White Papers

10 CheXNet Inference with Nvidia T4 on Dell EMC PowerEdge R7425 | Document ID
2.2 Test Setup
a) For the hardware, we selected PowerEdge 7425 which includes the Nvidia Tesla T4 GPU, the
most advanced accelerator for AI inference workloads. According to Nvidia, T4’s new Turing
Tensor cores accelerate int8 precision more than 2x faster than the previous generation low-
power offering [2].
b) For the framework and inference optimizer tools, we selected TensorFlow, TF-TRT integrated
and TensorRT C++ API, since they have better technical support and a wide variety of pre-
trained models are readily available.
c) Most of the tests were run in int8 precision mode, since it has significantly lower precision and
dynamic range than fp32, as well as lower memory requirements; therefore, it allows higher
throughput at lower latency.
Table 3 shows the software stack configuration on PowerEdge R7425
Table 3. OS and Software Stack Configuration
Software
Version
OS
Ubuntu 16.04.5 LTS
Kernel
GNU/Linux 4.4.0-133-generic x86_64
Nvidia-driver
410.79
CUDA
10.0
TensorFlow version
1.10
TensorRT™ version
5.0
Docker Image for TensorFlow CPU only
tensorflow:1.10.0-py3
Docker Image for TensorFlow GPU only
nvcr.io/nvidia/tensorflow:18.10-py3
Docker Image for TF-TRT integration
nvcr.io/nvidia/tensorflow:18.10-py3
Docker Image for TensorRT™ C++ API
nvcr.io/nvidia/tensorrt:18.11-py3
Script samples source
Samples included within the docker images
Test Date
February 2019