diff --git a/onnxruntime/python/tools/tensorrt/perf/build/build_image.py b/onnxruntime/python/tools/tensorrt/perf/build/build_image.py
index 2ae64a72d0..b95ad3c0a5 100644
--- a/onnxruntime/python/tools/tensorrt/perf/build/build_image.py
+++ b/onnxruntime/python/tools/tensorrt/perf/build/build_image.py
@@ -14,9 +14,10 @@ import sys
 from typing import List, Optional
 
 TRT_DOCKER_FILES = {
-    "8.4": "tools/ci_build/github/linux/docker/Dockerfile.ubuntu_cuda11_6_tensorrt8_4",
-    "8.5": "tools/ci_build/github/linux/docker/Dockerfile.ubuntu_cuda11_8_tensorrt8_5",
-    "8.6": "tools/ci_build/github/linux/docker/Dockerfile.ubuntu_cuda11_8_tensorrt8_6",
+    "8.4.cuda_11_6_cudnn_8": "tools/ci_build/github/linux/docker/Dockerfile.ubuntu_cuda11_6_tensorrt8_4",
+    "8.5.cuda_11_8_cudnn_8": "tools/ci_build/github/linux/docker/Dockerfile.ubuntu_cuda11_8_tensorrt8_5",
+    "8.6.cuda_11_8_cudnn_8": "tools/ci_build/github/linux/docker/Dockerfile.ubuntu_cuda11_8_tensorrt8_6",
+    "8.6.cuda_12_3_cudnn_9": "tools/ci_build/github/linux/docker/Dockerfile.ubuntu_cuda12_3_tensorrt8_6",
     "BIN": "tools/ci_build/github/linux/docker/Dockerfile.ubuntu_tensorrt_bin",
 }
 
@@ -99,18 +100,11 @@ def docker_build_trt(args: argparse.Namespace):
     :param args: The arguments to this script.
     """
 
-    if not is_valid_ver_str(args.trt_version, min_comps=2, max_comps=4):
-        print(f"[ERROR]: Invalid TensorRT version '{args.trt_version}'", file=sys.stderr)
-        sys.exit(1)
-
-    vers_comps = args.trt_version.split(".")
-    trt_ver_key = f"{vers_comps[0]}.{vers_comps[1]}"
-
-    if trt_ver_key not in TRT_DOCKER_FILES:
+    if args.trt_version not in TRT_DOCKER_FILES:
         print(f"[ERROR]: TensorRT version '{args.trt_version}' is currently unsupported", file=sys.stderr)
         sys.exit(1)
 
-    docker_file = TRT_DOCKER_FILES[trt_ver_key]
+    docker_file = TRT_DOCKER_FILES[args.trt_version]
     docker_file_path = os.path.normpath(os.path.join(args.repo_path, docker_file))
 
     if not os.path.isfile(docker_file_path):
@@ -144,11 +138,7 @@ def docker_build_trt_bin(args: argparse.Namespace):
         sys.exit(1)
 
     if not is_valid_ver_str(args.tar_cuda_version, 2, 2):
-        print("[ERROR]: Must specify a valid CUDA version for binary TensorRT installs (e.g., 11.x)", file=sys.stderr)
-        sys.exit(1)
-
-    if not is_valid_ver_str(args.tar_cudnn_version, 2, 2):
-        print("[ERROR]: Must specify a valid cuDNN version for binary TensorRT installs (e.g., 8.x)", file=sys.stderr)
+        print("[ERROR]: Must specify a valid CUDA version for binary TensorRT installs (e.g., 12.4)", file=sys.stderr)
         sys.exit(1)
 
     if not os.path.isfile(docker_file_path):
@@ -170,8 +160,6 @@ def docker_build_trt_bin(args: argparse.Namespace):
             "--build-arg",
             f"TAR_CUDA_VERSION={args.tar_cuda_version}",
             "--build-arg",
-            f"TAR_CUDNN_VERSION={args.tar_cudnn_version}",
-            "--build-arg",
             f"TRT_BINS_DIR={args.trt_bins_dir}",
             "-f",
             f"{docker_file_path}",
@@ -195,7 +183,9 @@ def parse_arguments() -> argparse.Namespace:
     parser.add_argument("-r", "--repo_path", required=True, help="Path to the onnxruntime repository")
     parser.add_argument("-i", "--image_name", required=True, help="The resulting Docker image name")
     parser.add_argument("-b", "--branch", default="main", help="Name of the onnxruntime git branch to checkout")
-    parser.add_argument("-t", "--trt_version", default="8.6.1.6", help="TensorRT version (e.g., 8.6.1.6)")
+    parser.add_argument(
+        "-t", "--trt_version", default="8.6.cuda_11_8_cudnn_8", help="TensorRT version (e.g., 8.6.cuda_11_8_cudnn_8)"
+    )
     parser.add_argument("-a", "--cuda_arch", default="75", help="CUDA architecture (e.g., 75)")
 
