diff --git a/library/include/ck_tile/library/tensor_operation_instance/gpu/tile_grouped_conv_bwd_weight_bf16_instances_opt.hpp b/library/include/ck_tile/library/tensor_operation_instance/gpu/tile_grouped_conv_bwd_weight_bf16_instances_opt.hpp
new file mode 100644
index 0000000000..1e82d59d51
--- /dev/null
+++ b/library/include/ck_tile/library/tensor_operation_instance/gpu/tile_grouped_conv_bwd_weight_bf16_instances_opt.hpp
@@ -0,0 +1,49 @@
+// Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+
+#include "ck_tile/library/tensor_operation_instance/gpu/tile_grouped_conv_instance_factory.hpp"
+#include "ck_tile/library/tensor_operation_instance/gpu/tile_grouped_conv_bwd_weight_invoker.hpp"
+
+namespace ck_tile {
+namespace ops {
+
+using BF16 = ck_tile::bfloat16_t;
+
+using DeviceOp2DBF16 = GroupedConvolutionBackwardWeightBaseInvoker<2,
+                                                 NHWGC,
+                                                 GKYXC,
+                                                 NHWGK,
+                                                 BF16,
+                                                 BF16,
+                                                 BF16,
+                                                 PassThrough,
+                                                 PassThrough,
+                                                 PassThrough,
+                                                 BF16,
+                                                 BF16>;
+
+template <ck_tile::index_t NDimSpatial,
+          typename ALayout,
+          typename BLayout,
+          typename ELayout>
+using tile_grouped_conv_bwd_weight_bf16_instances_opt = std::tuple<
+// clang-format off
+    //#####################################|       Num|  InLayout| WeiLayout| OutLayout| InData| WeiData| OutData|           In|         Wei|         Out|  K-block|  M-tile| N-tile | K-tile | M-warp| N-warp| K-warp| M-warp| N-warp| K-warp| Vector| Vector| Vector|  Double|                         GEMM|
+    //#####################################|       Dim|          |          |          |   Type|    Type|    Type|  Elementwise| Elementwise| Elementwise|      per|        |        |        |       |       |       |   tile|   tile|   tile|   size|   size|   size|    smem|                     pipeline|
+    //#####################################|   Spatial|          |          |          |       |        |        |    Operation|   Operation|   Operation|       CU|        |        |        |       |       |       |   size|   size|   size|      A|      B|      C|  buffer|                      version|
+    GroupedConvolutionBackwardWeightInvoker<NDimSpatial,   ALayout,   BLayout,   ELayout,   BF16,    BF16,    BF16,  PassThrough, PassThrough, PassThrough,        1,     128,     128,      64,      2,      2,      1,    32,     32,     16,      8,      8,       8,   false,  CK_TILE_PIPELINE_COMPUTE_V3>,
+    GroupedConvolutionBackwardWeightInvoker<NDimSpatial,   ALayout,   BLayout,   ELayout,   BF16,    BF16,    BF16,  PassThrough, PassThrough, PassThrough,        1,     128,     128,      64,      2,      2,      1,    32,     32,     16,      8,      8,       8,    true,  CK_TILE_PIPELINE_COMPUTE_V4>,
+    GroupedConvolutionBackwardWeightInvoker<NDimSpatial,   ALayout,   BLayout,   ELayout,   BF16,    BF16,    BF16,  PassThrough, PassThrough, PassThrough,        2,     128,     128,      64,      2,      2,      1,    32,     32,     16,      8,      8,       8,   false,  CK_TILE_PIPELINE_COMPUTE_V3>,
+    GroupedConvolutionBackwardWeightInvoker<NDimSpatial,   ALayout,   BLayout,   ELayout,   BF16,    BF16,    BF16,  PassThrough, PassThrough, PassThrough,        2,     128,     128,      64,      2,      2,      1,    32,     32,     16,      8,      8,       8,    true,  CK_TILE_PIPELINE_COMPUTE_V4>,
+
+    // Skinny kernels.
