Wave Computing Announces Early Access Program For Industry’s Fastest Dataflow Solution for Machine Learning Acceleration New Wave Compute Appliance Enables Up To 1000x Performance For Neural Network Training
Wave Computing, the Silicon Valley start-up that is revolutionizing machine learning, today announced an Early Access Program for its compute appliance, the industry’s fastest dataflow solution for accelerating machine learning training and inferencing. The Wave system overcomes the performance and scalability limitations of traditional machine learning approaches that rely on CPUs or co-processors, such as GPUs and FPGAs. By natively executing dataflow graphs, Wave’s compute appliance speeds the training of neural networks up to 1,000x and enables support for much larger datasets in a single data center node. Wave’s Early Access Program will provide a select number of data scientists the ability to work with Wave compute appliance prototypes before system sales begin in Q4 2017. Companies interested in applying for the Early Access Program can apply today.
Companies in data-driven industries - including retail, social media, entertainment, gaming, financial services and automotive - are using machine learning to redefine the way they do business and enable new revenue streams. However, while datasets are increasing in size and complexity, data scientists are under pressure to reduce the time needed to train and deploy neural networks to increase productivity and speed time to revenue. One leading U.S. car manufacturer recently reported its customers consumed 4.22 Petabytes of data in 2016, and according to Datafloq, self-driving cars will generate 2 Petabytes of data per car per year by 2020. Using a Wave compute appliance, training a neural network using this amount of data can be reduced from weeks to days while smaller networks can be reduced from hours to minutes.
“This is an exciting time for the machine learning industry as big data and analytics offer insight-driven enterprises the ability to change how they do business,” said Derek Meyer, CEO of Wave Computing. “Our new Wave compute appliance offers a breakthrough for companies wanting to speed their development and deployment of machine learning applications. By providing data scientists with early access to our dataflow solution, they will have the necessary performance to train neural networks in times never before imaginable.
About Wave's Compute Appliance
The Wave compute appliance is a “plug & play” machine learning solution that easily fits into existing data center environments. Employing a native dataflow architecture, Wave’s solution eliminates the need for a CPU or co-processor, thereby greatly improving the speed and scalability of machine learning training. Initially supporting TensorFlow, Wave’s dataflow software is framework agnostic; future releases are planned to support Microsoft’s Cognitive Toolkit (CNTK), MXNet and more. Additional supporting dataflow software includes tools and agent libraries with no need to learn new programming languages or reprogram to new, proprietary APIs.
A single Wave compute appliance can deliver up to 2.9 PetaOps per second of performance leveraging 256,000 interconnected Processing Elements (PEs), more than 2 TeraBytes of bulk memory and high-speed HMC memory, as well as up to 32 TeraBytes of storage. Up to four Wave compute appliances can be combined within a single node in the data center.
Availability
The Wave compute appliances will be available for purchase in Q4 2017. Companies interested in participating in Wave Computing’s Early Access Program can apply today.
About Wave Computing
Wave Computing is a VC-backed startup that is revolutionizing the machine learning industry. After years of dataflow architecture, hardware, software and tools development, the company’s world-class team has developed its patented, native dataflow solution that outperforms any other machine learning training product available today. Based in Campbell, California, Wave Computing is providing its solutions to customers globally.