By: Michael Feldman

Oracle announced it is now offering NVIDIA’s P100 GPU instances in its public cloud, with plans to add the more powerful V100 GPUs in the near future.

This is the first time Oracle has offered access to GPU acceleration, reflecting an industry-wide move to provide access to cloud hardware optimized for artificial intelligence and machine learning. The new instances, known as the X7, can also be used for more traditional HPC application, like genomics analysis or physics simulations, but it’s pretty clear that Oracle believes its major opportunity with these GPU-accelerated servers is in the AI realm.

 In a blog post by Oracle’s Vinay Shivagange Chandrashekar, he writes that the new offering will enable bare metal access to servers that house two Tesla P100 GPUs, which together provide more than 21 peak teraflops of 32-bit floating point performance. They are also working with NVIDIA to build an eight-GPU configuration using the much more powerful Volta-generation V100 accelerators.  Besides the hardware, the company is also going to provide a pre-configured software stack with the new instances, which includes both the CUDA SDK and the cuDNN deep learning library.

“This is going to be a game changer for customers allowing them to essentially rent a supercomputer by the hour,” writes Chandrashekar.  He also notes that AI and deep learning are “quickly becoming one of the most important and fastest growing workloads.”

One of Oracle’s early customers for the new GPU cloud instance is Fluent.ai, which offers speech recognition technology to consumer electronics manufacturers who want to add this capability into their products. The company has a customized neural network that they claim can understand speech in a variety of languages and accents, and in less than ideal acoustic environments.

“Running on new NVIDIA GPU instances has significantly optimized the training of our deep learning models compared to the previous-generation hardware,” said Vikrant Tomar, chief technology officer at Fluent.ai. “This allows us to train more sophisticated speech recognition models while reducing the overall job time from weeks to days.”

The GPU instances will initially be available in Oracle’s US-East Ashburn region, with access in the EU-Germany Frankurt region coming soon.