diff --git a/test/registered/cuda_graph/test_piecewise_cuda_graph_2_gpu.py b/test/registered/cuda_graph/test_piecewise_cuda_graph_2_gpu.py
index ac8176210..dfaca8270 100644
--- a/test/registered/cuda_graph/test_piecewise_cuda_graph_2_gpu.py
+++ b/test/registered/cuda_graph/test_piecewise_cuda_graph_2_gpu.py
@@ -12,7 +12,9 @@ from sglang.test.test_utils import (
 )
 
 # CI Registration - 2-GPU tests (80GB GPUs required)
-register_cuda_ci(est_time=160, suite="stage-b-test-large-2-gpu")
+register_cuda_ci(
+    est_time=160, suite="stage-b-test-large-2-gpu", disabled="see issue #16691"
+)
 
 
 class TestPiecewiseCudaGraphTP(CustomTestCase):