     # Command-line options for installing TensorRT from binaries.
@@ -208,12 +198,7 @@ def parse_arguments() -> argparse.Namespace:
     parser.add_argument(
         "--tar_cuda_version",
         default="",
-        help="CUDA version (e.g., 11.8) used to find TensorRT EA binary tar.gz package",
-    )
-    parser.add_argument(
-        "--tar_cudnn_version",
-        default="",
-        help="CUDA version (e.g., 8.6) used to find TensorRT EA binary tar.gz package",
+        help="CUDA version (e.g., 12.4) used to find TensorRT EA binary tar.gz package",
     )
     parser.add_argument("--trt_bins_dir", default="", help="Directory containing TensorRT tar.gz package")
     parser.add_argument(
diff --git a/tools/ci_build/github/azure-pipelines/linux-gpu-tensorrt-daily-perf-pipeline.yml b/tools/ci_build/github/azure-pipelines/linux-gpu-tensorrt-daily-perf-pipeline.yml
index 9f3a127262..15f558e6f9 100644
--- a/tools/ci_build/github/azure-pipelines/linux-gpu-tensorrt-daily-perf-pipeline.yml
+++ b/tools/ci_build/github/azure-pipelines/linux-gpu-tensorrt-daily-perf-pipeline.yml
@@ -8,15 +8,16 @@ parameters:
 - name: TrtVersion
   displayName: TensorRT Version
   type: string
-  default: 8.6.1.6
+  default: 8.6.cuda_11_8_cudnn_8
   values:
-  - 8.4.1.5
-  - 8.5.1.1
-  - 8.6.1.6
+  - 8.4.cuda_11_6_cudnn_8
+  - 8.5.cuda_11_8_cudnn_8
+  - 8.6.cuda_11_8_cudnn_8
+  - 8.6.cuda_12_3_cudnn_9
   - BIN
 
 - name: UseTensorrtOssParser
-  displayName: Use TensorRT-OSS Parser
+  displayName: Use TensorRT-OSS Parser (not compatible with BIN)
   type: boolean
   default: false
 
@@ -86,11 +87,11 @@ jobs:
       - script: 'ls -al $(trtBinsDir)'
         displayName: 'Show available TensorRT .tar.gz packages'
 
-      - script: 'cp $(trtBinsDir)/TensorRT-$(trtVersion).Linux.x86_64-gnu.cuda-$(tarCudaVersion).cudnn$(tarCudnnVersion).tar.gz $(Build.SourcesDirectory)/onnxruntime/python/tools/tensorrt/perf/build/'
+      - script: 'cp $(trtBinsDir)/TensorRT-$(trtVersion).Linux.x86_64-gnu.cuda-$(tarCudaVersion).tar.gz $(Build.SourcesDirectory)/onnxruntime/python/tools/tensorrt/perf/build/'
         displayName: 'Copy TensorRT .tar.gz package into Docker build directory'
 
-      - script: 'python3 $(Build.SourcesDirectory)/onnxruntime/python/tools/tensorrt/perf/build/build_image.py -r $(Build.SourcesDirectory) -i $(image) -b $(branchName) -t $(trtVersion) -a 75 --install_bin --tar_cuda_version=$(tarCudaVersion) --tar_cudnn_version=$(tarCudnnVersion) --trt_bins_dir=.'
-        displayName: 'Install TensorRT from binaries and build latest ORT Image'
+      - script: 'python3 $(Build.SourcesDirectory)/onnxruntime/python/tools/tensorrt/perf/build/build_image.py -r $(Build.SourcesDirectory) -i $(image) -b $(branchName) -t $(trtVersion) -a 75 --install_bin --tar_cuda_version=$(tarCudaVersion) --trt_bins_dir=.'
+        displayName: 'Install TensorRT $(tarTrtVersion) from binaries and build latest ORT Image'
         workingDirectory: '$(Build.SourcesDirectory)/onnxruntime/python/tools/tensorrt/perf/build'
     
