Skip to content

Latest commit

 

History

History
132 lines (114 loc) · 5.98 KB

README.04-benchmarks.md

File metadata and controls

132 lines (114 loc) · 5.98 KB
Raw

Performance Benchmarks

Caffe

Caffe is a deep learning framework made with expression, speed, and modularity in mind. It is developed by Berkeley AI Research (BAIR)/The Berkeley Vision and Learning Center (BVLC) and community contributors.

# install prerequisites
$ sudo apt-get install libprotobuf-dev libleveldb-dev libsnappy-dev libhdf5-serial-dev protobuf-compiler
$ sudo apt-get install --no-install-recommends libboost-all-dev
$ sudo apt-get install libgflags-dev libgoogle-glog-dev liblmdb-dev
# BLAS
$ sudo apt-get install libatlas-base-dev

# add caffe as subtree
$ git remote add caffe https://github.com/BVLC/caffe
$ git subtree add --prefix=benchmarks/caffe/ --squash caffe master
$ git push
# check caffe subtree
$ git fetch https://github.com/BVLC/caffe master
$ ./contrib/devtools/git-subtree-check.sh benchmarks/caffe
# sync caffe subtree
$ git remote add caffe-remote https://github.com/BVLC/caffe
$ git subtree pull --prefix=benchmarks/caffe/ --squash caffe-remote master

AlexNet

Benchmark AlexNet described in Google's ImageNet Classification with Deep Convolutional Neural Networks paper

$ caffe/build/tools/caffe time --model=caffe/models/bvlc_alexnet/deploy.prototxt -gpu 0 -iterations 200

GoogleNet

Benchmark GoogleNet described in Google's Going Deeper with Convolutions paper

$ caffe/build/tools/caffe time --model=caffe/models/bvlc_googlenet/deploy.prototxt -gpu 0 -iterations 200

Nheqminer

Equihash miner for NiceHash

# add nheqminer as subtree
$ git remote add nheqminer https://github.com/nicehash/nheqminer.git
$ git subtree add --prefix=benchmarks/nheqminer/ --squash nheqminer master
$ git push
# check nheqminer subtree
$ git fetch https://github.com/nicehash/nheqminer.git master
$ ./contrib/devtools/git-subtree-check.sh benchmarks/nheqminer
# sync nheqminer subtree
$ git remote add nheqminer-remote https://github.com/nicehash/nheqminer.git
$ git subtree pull --prefix=benchmarks/nheqminer/ --squash nheqminer-remote master

Periodic Task Releaser

This is a collection of CUDA programs intended to measure interference between GPU processes. It was created as part of ongoing research in real-time systems (see paper) at UNC Chapel Hill

# add periodictaskreleaser as subtree
$ git remote add periodictaskreleaser https://github.com/yalue/PeriodicTaskReleaser.git
$ git subtree add --prefix=benchmarks/periodictaskreleaser/ --squash periodictaskreleaser master
$ git push
# check periodictaskreleaser subtree
$ git fetch https://github.com/yalue/PeriodicTaskReleaser.git master
$ ./contrib/devtools/git-subtree-check.sh benchmarks/periodictaskreleaser
# sync periodictaskreleaser subtree
$ git remote add periodictaskreleaser-remote https://github.com/yalue/PeriodicTaskReleaser.git
$ git subtree pull --prefix=benchmarks/periodictaskreleaser/ --squash periodictaskreleaser-remote master

Commands below create graphs from periodictaskreleaser benchmarks

# install prerequisites
$ sudo apt install libfreetype6-dev
$ sudo apt install python-pip
$ sudo apt install python-gi-cairo
$ pip install matplotlib

# create performance test graphs
$ python generate_plots.py

CUDA Memory Experiments

Simple programs that run memory experiments on CUDA, created at UNC - Chapel Hill

# add cudamemoryexperiments as subtree
$ git remote add cudamemoryexperiments https://github.com/Sarahild/CudaMemoryExperiments.git
$ git subtree add --prefix=benchmarks/cudamemoryexperiments/ --squash cudamemoryexperiments master
$ git push
# check cudamemoryexperiments subtree
$ git fetch https://github.com/Sarahild/CudaMemoryExperiments.git master
$ ./contrib/devtools/git-subtree-check.sh benchmarks/cudamemoryexperiments
# sync cudamemoryexperiments subtree
$ git remote add cudamemoryexperiments-remote https://github.com/Sarahild/CudaMemoryExperiments.git
$ git subtree pull --prefix=benchmarks/cudamemoryexperiments/ --squash cudamemoryexperiments-remote master

Mixbench

A GPU benchmark tool for evaluating GPUs on mixed operational intensity kernels (CUDA, OpenCL, HIP)

# add mixbench as subtree
$ git remote add mixbench https://github.com/ekondis/mixbench.git
$ git subtree add --prefix=benchmarks/mixbench/ --squash mixbench master
$ git push
# check mixbench subtree
$ git fetch https://github.com/ekondis/mixbench.git master
$ ./contrib/devtools/git-subtree-check.sh benchmarks/mixbench
# sync mixbench subtree
$ git remote add mixbench-remote https://github.com/ekondis/mixbench.git
$ git subtree pull --prefix=benchmarks/mixbench/ --squash mixbench-remote master

Rodinia

The University of Virginia Rodinia Benchmark Suite is a collection of parallel programs which targets heterogeneous computing platforms with both multicore CPUs and GPUs.

# install prerequisites
$ sudo apt install libglew-dev

# add rodinia as subtree
$ git remote add rodinia https://github.com/yuhc/gpu-rodinia.git
$ git subtree add --prefix=benchmarks/rodinia/ --squash rodinia master
$ git push
# check rodinia subtree
$ git fetch https://github.com/yuhc/gpu-rodinia.git master
$ ./contrib/devtools/git-subtree-check.sh benchmarks/rodinia
# sync rodinia subtree
$ git remote add rodinia-remote https://github.com/yuhc/gpu-rodinia.git
$ git subtree pull --prefix=benchmarks/rodinia/ --squash rodinia-remote master

Please check nvpmodel and explore Max-Q, Max-P and Max-N power modes.