diff --git a/python/src/ir.cc b/python/src/ir.cc
index abbf7ab41828..2ef6c4410e59 100644
--- a/python/src/ir.cc
+++ b/python/src/ir.cc
@@ -1469,7 +1469,8 @@ void init_triton_ir(py::module &&m) {
            })
       .def("create_log",
            [](TritonOpBuilder &self, Value &val) -> Value {
-             return approxMath(self.create<math::LogOp>(val));
+             // TODO: switch to approxMath.
+             return self.create<math::LogOp>(val);
            })
       .def("create_log2",
            [](TritonOpBuilder &self, Value &val) -> Value {
diff --git a/python/test/unit/language/test_core.py b/python/test/unit/language/test_core.py
index 5cec71b40175..5b9a0b20bc31 100644
--- a/python/test/unit/language/test_core.py
+++ b/python/test/unit/language/test_core.py
@@ -4385,7 +4385,7 @@ def kernel(X, Y, BLOCK: tl.constexpr):
     k = kernel[(1, )](x, y, BLOCK=shape[0])
     torch.allclose(getattr(torch, func_str)(x), y, rtol=1e-3)
 
-    if func_str in ['log', 'log2'] and is_cuda():
+    if func_str in ['log2'] and is_cuda():
         assert 'lg2.approx.ftz.f32' in k.asm['ptx']
     if func_str in ['exp', 'exp2'] and is_cuda():
         assert 'ex2.approx.ftz.f32' in k.asm['ptx']