Deep Learning Blog | Bitfusion

Bitfusion Scientific Computing AMI 2016.08

Written by maciej | Aug 23, 2016 3:05:31 PM

Last week we released an update for our popular Bitfusion Ubuntu 14 Scientific Computing AMI which upgrades all of the commonly used applications for scientific and statistical computing to the most recent versions, including:

  • Python 2 / Python 3
  • Numpy 1.11.1 / SciPy 0.18.0 / SciKit 0.17.1, Pandas 0.18.1., SymPy 1.0, and Matplotlib 1.5.1
  • Julia 0.4.6
  • Octave 4.4.1
  • R 3.3.1 / RStudio Server 0.99.9
  • Jupyter 4.1.0
  • OpenBlas 0.2.18

As a new feature, we added Jupyter which is by default available on port 8888 and comes pre-installed with kernels for Python 2, Python 3, Julia, Octave,  and R. Of course, if you prefer RStudio Server for R development you can access it on Port 8787 as usual. Both of these development environments are secured by a default password that is set to the AWS instance-id, which can be easily changed as documented here. We've also included Nvidia drivers and the Cuda Toolkit on this AMI should you want to run these applications on a GPU instance and install GPU enabled libraries.

Finally, we pre-compiled and pre-configured the applications to take advantage of the OpenBLAS library to give you optimal performance across all the various EC2 instances on AWS. Why is this important? A non-optimized  BLAS library or an improperly configured BLAS library can deliver poor performance, even on a powerful processor - resulting in long run-times and wasted money.

The graphs below illustrates this fact by showing results on an m4.4xlarge instance that features 16 vCPUs - 8 physical cores which are dual threaded. The first chart shows the result for the R Benchmark where R is using the optimized BLAS version, it shows a 2.5x run-time improvement over R using the default non-optimized version.

The second chart shows the results for an Octave Benchmark where Octave using the optimized BLAS version shows a 6.4x improvement over Octave using the default non-optimized version. Both benchmarks are discussed in more detailed on our documentation page for the AMI.

We're already looking forward to our future releases of this AMI. Some features you can expect are the following:

  • Automatic switching to GPU acceleration when appropriate on GPU equipped instances
  • Integration of our Bitfusion Boost technology for seamless cluster computing across GPU instances.

Questions or comments? Please post them in the comment section below or join our community Bitfusion-AWS Slack Channel.

Get Started!