+    GroupedConvolutionBackwardWeightInvoker<NDimSpatial,   ALayout,   BLayout,   ELayout,   BF16,    BF16,    BF16,  PassThrough, PassThrough, PassThrough,        2,     128,     128,       16,      2,      2,      1,    32,     32,     16,      8,      8,       4,   false,  CK_TILE_PIPELINE_COMPUTE_V3>,
+    GroupedConvolutionBackwardWeightInvoker<NDimSpatial,   ALayout,   BLayout,   ELayout,   BF16,    BF16,    BF16,  PassThrough, PassThrough, PassThrough,        2,     128,     128,       16,      2,      2,      1,    32,     32,     16,      8,      8,       4,    true,  CK_TILE_PIPELINE_COMPUTE_V4>,
+    GroupedConvolutionBackwardWeightInvoker<NDimSpatial,   ALayout,   BLayout,   ELayout,   BF16,    BF16,    BF16,  PassThrough, PassThrough, PassThrough,        1,     128,     128,       16,      2,      2,      1,    32,     32,     16,      8,      8,       4,   false,  CK_TILE_PIPELINE_COMPUTE_V3>,
+    GroupedConvolutionBackwardWeightInvoker<NDimSpatial,   ALayout,   BLayout,   ELayout,   BF16,    BF16,    BF16,  PassThrough, PassThrough, PassThrough,        1,     128,     128,       16,      2,      2,      1,    32,     32,     16,      8,      8,       4,    true,  CK_TILE_PIPELINE_COMPUTE_V4>
+    // clang-format on
+>;
+
+} // namespace ops
+} // namespace ck_tile
diff --git a/library/include/ck_tile/library/tensor_operation_instance/gpu/tile_grouped_conv_bwd_weight_factory.hpp b/library/include/ck_tile/library/tensor_operation_instance/gpu/tile_grouped_conv_bwd_weight_factory.hpp
index fa680930ba..0e776edd81 100644
--- a/library/include/ck_tile/library/tensor_operation_instance/gpu/tile_grouped_conv_bwd_weight_factory.hpp
+++ b/library/include/ck_tile/library/tensor_operation_instance/gpu/tile_grouped_conv_bwd_weight_factory.hpp
@@ -56,6 +56,7 @@ using DeviceOp2DF32 = GroupedConvolutionBackwardWeightBaseInvoker<2,
 // Forward declarations for instance factory functions
 void add_grouped_conv2d_bwd_weight_f16_instances(std::vector<std::unique_ptr<DeviceOp2DF16>>& instances);
 void add_grouped_conv2d_bwd_weight_bf16_instances(std::vector<std::unique_ptr<DeviceOp2DBF16>>& instances);
+void add_grouped_conv2d_bwd_weight_bf16_instances_opt(std::vector<std::unique_ptr<DeviceOp2DBF16>>& instances);
 
 template <ck_tile::index_t NumDimSpatial,
           typename InLayout,
@@ -117,6 +118,7 @@ struct DeviceOperationInstanceFactory<GroupedConvolutionBackwardWeightBaseInvoke
                              std::is_same_v<ComputeTypeB, ck_tile::bfloat16_t>)
                 {
                     add_grouped_conv2d_bwd_weight_bf16_instances(op_ptrs);
+                    add_grouped_conv2d_bwd_weight_bf16_instances_opt(op_ptrs);
                 }
             }
         }
diff --git a/library/src/ck_tile/tensor_operation_instance/gpu/CMakeLists.txt b/library/src/ck_tile/tensor_operation_instance/gpu/CMakeLists.txt
index 6331bec202..ac97cd0b97 100644
