diff --git a/ci/docker/Dockerfile.build.centos7_gpu b/ci/docker/Dockerfile.build.centos7_gpu
index 8bf2442731af..d9878743fd58 100644
--- a/ci/docker/Dockerfile.build.centos7_gpu
+++ b/ci/docker/Dockerfile.build.centos7_gpu
@@ -18,7 +18,7 @@
 #
 # Dockerfile to build and run MXNet on CentOS 7 for GPU
 
-FROM nvidia/cuda:9.1-cudnn7-devel-centos7
+FROM nvidia/cuda:10.0-cudnn7-devel-centos7
 
 WORKDIR /work/deps
 
@@ -34,6 +34,8 @@ COPY install/centos7_adduser.sh /work/
 RUN /work/centos7_adduser.sh
 
 ENV PYTHONPATH=./python/
+ENV CUDA_ARCH_LIST="5.2 7.0"
+
 WORKDIR /work/mxnet
 
 COPY runtime_functions.sh /work/
diff --git a/ci/docker/Dockerfile.build.ubuntu_base_gpu b/ci/docker/Dockerfile.build.ubuntu_base_gpu
index 99b79f513bee..61e3c5cb1b3f 100644
--- a/ci/docker/Dockerfile.build.ubuntu_base_gpu
+++ b/ci/docker/Dockerfile.build.ubuntu_base_gpu
@@ -19,7 +19,7 @@
 # Dockerfile to run the MXNet Installation Tests on Ubuntu 16.04
 # This should run in an empty docker with ubuntu and cuda.
 
-FROM nvidia/cuda:9.1-cudnn7-devel
+FROM nvidia/cuda:10.0-cudnn7-devel-ubuntu16.04
 
 WORKDIR /work/deps
 
diff --git a/ci/docker/Dockerfile.build.ubuntu_build_cuda b/ci/docker/Dockerfile.build.ubuntu_build_cuda
index 9ed0cbbe3e52..0d1d02d1636c 100644
--- a/ci/docker/Dockerfile.build.ubuntu_build_cuda
+++ b/ci/docker/Dockerfile.build.ubuntu_build_cuda
@@ -21,7 +21,7 @@
 # package generation, requiring the actual CUDA library to be
 # present
 
-FROM nvidia/cuda:9.1-cudnn7-devel
+FROM nvidia/cuda:10.0-cudnn7-devel-ubuntu16.04
 
 WORKDIR /work/deps
 
diff --git a/ci/docker/Dockerfile.build.ubuntu_gpu b/ci/docker/Dockerfile.build.ubuntu_gpu
index 8fcbcbbb9674..72af6cb87204 100644
--- a/ci/docker/Dockerfile.build.ubuntu_gpu
+++ b/ci/docker/Dockerfile.build.ubuntu_gpu
@@ -18,7 +18,7 @@
 #
 # Dockerfile to run MXNet on Ubuntu 16.04 for GPU
 
-FROM nvidia/cuda:9.1-cudnn7-devel
+FROM nvidia/cuda:10.0-cudnn7-devel-ubuntu16.04
 
 WORKDIR /work/deps
 
diff --git a/ci/docker/Dockerfile.build.ubuntu_nightly_gpu b/ci/docker/Dockerfile.build.ubuntu_nightly_gpu
index deeed8b0d52a..a5fd804650ab 100644
--- a/ci/docker/Dockerfile.build.ubuntu_nightly_gpu
+++ b/ci/docker/Dockerfile.build.ubuntu_nightly_gpu
@@ -15,10 +15,8 @@
 # KIND, either express or implied.  See the License for the
 # specific language governing permissions and limitations
 # under the License.
-#
-# Dockerfile to run MXNet on Ubuntu 16.04 for CPU
 
-FROM nvidia/cuda:9.1-cudnn7-devel
+FROM nvidia/cuda:10.0-cudnn7-devel-ubuntu16.04
 
 WORKDIR /work/deps
 
diff --git a/ci/docker/runtime_functions.sh b/ci/docker/runtime_functions.sh
index 27c93b5049ef..104c5aa6a176 100755
--- a/ci/docker/runtime_functions.sh
+++ b/ci/docker/runtime_functions.sh
@@ -292,20 +292,18 @@ build_centos7_mkldnn() {
 build_centos7_gpu() {
     set -ex
     cd /work/mxnet
-    # unfortunately this build has problems in 3rdparty dependencies with ccache and make
-    # build_ccache_wrappers
-    make \
-        DEV=1                                     \
-        ENABLE_TESTCOVERAGE=1                     \
-        USE_LAPACK=1                              \
-        USE_LAPACK_PATH=/usr/lib64/liblapack.so   \
-        USE_BLAS=openblas                         \
-        USE_CUDA=1                                \
-        USE_CUDA_PATH=/usr/local/cuda             \
-        USE_CUDNN=1                               \
-        USE_DIST_KVSTORE=1                        \
-        CUDA_ARCH="$CI_CUDA_COMPUTE_CAPABILITIES" \
-        -j$(nproc)
+    cmake \
+        -DCMAKE_CXX_COMPILER_LAUNCHER=ccache \
+        -DCMAKE_C_COMPILER_LAUNCHER=ccache \
+        -DENABLE_TESTCOVERAGE=ON \
+        -DUSE_CUDA=ON \
+        -DUSE_CUDNN=ON \
+        -DCUDA_ARCH_LIST="${CUDA_ARCH_LIST}" \
+        -DUSE_SIGNAL_HANDLER=ON \
+        -DCMAKE_BUILD_TYPE=Release \
+        -DUSE_DIST_KVSTORE=ON \
+        -G Ninja /work/mxnet
+    ninja -v
 }
 
 build_ubuntu_cpu() {