diff --git a/flashinfer/jit/attention/modules.py b/flashinfer/jit/attention/modules.py
index bb6962b791..ed3d2b8d0f 100755
--- a/flashinfer/jit/attention/modules.py
+++ b/flashinfer/jit/attention/modules.py
@@ -1751,7 +1751,7 @@ def gen_fmha_cutlass_sm100a_module(
     ]
 
     nvcc_flags = current_compilation_context.get_nvcc_flags_list(
-        supported_major_versions=[10, 11, 12]
+        supported_major_versions=[10, 11]
     )
     return gen_jit_spec(
         uri,
diff --git a/flashinfer/prefill.py b/flashinfer/prefill.py
index ab6881b786..8c18d68041 100755
--- a/flashinfer/prefill.py
+++ b/flashinfer/prefill.py
@@ -53,7 +53,6 @@
     canonicalize_torch_dtype,
     determine_attention_backend,
     device_support_pdl,
-    get_compute_capability,
     get_device_sm_count,
     is_float8,
     is_sm100a_supported,
@@ -100,11 +99,7 @@ def get_fmha_module(
     device: torch.device,
     use_fp16_qk_reduction: bool = False,
 ):
-    if (
-        is_sm100a_supported(device)
-        or is_sm110a_supported(device)
-        or is_sm12x_supported(device)
-    ):
+    if is_sm100a_supported(device) or is_sm110a_supported(device):
         return gen_fmha_cutlass_sm100a_module(
             dtype_q,
             dtype_kv,
@@ -117,15 +112,10 @@ def get_fmha_module(
             use_logits_soft_cap,
         ).build_and_load()
     else:
-        major, minor = get_compute_capability(device)
-        if major == 12:
-            min_cuda = "13.0" if minor >= 1 else "12.8"
-            raise ValueError(
-                f"SM12x detected but CUDA version is too old; "
-                f"SM12{minor}x requires CUDA >= {min_cuda}."
-            )
         raise ValueError(
-            "This device is not supported; requires SM100a, SM110a, or SM12x."
+            "CUTLASS FMHA requires SM100a (B200/GB200) or SM110a. "
+            "SM12x (RTX 5090/DGX Spark) lacks tcgen05 MMA required by this kernel. "
+            "Use backend='fa2' instead."
         )
 
 
@@ -3838,13 +3828,6 @@ def fmha_v2_prefill_deepseek(
         If return_lse is False, the output will be a single tensor.
     """
     if not is_sm12x_supported(query.device):
-        major, minor = get_compute_capability(query.device)
-        if major == 12:
-            min_cuda = "13.0" if minor >= 1 else "12.8"
-            raise ValueError(
-                f"fmha_v2_prefill_deepseek requires CUDA >= {min_cuda} "
-                f"for SM12{minor}x GPUs."
-            )
         raise ValueError("fmha_v2_prefill_deepseek is only supported on SM12x GPUs.")
     assert query.shape[3] == 192 and key.shape[3] == 192 and value.shape[3] == 128, (
         "currently only support deepseek r1 192 query and 128 value"