--- a/library/src/ck_tile/tensor_operation_instance/gpu/CMakeLists.txt
+++ b/library/src/ck_tile/tensor_operation_instance/gpu/CMakeLists.txt
@@ -11,6 +11,10 @@ add_library(ck_tile_grouped_conv_bwd_weight_bf16_instances STATIC
     tile_grouped_conv_bwd_weight_bf16_instances.cpp
 )
 
+add_library(ck_tile_grouped_conv_bwd_weight_bf16_instances_opt STATIC
+    tile_grouped_conv_bwd_weight_bf16_instances_opt.cpp
+)
+
 add_library(ck_tile_grouped_conv_fwd_fp16_instances STATIC
     tile_grouped_conv_fwd_fp16_instances.cpp
 )
@@ -29,6 +33,11 @@ target_include_directories(ck_tile_grouped_conv_bwd_weight_bf16_instances PRIVAT
     ${PROJECT_SOURCE_DIR}/library/include
 )
 
+target_include_directories(ck_tile_grouped_conv_bwd_weight_bf16_instances_opt PRIVATE
+    ${PROJECT_SOURCE_DIR}/include
+    ${PROJECT_SOURCE_DIR}/library/include
+)
+
 target_include_directories(ck_tile_grouped_conv_fwd_fp16_instances PRIVATE
     ${PROJECT_SOURCE_DIR}/include
     ${PROJECT_SOURCE_DIR}/library/include
diff --git a/library/src/ck_tile/tensor_operation_instance/gpu/tile_grouped_conv_bwd_weight_bf16_instances_opt.cpp b/library/src/ck_tile/tensor_operation_instance/gpu/tile_grouped_conv_bwd_weight_bf16_instances_opt.cpp
new file mode 100644
index 0000000000..8d315771ac
--- /dev/null
+++ b/library/src/ck_tile/tensor_operation_instance/gpu/tile_grouped_conv_bwd_weight_bf16_instances_opt.cpp
@@ -0,0 +1,20 @@
+// Copyright (c) 2025, Advanced Micro Devices, Inc. All rights reserved.
+
+#include "ck_tile/library/tensor_operation_instance/gpu/tile_grouped_conv_instance_factory.hpp"
+#include "ck_tile/library/tensor_operation_instance/gpu/tile_grouped_conv_bwd_weight_bf16_instances_opt.hpp"
+
+namespace ck_tile {
+namespace ops {
+
+void add_grouped_conv2d_bwd_weight_bf16_instances_opt(std::vector<std::unique_ptr<DeviceOp2DBF16>>& instances)
+{
+    add_device_operation_instances(instances,
+                                   tile_grouped_conv_bwd_weight_bf16_instances_opt<
+                                       2,
+                                       NHWGC,
+                                       GKYXC,
+                                       NHWGK>{});
+}
+
+} // namespace ops
+} // namespace ck_tile
diff --git a/profiler/ck_tile/src/CMakeLists.txt b/profiler/ck_tile/src/CMakeLists.txt
index baa840cf59..26db40b9fa 100644
--- a/profiler/ck_tile/src/CMakeLists.txt
+++ b/profiler/ck_tile/src/CMakeLists.txt
@@ -41,6 +41,7 @@ set(DEVICE_INSTANCES "")
 if(SUPPORTED_GPU_TARGETS MATCHES "gfx9" OR SUPPORTED_GPU_TARGETS MATCHES "gfx1[12]")
   list(APPEND DEVICE_INSTANCES ck_tile_grouped_conv_bwd_weight_fp16_instances)
   list(APPEND DEVICE_INSTANCES ck_tile_grouped_conv_bwd_weight_bf16_instances)
+  list(APPEND DEVICE_INSTANCES ck_tile_grouped_conv_bwd_weight_bf16_instances_opt)
   list(APPEND DEVICE_INSTANCES ck_tile_grouped_conv_fwd_fp16_instances)
   list(APPEND DEVICE_INSTANCES ck_tile_grouped_conv_fwd_bf16_instances)
 endif()