ne[1] == 1 && dst->ne[3] == 1 && (src0->ne[1] < MMV_MAX_ROWS || any_gpus_without_fp16_mma)) {
+    if (!split && use_mul_mat_vec && dst->ne[3] == 1 && (src0->ne[1] < MMV_MAX_ROWS || any_gpus_without_fp16_mma)) {
+        // the custom F16 vector kernel can be used over batched cuBLAS GEMM
+        // but this is only faster for GPUs without tensor cores or with a thin src0 matrix (particularly KQV in attention)
         ggml_cuda_mul_mat_vec(ctx, src0, src1, dst);
     } else if (!split && src0->type == GGML_TYPE_F16 && (src1->type == GGML_TYPE_F16 || !any_gpus_with_slow_fp16)
                && !ggml_is_transposed(src0) && !ggml_is_transposed(src1) && src1->ne[2]*src1->ne[3] > 1) {
-        // KQ + KQV multi-batch without FlashAttention
+        // general KQ + KQV multi-batch without FlashAttention
         ggml_cuda_mul_mat_batched_cublas(ctx, src0, src1, dst);
     } else if (use_mul_mat_vec) {
         ggml_cuda_op_mul_mat(ctx, src0, src1, dst, ggml_cuda_op_mul_mat_vec, nullptr);