diff --git a/genesis/engine/solvers/rigid/rigid_solver.py b/genesis/engine/solvers/rigid/rigid_solver.py
index 053d275ac..42e926cd1 100644
--- a/genesis/engine/solvers/rigid/rigid_solver.py
+++ b/genesis/engine/solvers/rigid/rigid_solver.py
@@ -454,6 +454,9 @@ def _build_static_config(self):
         # Prefer the monolith solver on CPU (always faster there, perf dispatch is a waste of effort)
         if gs.backend == gs.cpu or self.sim.options.requires_grad:
             static_rigid_sim_config["prefer_decomposed_solver"] = 0
+        # Prefer the decomposed solver on CUDA, which supports hardware accelerated graph and do-while condition
+        elif gs.backend == gs.cuda:
+            static_rigid_sim_config["prefer_decomposed_solver"] = 1
 
         if self.is_active:
             # TODO: These alternative tiled algorithms are designed to reduce the impact of latency. However, naive
diff --git a/genesis/ext/_trimesh_patch.py b/genesis/ext/_trimesh_patch.py
index 95d0afd31..906f23d65 100644
--- a/genesis/ext/_trimesh_patch.py
+++ b/genesis/ext/_trimesh_patch.py
@@ -1,5 +1,6 @@
 import os
 import re
+import warnings
 from collections import defaultdict, deque
 
 import numpy as np
@@ -25,6 +26,14 @@
 import genesis as gs
 
 
+# Silence trimesh mass_properties divide-by-zero on degenerate meshes.
+warnings.filterwarnings(
+    "ignore",
+    category=RuntimeWarning,
+    message=r"(divide by zero|invalid value) encountered in (divide|scalar multiply)",
+)
+
+
 def load_obj(
     file_obj,
     resolver=None,
diff --git a/pyproject.toml b/pyproject.toml
index d0d94360c..094aad52c 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -178,7 +178,9 @@ ignore = [
 [tool.pytest.ini_options]
 addopts = [
     "--verbose",
-    "-r", "a",
+    # Exclude 'f' (failed) from short test summary because it duplicates the FAILURES section in verbose mode.
+    # A plain list of failed test IDs is printed instead by 'pytest_terminal_summary' hook in tests/conftest.py.
+    "-r", "EsxXw",
     "--color=yes",
     "--import-mode=importlib",
     "--pdbcls=IPython.terminal.debugger:TerminalPdb",
diff --git a/tests/conftest.py b/tests/conftest.py
index 6715188aa..63497ccbb 100644
--- a/tests/conftest.py
+++ b/tests/conftest.py
@@ -487,6 +487,29 @@ def pytest_runtest_makereport(item, call):
             report.longrepr = (os.path.relpath(__file__), lineno, reason)
 
 
+def pytest_terminal_summary(terminalreporter, exitstatus, config):
+    """Print a plain list of failed test IDs at the end of the run.
+
+    Replaces pytest's default 'FAILED test_id - msg' short summary lines (which in verbose mode
+    duplicate the FAILURES section) with a compact, easy-to-scan ID-only list.
+    """
+    failed = terminalreporter.stats.get("failed")
+    if not failed:
+        return
+    terminalreporter.write_sep("=", "Failed tests")
+    fullwidth = terminalreporter._tw.fullwidth
+    for report in failed:
+        reprcrash = getattr(report.longrepr, "reprcrash", None)
+        msg = " ".join(reprcrash.message.split()) if reprcrash is not None else ""
+        if not msg:
+            terminalreporter.write_line(report.nodeid)
+            continue
+        line = f"{report.nodeid} - {msg}"
+        if len(line) > fullwidth:
+            line = line[: fullwidth - 3] + "..."
+        terminalreporter.write_line(line)
+
+
 def pytest_addoption(parser: pytest.Parser) -> None:
     parser.addoption("--backend", action="store", default=None, help="Default simulation backend.")
     parser.addoption(