UPDATED 20:54 EDT / MARCH 29 2018

CLOUD

Oracle puts GPU-powered bare-metal servers in the cloud

Oracle Corp. is expanding its bare-metal-computing-in-the-cloud offerings with a service that incorporates Nvidia Corp.’s Tesla graphics processing unit chips.

The platform, announced Tuesday, is intended primarily for organizations that need high-performance computing capacity for artificial intelligence, deep learning and machine learning projects.

Nvidia has been on a recent campaign to make its chips a standard for machine learning development both 0n-premises and in the cloud, and Oracle is the latest endorsement. Bare-metal services essentially give customers a dedicated physical server with pay-per-use pricing.

Unlike colocation arrangements, the customer doesn’t buy the hardware but rents it. Several other vendors offer bare-metal options, including Amazon Web Services Inc., IBM Corp. and Rackspace Inc., but Oracle claimed its service is easier to provision and cheaper. “Our ability to get an entire physical server on demand in minutes without a phone call is unique,” said Leo Leung, senior director of products and strategy for Oracle Cloud Infrastructure.

For customers that don’t need bare-metal performance, Oracle will release virtual machine support for Nvidia Tesla V100 chips within a few weeks, enabling customers to provision virtual machines with between one and four GPUs each.

Oracle launched its bare-metal offerings in 2016 and announced a version that uses the Nvidia Pascal GPUs last October. The new service uses Nvidia’s more recent Tesla V100 chip, which is based on the Volta architecture that was introduced last spring.

The processor packs 21 billion transistors into a single package about the size of an Apple Watch face and is claimed to be five times as fast as Pascal. The Oracle service combines eight Tesla V100 GPUs with Nvidia’s NVLink interconnect technology, which delivers up to 12 times the data transfer rate of the industry-standard PCI Express interface.

Oracle is pricing its offering below competitors’. An eight-GPU server with 768 gigabytes of memory and up to 512 terabytes of block storage is priced at $2.25 per CPU hour, compared with more than $3 for competitors, Leung said. The company also leverages a networking architecture that prevents contention, giving users effectively full bandwidth regardless of other resources on the network. “Once you’re inside the cloud, the network between resources aren’t oversubscribed,” he said.

One dividend of that dedicated networking infrastructure is that Oracle can guarantee service levels that others can’t, Leung said. “We’ll have performance guarantees for how storage is going to perform, and if performance falls outside the [service-level agreements], you can claim credits.”

As part of the Nvidia partnership, Oracle is making the Nvidia GPU Cloud software library available to its cloud customers. The company is also launching limited availability of Nvidia Grid, a virtual desktop service that delivers GPU-accelerated graphics performance via the Oracle cloud.

Image: Pixabay

A message from John Furrier, co-founder of SiliconANGLE:

Your vote of support is important to us and it helps us keep the content FREE.

One click below supports our mission to provide free, deep, and relevant content.  

Join our community on YouTube

Join the community that includes more than 15,000 #CubeAlumni experts, including Amazon.com CEO Andy Jassy, Dell Technologies founder and CEO Michael Dell, Intel CEO Pat Gelsinger, and many more luminaries and experts.

“TheCUBE is an important partner to the industry. You guys really are a part of our events and we really appreciate you coming and I know people appreciate the content you create as well” – Andy Jassy

THANK YOU