lower vs_theoretical conv tflops threshold for nv (#7811)

less flaky
This commit is contained in:
chenyu
2024-11-20 20:03:49 -05:00
committed by GitHub
parent 46aa23539f
commit 11cea00090

View File

@@ -89,7 +89,7 @@ class TestKernelSpeed(unittest.TestCase):
def test_gemv_16384_4096(self): self._test_matmul(16384, 4096, 1, nv_gbs=430, amd_gbs=400)
def test_gemv_4096_16384(self): self._test_matmul(4096, 16384, 1, nv_gbs=430, amd_gbs=400)
def test_conv_3x3_256_32_32_256_256(self): self._test_conv_3x3(256, 32, 32, 256, 256, nv_tflops=30, amd_tflops=22)
def test_conv_3x3_256_32_32_256_256(self): self._test_conv_3x3(256, 32, 32, 256, 256, nv_tflops=27, amd_tflops=22)
if __name__ == '__main__':
unittest.main()