diff --git a/python/cudnn/grouped_gemm/grouped_gemm_quant/api.py b/python/cudnn/grouped_gemm/grouped_gemm_quant/api.py
index a89bcf004..eaf7d7d98 100644
--- a/python/cudnn/grouped_gemm/grouped_gemm_quant/api.py
+++ b/python/cudnn/grouped_gemm/grouped_gemm_quant/api.py
@@ -1172,6 +1172,7 @@ def grouped_gemm_quant_wrapper_sm100(
     row_scale_tensor: Optional[torch.Tensor] = None,
     acc_dtype: torch.dtype = torch.float32,
     d_dtype: torch.dtype = torch.bfloat16,
+    d_tensor: Optional[torch.Tensor] = None,
     cd_major: str = "n",
     mma_tiler_mn: Tuple[int, int] = (256, 256),
     cluster_shape_mn: Optional[Tuple[int, int]] = None,
@@ -1212,6 +1213,9 @@ def grouped_gemm_quant_wrapper_sm100(
             conversion.
         acc_dtype: Accumulator data type
         d_dtype: Output D tensor data type
+        d_tensor: Optional preallocated output tensor to write into instead of
+            allocating. Must match the internal layout: shape (valid_m, n_out, 1),
+            stride (n_out, 1, valid_m * n_out), dtype d_dtype, on a_tensor.device.
         cd_major: CD major dimension (only "n"-major layout is supported)
         mma_tiler_mn: MMA tiler shape
         cluster_shape_mn: Cluster shape
@@ -1288,7 +1292,21 @@ def grouped_gemm_quant_wrapper_sm100(
     _logger.debug("grouped_gemm_quant_wrapper_sm100: Creating output tensors")
 
     if cd_major == "n":
-        d_tensor = torch.empty_strided((valid_m, n_out, 1), (n_out, 1, valid_m * n_out), dtype=d_dtype, device=a_tensor.device)
+        expected_shape = (valid_m, n_out, 1)
+        expected_stride = (n_out, 1, valid_m * n_out)
+        if d_tensor is None:
+            d_tensor = torch.empty_strided(expected_shape, expected_stride, dtype=d_dtype, device=a_tensor.device)
+        elif (
+            tuple(d_tensor.shape) != expected_shape
+            or tuple(d_tensor.stride()) != expected_stride
+            or d_tensor.dtype != d_dtype
+            or d_tensor.device != a_tensor.device
+        ):
+            raise ValueError(
+                f"d_tensor must have shape {expected_shape}, stride {expected_stride}, "
+                f"dtype {d_dtype}, device {a_tensor.device}, but got shape {tuple(d_tensor.shape)}, "
+                f"stride {tuple(d_tensor.stride())}, dtype {d_tensor.dtype}, device {d_tensor.device}."
+            )
         d_col_tensor = (
             torch.empty_strided((valid_m, n_out, 1), (n_out, 1, valid_m * n_out), dtype=d_dtype, device=a_tensor.device)
             if is_low_precision_output_config
diff --git a/test/python/fe_api/test_grouped_gemm_quant.py b/test/python/fe_api/test_grouped_gemm_quant.py
index 03fcae24c..88b9092ab 100644
--- a/test/python/fe_api/test_grouped_gemm_quant.py
+++ b/test/python/fe_api/test_grouped_gemm_quant.py
@@ -246,6 +246,7 @@ def test_grouped_gemm_quant_wrapper_fp8(
         discrete_col_sfd=discrete_col_sfd,
         use_dynamic_sched=use_dynamic_sched,
         request=request,
+        provide_d_tensor=True,
     )
 
 
@@ -1110,6 +1111,7 @@ def _test_grouped_gemm_quant_wrapper(
     input_mutator=None,
     use_dynamic_sched=False,
     enable_bias=False,
+    provide_d_tensor=False,
 ):
     """Test GroupedGemmQuant API via the wrapper function (with caching)."""
     try:
@@ -1151,6 +1153,16 @@ def _test_grouped_gemm_quant_wrapper(
     if input_mutator is not None:
         input_mutator(inputs, cfg)
 
+    caller_d_tensor = None
+    if provide_d_tensor:
+        valid_m, n_out = inputs["a_tensor"].shape[0], cfg["n"]
+        caller_d_tensor = torch.empty_strided(
+            (valid_m, n_out, 1),
+            (n_out, 1, valid_m * n_out),
+            dtype=cfg["d_dtype"],
+            device=inputs["a_tensor"].device,
+        )
+
     try:
         for _ in range(2):  # Run twice to test caching path
             outputs = grouped_gemm_quant_wrapper_sm100(
@@ -1166,6 +1178,7 @@ def _test_grouped_gemm_quant_wrapper(
                 row_scale_tensor=inputs.get("row_scale_tensor"),
                 acc_dtype=cfg["acc_dtype"],
                 d_dtype=cfg["d_dtype"],
+                d_tensor=caller_d_tensor,
                 cd_major=cfg["cd_major"],
                 mma_tiler_mn=cfg["mma_tiler_mn"],
                 cluster_shape_mn=cfg["cluster_shape_mn"],
@@ -1179,6 +1192,9 @@ def _test_grouped_gemm_quant_wrapper(
     except (ValueError, NotImplementedError) as e:
         pytest.skip(f"Unsupported testcase: {e}")
 
+    if provide_d_tensor:
+        assert outputs["d_tensor"].data_ptr() == caller_d_tensor.data_ptr()
+
     check_ref_grouped_gemm_quant(
         inputs,
         outputs,