Finding up-to-date, standardised can be quite difficult among the spate of recent and Nvidia library updates. As such I am proposing we pool our own results using the nightly-build benchmark script:

https://github.com/tensorflow/benchmarks/tree/master/scripts/tf_cnn_benchmarks

It would be interesting to see if we can get any Turing cards benchmarked.

The benchmarks themselves only take a couple of minutes to run, and require no installation or setup. Once you have cloned the repo, simply run the following snippet:

#!/bin/bash

### Change this to match the repo root dir ###
RPATH="$/repos/benchmarks"

#Ouput results to  dir
RESULTS="$/tf_cnn_bench_results"
# 
FPATH="$RPATH/scripts/tf_cnn_benchmarks"
#Run the benches
python "$FPATH/tf_cnn_benchmarks.py" --num_gpus=1 --batch_size=64 --variable_update=parameter_server --optimizer=sgd --model=resnet50 | tee -a "$RESULTS"
sleep 5
python3 "$FPATH/tf_cnn_benchmarks.py" --num_gpus=1 --batch_size=32 --variable_update=parameter_server --optimizer=sgd --model=resnet152 | tee -a"$RESULTS"
sleep 5
python "$FPATH/tf_cnn_benchmarks.py" --num_gpus=1 --batch_size=512 --variable_update=parameter_server --optimizer=sgd --model=alexnet | tee -a "$RESULTS"
sleep 5
python "$FPATH/tf_cnn_benchmarks.py" --num_gpus=1 --batch_size=32 --variable_update=parameter_server --optimizer=sgd --model=vgg16 | tee -a "$RESULTS"
sleep 5
python "$FPATH/tf_cnn_benchmarks.py" --num_gpus=1 --batch_size=64 --variable_update=parameter_server --optimizer=sgd --model=inception3 | tee -a "$RESULTS"

Proposed testing setup

Tensorflow 1..0+ (v1.11.0-rc2 latest)

CUDA 9.0+ (10.0 latest)

CuDNN 7.0+ (7.3 latest)

Results

Model Resnet50 Resnet152 Alexnet VGG16 InceptionV3
Batch Size 64 32 512 32 64
Images / Second

System

System component Spec
Operating System
GPU
CPU
RAM

Environment

Dependency Version / Tag / Commit
Benchmarks commit 33883211be3c3d44dd30fe3f54719c65326c52f4
Tensorflow tag v1.10.0+
CUDA 9.0+
CuDNN 7.0+
Additional dependencies
**Results**

|Model|Resnet50|Resnet152|Alexnet|VGG16|InceptionV3|
|:-|:-|:-|:-|:-|:-|
|Batch Size|64|32|512|32|64|
|Images / Second||||||

**System**

|System component|Spec|
|:-|:-|
|Operating System||
|GPU||
|CPU||
|RAM||

**Environment**

|Dependency|Version / Tag / Commit|
|:-|:-|
|Benchmarks|commit 33883211be3c3d44dd30fe3f54719c65326c52f4|
|Tensorflow|tag v1.10.0+|
|CUDA|9.0+|
|CuDNN|7.0+|
|Additional dependencies||



Source link
thanks you RSS link
( https://www.reddit.com/r//comments/9i3hy8/d__benchmarks_for_tensorflow_/)

LEAVE A REPLY

Please enter your comment!
Please enter your name here