TensorRT Rotate ops failed unit test on RTX 4090

[BaophanN]

Dear author, thanks for the amazing work. I built and run the unit test of tensorRT on RTX 2070 and RTX 4090. The former passed all test while the latter failed all test case in Rotate operators. What is the reason for this? Thank you.
Here are some samples failed test cases, given the build process happens without any problems.

======================================================================
FAIL: test_fp16_nearest (det2trt.models.utils.test_trt_ops.test_rotate.RotateTestCase2)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/workspace/source/BEVFormer_tensorrt/./det2trt/models/utils/test_trt_ops/test_rotate.py", line 223, in test_fp16_nearest
    self.fp16_case(0.6)
  File "/workspace/source/BEVFormer_tensorrt/./det2trt/models/utils/test_trt_ops/test_rotate.py", line 195, in fp16_case
    self.assertLessEqual(cost, delta)
AssertionError: 1.01 not less than or equal to 0.6

======================================================================
FAIL: test_int8_fp16_bilinear (det2trt.models.utils.test_trt_ops.test_rotate.RotateTestCase2)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/workspace/source/BEVFormer_tensorrt/./det2trt/models/utils/test_trt_ops/test_rotate.py", line 232, in test_int8_fp16_bilinear
    self.int8_fp16_case(0.3)
  File "/workspace/source/BEVFormer_tensorrt/./det2trt/models/utils/test_trt_ops/test_rotate.py", line 211, in int8_fp16_case
    self.assertLessEqual(cost, delta)
AssertionError: 0.6772 not less than or equal to 0.3

======================================================================
FAIL: test_int8_fp16_nearest (det2trt.models.utils.test_trt_ops.test_rotate.RotateTestCase2)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/workspace/source/BEVFormer_tensorrt/./det2trt/models/utils/test_trt_ops/test_rotate.py", line 235, in test_int8_fp16_nearest
    self.int8_fp16_case(0.5)
  File "/workspace/source/BEVFormer_tensorrt/./det2trt/models/utils/test_trt_ops/test_rotate.py", line 211, in int8_fp16_case
    self.assertLessEqual(cost, delta)
AssertionError: 1.01 not less than or equal to 0.5

----------------------------------------------------------------------

Here are the full logs I record while running unit test on each GPU.

rtx4090_test_trt_ops.log
rtx2070_test_trt_ops.log