diff --git a/csrc/ops.h b/csrc/ops.h
index 171b70eb80aee..363ddec3d0729 100644
--- a/csrc/ops.h
+++ b/csrc/ops.h
@@ -155,8 +155,6 @@ void cutlass_scaled_mm_azp(torch::Tensor& out, torch::Tensor const& a,
                            c10::optional<torch::Tensor> const& azp,
                            c10::optional<torch::Tensor> const& bias);
 
-bool cutlass_scaled_sparse_mm_supports_fp8(int64_t cuda_device_capability);
-
 void cutlass_scaled_sparse_mm(torch::Tensor& out, torch::Tensor const& a,
                               torch::Tensor const& e, torch::Tensor const& b,
                               torch::Tensor const& a_scales,
diff --git a/csrc/sparse/cutlass/sparse_scaled_mm_entry.cu b/csrc/sparse/cutlass/sparse_scaled_mm_entry.cu
index 9e23df5a05d69..5075c342098ba 100644
--- a/csrc/sparse/cutlass/sparse_scaled_mm_entry.cu
+++ b/csrc/sparse/cutlass/sparse_scaled_mm_entry.cu
@@ -12,22 +12,6 @@ void cutlass_scaled_sparse_mm_sm90(torch::Tensor& c, torch::Tensor const& a,
                                    c10::optional<torch::Tensor> const& bias);
 #endif
 
-bool cutlass_scaled_sparse_mm_supports_fp8(int64_t cuda_device_capability) {
-  // CUTLASS FP8 kernels need at least
-  //   CUDA 12.0 on SM90 systems (Hopper)
-  //   CUDA 12.4 on SM89 systems (Lovelace)
-
-#if defined CUDA_VERSION
-  if (cuda_device_capability >= 90) {
-    return CUDA_VERSION >= 12000;
-  } else if (cuda_device_capability >= 89) {
-    return CUDA_VERSION >= 12040;
-  }
-#endif
-
-  return false;
-}
-
 int32_t test_get_sm_version_num() {
   int32_t major_capability, minor_capability;
   cudaDeviceGetAttribute(&major_capability, cudaDevAttrComputeCapabilityMajor,
diff --git a/csrc/torch_bindings.cpp b/csrc/torch_bindings.cpp
index 667aa94db3218..3cbc843dc501a 100644
--- a/csrc/torch_bindings.cpp
+++ b/csrc/torch_bindings.cpp
@@ -313,7 +313,8 @@ TORCH_LIBRARY_EXPAND(TORCH_EXTENSION_NAME, ops) {
   ops.def("cutlass_scaled_mm_supports_fp8(int cuda_device_capability) -> bool");
   ops.impl("cutlass_scaled_mm_supports_fp8", &cutlass_scaled_mm_supports_fp8);
 
-  // Test
+  // CUTLASS sparse GEMM, supporting symmetric per-tensor or per-row/column
+  // quantization, as well as bias
   ops.def(
       "cutlass_scaled_sparse_mm(Tensor! out, Tensor a,"
       "                  Tensor e,"
@@ -321,14 +322,7 @@ TORCH_LIBRARY_EXPAND(TORCH_EXTENSION_NAME, ops) {
       "                  Tensor b_scales, Tensor? bias) -> ()");
   ops.impl("cutlass_scaled_sparse_mm", torch::kCUDA, &cutlass_scaled_sparse_mm);
 
-  // Test
-  ops.def(
-      "cutlass_scaled_sparse_mm_supports_fp8(int cuda_device_capability) -> "
-      "bool");
-  ops.impl("cutlass_scaled_sparse_mm_supports_fp8",
-           &cutlass_scaled_sparse_mm_supports_fp8);
-
-  // Test
+  // CUTLASS sparse matrix compressor
   ops.def(
       "cutlass_compress_entry(Tensor! a_compressed, Tensor! e,"
       " Tensor a) -> bool");
diff --git a/vllm/_custom_ops.py b/vllm/_custom_ops.py
index 6087247de5a94..d99d2340275af 100644
--- a/vllm/_custom_ops.py
+++ b/vllm/_custom_ops.py
@@ -532,11 +532,6 @@ def cutlass_scaled_mm_azp(a: torch.Tensor,
     return out
 
 
-def cutlass_scaled_sparse_mm_supports_fp8(cuda_device_capability: int) -> bool:
-    return torch.ops._C.cutlass_scaled_sparse_mm_supports_fp8(
-        cuda_device_capability)
-
-
 def cutlass_compress_entry(a: torch.Tensor) \
     -> Tuple[torch.Tensor, torch.Tensor]:
     assert (a.dtype is torch.int8 or a.dtype is torch.float8_e4m3fn or \