Inspired by NVIDIA's DIGITS DevBox at GTC 2015, Microway is pleased to announce a new WhisperStation - Deep Learning Preconfigured System. Our configuration closely mirrors the switched PCI-E architecture that is favorable for Deep Learning applications.
It has been designed to provide maximum performance when training deep neural networks. It is the perfect system for a researcher exploring the applications of machine learning before such applications move into production. Using NVIDIA's DIGITS interface for popular deep learning tools, such as Caffe, researchers can collaborate and share GPU resources as they fine-tune their neural networks.
Struggling with the definition of Deep Learning? So is machine learning luminary Yann LeCun of Facebook. A recent interview explores this, how machine learning algorithims are developed, and the line between intuitive insight and empirical modeling in this growing field.
Yann also explains the work of the Deep Learning group at Facebook. If you've ever been impressed by the automatic tagging suggestions in Facebook, you've seen the refined products of this type of research.
We're pleased to announce the availability of our new 4-socket server with Xeon E7-4800v3/8800v3 CPUs (Haswell-EX). These platforms are exellent small-cluster replacements.
CUDA 7.0 brings with it C++11 features, a brand-new Thrust 1.8 algorithim library with faster sort, merge, reduce, and scan support, a new cuSOLVER linear solver library, and an improved cuFFT library. Individual feature performance gains range from 15-300% in CUDA 7.0. A new runtime compilation feature is also previewed for future production release–enabling developers to create highly architecture dependent yet flexible code. There are a number of beneficial changes to the installers (full runfile installers and full local package installers now available), and NVIDIA has also delivered much broader OS compatibility and support for POWER8 CPUs.
If you missed GTC 2015, we recommend watching the CUDA 7 Features and Overview session. As always, we strongly suggest you plan upgrades carefully and proceed cautiously with any CUDA updates. Especially when migrating from much older versions, upgrades can go and have gone wrong. Please inquire with Tech Support before you execute an update if you are unsure of what you may encounter along the way.
Intel has launched new Xeon E5-4600v3 CPUs that extend the performance enhancements of the Xeon E5-2600v3 "Haswell-EP" platform, including new instructions, more cores, and better memory bandwidth, to the 4-socket platform. Our review explores these advantages and helps you evaluate the performance benefits of a new system based upon this platform.Read the full post on Xeon E5-4600v3 CPUs
Our blog's series on how to run GPU-accelerated molecular dynamics applications continues with two posts. First, our guide on running HOOMD-blue shows how two K40s deliver up to 13x performance gains over CPUs alone. If you're a fan of python and interested in running molecular dynamics simulations, HOOMD-blue is a flexible and powerful tool that scales up to thousands of GPUs. As in our previous guides, we show you how to both run the default benchmarks and how to test your own code.
Similarly, we've posted instructions on how to run GROMACS on a GPU cluster. If you're interested in seeing how well your code runs on GPUs, sign up for Test Drive access on our cluster. We have a variety of applications, libraries, and compilers installed and ready for your work.
We've collected a number of the latest HPC technical resources for your perusal:
Eliot Eshelman at 508-732-5534
Ed Hinkel at 508-732-5523
Sam Wheeler at 508-732-5526
GSA Schedule GS-35F-0431N