     # Build ORT with TensorRT built-in parser 
diff --git a/tools/ci_build/github/linux/docker/Dockerfile.ubuntu_cuda12_3_tensorrt8_6 b/tools/ci_build/github/linux/docker/Dockerfile.ubuntu_cuda12_3_tensorrt8_6
new file mode 100644
index 0000000000..9493480784
--- /dev/null
+++ b/tools/ci_build/github/linux/docker/Dockerfile.ubuntu_cuda12_3_tensorrt8_6
@@ -0,0 +1,96 @@
+# --------------------------------------------------------------
+# Copyright (c) Microsoft Corporation. All rights reserved.
+# Licensed under the MIT License.
+# --------------------------------------------------------------
+# Dockerfile to run ONNXRuntime with TensorRT integration
+
+# Build base image with required system packages
+FROM nvidia/cuda:12.3.1-devel-ubuntu20.04 AS base
+
+# The local directory into which to build and install CMAKE
+ARG ONNXRUNTIME_LOCAL_CODE_DIR=/code
+
+ENV PATH /usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/src/tensorrt/bin:${ONNXRUNTIME_LOCAL_CODE_DIR}/cmake-3.27.3-linux-x86_64/bin:/opt/miniconda/bin:${PATH}
+ENV DEBIAN_FRONTEND=noninteractive
+
+RUN apt-get update &&\
+    apt-get install -y sudo git bash unattended-upgrades wget
+RUN unattended-upgrade
+
+# Install python3
+RUN apt-get install -y --no-install-recommends \
+    python3 \
+    python3-pip \
+    python3-dev \
+    python3-wheel &&\
+    cd /usr/local/bin &&\
+    ln -s /usr/bin/python3 python &&\
+    ln -s /usr/bin/pip3 pip;
+
+RUN pip install --upgrade pip 
+RUN pip install setuptools>=68.2.2
+
+# Install cuDNN v9
+RUN apt-get -y install cudnn9-cuda-12
+
+# Install TensorRT
+RUN v="8.6.1.6-1+cuda12.0" &&\
+    apt-key adv --fetch-keys https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2004/x86_64/7fa2af80.pub &&\
+    apt-get update &&\
+    sudo apt-get install -y libnvinfer8=${v} libnvonnxparsers8=${v} libnvparsers8=${v} libnvinfer-plugin8=${v} libnvinfer-lean8=${v} libnvinfer-vc-plugin8=${v} libnvinfer-dispatch8=${v}\
+        libnvinfer-headers-dev=${v} libnvinfer-headers-plugin-dev=${v} libnvinfer-dev=${v} libnvonnxparsers-dev=${v} libnvparsers-dev=${v} libnvinfer-plugin-dev=${v} libnvinfer-lean-dev=${v} libnvinfer-vc-plugin-dev=${v}  libnvinfer-dispatch-dev=${v}\
+        python3-libnvinfer=${v} libnvinfer-samples=${v} tensorrt-dev=${v} tensorrt-libs=${v}
+
+# Compile trtexec
+RUN cd /usr/src/tensorrt/samples/trtexec && make
+
+# Install Valgrind
+RUN apt-get install -y valgrind
+
+# Build final image from base. Builds ORT.
+FROM base as final
+ARG BUILD_USER=onnxruntimedev
+ARG BUILD_UID=1000
+RUN adduser --gecos 'onnxruntime Build User' --disabled-password $BUILD_USER --uid $BUILD_UID
+USER $BUILD_USER
+
+# ONNX Runtime arguments
+
+# URL to the github repo from which to clone ORT.
+ARG ONNXRUNTIME_REPO=https://github.com/Microsoft/onnxruntime
+
+# The local directory into which to clone ORT.
+ARG ONNXRUNTIME_LOCAL_CODE_DIR=/code
+
+# The git branch of ORT to checkout and build.
+ARG ONNXRUNTIME_BRANCH=main
+
+# Optional. The specific commit to pull and build from. If not set, the latest commit is used.
+ARG ONNXRUNTIME_COMMIT_ID
+
+# The supported CUDA architecture
+ARG CMAKE_CUDA_ARCHITECTURES=75
+
+WORKDIR ${ONNXRUNTIME_LOCAL_CODE_DIR}
+
+# Clone ORT repository with branch
+RUN git clone --single-branch --branch ${ONNXRUNTIME_BRANCH} --recursive ${ONNXRUNTIME_REPO} onnxruntime &&\
+    /bin/sh onnxruntime/dockerfiles/scripts/install_common_deps.sh
+
+WORKDIR ${ONNXRUNTIME_LOCAL_CODE_DIR}/onnxruntime
+
+# Reset to a specific commit if specified by build args.
+RUN if [ -z "$ONNXRUNTIME_COMMIT_ID" ] ; then echo "Building branch ${ONNXRUNTIME_BRANCH}" ;\
+    else echo "Building branch ${ONNXRUNTIME_BRANCH} @ commit ${ONNXRUNTIME_COMMIT_ID}" &&\
+    git reset --hard ${ONNXRUNTIME_COMMIT_ID} && git submodule update --recursive ; fi
+
+# Build ORT
+ENV CUDA_MODULE_LOADING "LAZY" 
+ARG PARSER_CONFIG=""
+RUN /bin/sh build.sh ${PARSER_CONFIG} --parallel --build_shared_lib --cuda_home /usr/local/cuda --cudnn_home /usr/lib/x86_64-linux-gnu/ --use_tensorrt --tensorrt_home /usr/lib/x86_64-linux-gnu/ --config Release --build_wheel --skip_tests --skip_submodule_sync --cmake_extra_defines '"CMAKE_CUDA_ARCHITECTURES='${CMAKE_CUDA_ARCHITECTURES}'"'
+
+# Switch to root to continue following steps of CI
+USER root
+
+# Intall ORT wheel
+RUN pip install ${ONNXRUNTIME_LOCAL_CODE_DIR}/onnxruntime/build/Linux/Release/dist/*.whl
\ No newline at end of file
diff --git a/tools/ci_build/github/linux/docker/Dockerfile.ubuntu_tensorrt_bin b/tools/ci_build/github/linux/docker/Dockerfile.ubuntu_tensorrt_bin
index 21b09b2d89..a26bf88fbb 100644
--- a/tools/ci_build/github/linux/docker/Dockerfile.ubuntu_tensorrt_bin
+++ b/tools/ci_build/github/linux/docker/Dockerfile.ubuntu_tensorrt_bin
@@ -4,29 +4,15 @@
 # --------------------------------------------------------------
 # Dockerfile to run ONNXRuntime with TensorRT installed from provided binaries
 
