diff --git a/tensorflow/core/profiler/internal/gpu/BUILD b/tensorflow/core/profiler/internal/gpu/BUILD
index 6fc78e46862..bfe855ef417 100644
--- a/tensorflow/core/profiler/internal/gpu/BUILD
+++ b/tensorflow/core/profiler/internal/gpu/BUILD
@@ -53,7 +53,10 @@ tf_cc_test_gpu(
     srcs = ["device_tracer_test.cc"],
     args = ["--heap_check=local"],
     linkstatic = tf_kernel_tests_linkstatic(),
-    tags = tf_cuda_tests_tags() + ["nomac"],
+    tags = tf_cuda_tests_tags() + [
+        "nomac",
+        "gpu_cupti",
+    ],
     deps = [
         ":device_tracer",
         "//tensorflow/cc:cc_ops",
diff --git a/tensorflow/python/BUILD b/tensorflow/python/BUILD
index 2f65cb57eca..d8d9c3a8872 100644
--- a/tensorflow/python/BUILD
+++ b/tensorflow/python/BUILD
@@ -6409,6 +6409,9 @@ cuda_py_test(
     size = "small",
     srcs = ["client/timeline_test.py"],
     python_version = "PY3",
+    tags = [
+        "gpu_cupti",
+    ],
     xla_enable_strict_auto_jit = False,  # Graph structure is different with autojit
     deps = [
         ":client",
diff --git a/tensorflow/tools/ci_build/presubmit/ubuntu_16/gpu_py36_full/build.sh b/tensorflow/tools/ci_build/presubmit/ubuntu_16/gpu_py36_full/build.sh
index 1498063630a..6a7e4c74576 100644
--- a/tensorflow/tools/ci_build/presubmit/ubuntu_16/gpu_py36_full/build.sh
+++ b/tensorflow/tools/ci_build/presubmit/ubuntu_16/gpu_py36_full/build.sh
@@ -45,12 +45,14 @@ function run_build () {
   export ACTION_PATH="/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin"
   export PYTHON_BIN_PATH="/usr/bin/python3"
   export TF2_BEHAVIOR=1
-  tag_filters="gpu,-no_gpu,-nogpu,-benchmark-test,-no_oss,-oss_serial,-no_gpu_presubmit""$(maybe_skip_v1)"
+  # TODO(b/152356894):
+  # Remove -gpu_cupti once RBE supports cupti tests.
+  tag_filters="gpu,-no_gpu,-nogpu,-benchmark-test,-no_oss,-oss_serial,-no_gpu_presubmit,-gpu_cupti""$(maybe_skip_v1)"
 
   # Get the default test targets for bazel.
   source tensorflow/tools/ci_build/build_scripts/PRESUBMIT_BUILD_TARGETS.sh
 
-  RBE_CONFIG="@ubuntu16.04-py3-gcc7_manylinux2010-cuda10.0-cudnn7-tensorrt5.1"
+  RBE_CONFIG="@ubuntu16.04-py3-gcc7_manylinux2010-cuda10.1-cudnn7-tensorrt6.0"
   TF_CUDA_CONFIG_REPO="${RBE_CONFIG}_config_cuda"
   TF_TENSORRT_CONFIG_REPO="${RBE_CONFIG}_config_tensorrt"
   TF_PYTHON_CONFIG_REPO="${RBE_CONFIG}_config_python"