-FROM nvidia/cuda:11.7.1-cudnn8-devel-ubuntu20.04
+# Build base image with required system packages
+FROM nvidia/cuda:12.3.1-devel-ubuntu20.04 AS base
 
+# The local directory into which to build and install CMAKE
+ARG ONNXRUNTIME_LOCAL_CODE_DIR=/code
 
-# ONNX Runtime Variables
-ARG ONNXRUNTIME_REPO=https://github.com/Microsoft/onnxruntime
-ARG ONNXRUNTIME_BRANCH=main
-ARG CMAKE_CUDA_ARCHITECTURES=37;50;52;60;61;70;75;80
-
-# Must provide version numbers used to build the name of the tar file containing TensorRT binaries.
-# See: https://docs.nvidia.com/deeplearning/tensorrt/install-guide/index.html#installing-tar
-ARG TAR_TRT_VERSION
-ARG TAR_CUDA_VERSION
-ARG TAR_CUDNN_VERSION
-
-# Directory containing TensorRT tar.gz installation package
-ARG TRT_BINS_DIR=.
-
-ENV PATH /usr/local/nvidia/bin:/usr/local/cuda/bin:/code/cmake-3.27.3-linux-x86_64/bin:/opt/miniconda/bin:${PATH}
-
+ENV PATH /usr/local/nvidia/bin:/usr/local/cuda/bin:${ONNXRUNTIME_LOCAL_CODE_DIR}/cmake-3.27.3-linux-x86_64/bin:/opt/miniconda/bin:${PATH}
 ENV DEBIAN_FRONTEND=noninteractive
 
-COPY ${TRT_BINS_DIR}/TensorRT-${TAR_TRT_VERSION}.Linux.x86_64-gnu.cuda-${TAR_CUDA_VERSION}.cudnn${TAR_CUDNN_VERSION}.tar.gz /TensorRT-${TAR_TRT_VERSION}.tar.gz
-
 RUN apt-get update &&\
     apt-get install -y sudo git bash unattended-upgrades wget
 RUN unattended-upgrade
@@ -44,22 +30,77 @@ RUN apt-get install -y --no-install-recommends \
 RUN pip install --upgrade pip 
 RUN pip install setuptools>=68.2.2
 
+# Install cuDNN v9
+RUN apt-get -y install cudnn9-cuda-12
+
+# Install TensorRT
+# Must provide version numbers used to build the name of the tar file containing TensorRT binaries.
+# See: https://docs.nvidia.com/deeplearning/tensorrt/install-guide/index.html#installing-tar
+ARG TAR_TRT_VERSION
+ARG TAR_CUDA_VERSION
+
+# Directory containing TensorRT tar.gz installation package
+ARG TRT_BINS_DIR=.
+COPY ${TRT_BINS_DIR}/TensorRT-${TAR_TRT_VERSION}.Linux.x86_64-gnu.cuda-${TAR_CUDA_VERSION}.tar.gz /TensorRT-${TAR_TRT_VERSION}.tar.gz
+
 # Install TensorRT from tar.gz
 RUN tar -xzvf /TensorRT-${TAR_TRT_VERSION}.tar.gz
 
 RUN cd /TensorRT-${TAR_TRT_VERSION}/python &&\
-    python3 -m pip install tensorrt-${TAR_TRT_VERSION}-cp38-none-linux_x86_64.whl
+    python3 -m pip install tensorrt*cp38*.whl
 
 RUN cp -r /TensorRT-${TAR_TRT_VERSION}/lib/* /usr/lib/x86_64-linux-gnu/
 RUN cp /TensorRT-${TAR_TRT_VERSION}/include/* /usr/local/include/
 RUN cp /TensorRT-${TAR_TRT_VERSION}/bin/* /usr/local/bin/
 
-WORKDIR /code
+# Install Valgrind
+RUN apt-get install -y valgrind
+
+# Build final image from base. Builds ORT.
+FROM base as final
+ARG BUILD_USER=onnxruntimedev
+ARG BUILD_UID=1000
+RUN adduser --gecos 'onnxruntime Build User' --disabled-password $BUILD_USER --uid $BUILD_UID
+USER $BUILD_USER
+
+# ONNX Runtime arguments
+
+# URL to the github repo from which to clone ORT.
+ARG ONNXRUNTIME_REPO=https://github.com/Microsoft/onnxruntime
+
+# The local directory into which to clone ORT.
+ARG ONNXRUNTIME_LOCAL_CODE_DIR=/code
+
+# The git branch of ORT to checkout and build.
+ARG ONNXRUNTIME_BRANCH=main
+
+# Optional. The specific commit to pull and build from. If not set, the latest commit is used.
+ARG ONNXRUNTIME_COMMIT_ID
+
+# The supported CUDA architecture
+ARG CMAKE_CUDA_ARCHITECTURES=75
 
 # Prepare onnxruntime repository & build onnxruntime with TensorRT
+WORKDIR ${ONNXRUNTIME_LOCAL_CODE_DIR}
+
+# Clone ORT repository with branch
 RUN git clone --single-branch --branch ${ONNXRUNTIME_BRANCH} --recursive ${ONNXRUNTIME_REPO} onnxruntime &&\
-    /bin/sh onnxruntime/dockerfiles/scripts/install_common_deps.sh &&\
-    cd onnxruntime &&\
-    /bin/sh build.sh --parallel --build_shared_lib --cuda_home /usr/local/cuda --cudnn_home /usr/lib/x86_64-linux-gnu/ --use_tensorrt --tensorrt_home /usr/lib/x86_64-linux-gnu/ --config Release --build_wheel --skip_tests --skip_submodule_sync --cmake_extra_defines '"CMAKE_CUDA_ARCHITECTURES='${CMAKE_CUDA_ARCHITECTURES}'"' &&\
-    pip install /code/onnxruntime/build/Linux/Release/dist/*.whl &&\
-    cd .. 
+    /bin/sh onnxruntime/dockerfiles/scripts/install_common_deps.sh
+
+WORKDIR ${ONNXRUNTIME_LOCAL_CODE_DIR}/onnxruntime
+
+# Reset to a specific commit if specified by build args.
+RUN if [ -z "$ONNXRUNTIME_COMMIT_ID" ] ; then echo "Building branch ${ONNXRUNTIME_BRANCH}" ;\
+    else echo "Building branch ${ONNXRUNTIME_BRANCH} @ commit ${ONNXRUNTIME_COMMIT_ID}" &&\
+    git reset --hard ${ONNXRUNTIME_COMMIT_ID} && git submodule update --recursive ; fi
+
+# Build ORT
+ENV CUDA_MODULE_LOADING "LAZY" 
+ARG PARSER_CONFIG=""
+RUN /bin/sh build.sh ${PARSER_CONFIG} --parallel --build_shared_lib --cuda_home /usr/local/cuda --cudnn_home /usr/lib/x86_64-linux-gnu/ --use_tensorrt --tensorrt_home /usr/lib/x86_64-linux-gnu/ --config Release --build_wheel --skip_tests --skip_submodule_sync --cmake_extra_defines '"CMAKE_CUDA_ARCHITECTURES='${CMAKE_CUDA_ARCHITECTURES}'"'
+
+# Switch to root to continue following steps of CI
+USER root
+
+# Intall ORT wheel
+RUN pip install ${ONNXRUNTIME_LOCAL_CODE_DIR}/onnxruntime/build/Linux/Release/dist/*.whl
\ No newline at end of file