Rackspace Brings GPU-as-a-Service to Spot Cloud Platform
Rackspace Technology is adding a GPU-as-a-service offering to its open market for cloud servers to help meet the rapidly growing demand for compute power in the AI age.
The San Antonio, Texas-based company also is opening a new location in San Jose, California, for its Rackspace Spot service, adding to six others located around the world.
The new service and expanded reach are only the latest efforts by the cloud computing and managed hosting company to grow the capabilities of Rackspace Spot, an online auction where companies can bid on access to cloud servers. Rackspace launched the open-market auction in March, giving organizations a way to leverage unused server space at discounted prices.
At the time Rackspace Open launched, Lance Weaver, chief product and technology officer for Rackspace’s Private Cloud Business Unit, said the service was designed to offer access to powerful cloud infrastructure solutions to such organizations as small and medium enterprises, startups, digital companies, and developers without them having to shoulder a lot of upfront costs.
The infrastructure offerings via Rackspace Spot are fully managed Kubernetes clusters, which companies can bid on in hopes of grabbing compute capacity relatively cheaply.
“With high-availability Kubernetes clusters and market-based dynamic pricing, organizations can scale resources up or down as needed and obtain cloud infrastructure at the best possible price,” Weaver said in a statement. “The transparency of the auction process provides information to compare prices and choose the most cost-effective options.”
Pros and Cons of Spot Services
Spot cloud instances aren’t new. Most large cloud infrastructure service providers – including Amazon Web Services (AWS), Microsoft Azure and Google Cloud Platform – have been offering spot instances for years.
The attraction is clear. Organizations can tap into scalable cloud infrastructure at a savings of as much as 90% to run their power-hungry AI and data analytics workloads. That said, the cloud providers can reclaim the compute capacity with little notice, potentially interrupting the job, and availability is unpredictable.
Access to GPUs
With the GPU-as-a-service offering, Rackspace is now offering access to Nvidia GPUs, which are critical for running AI inferencing and training jobs. The rapid innovation and adoption of generative AI fueled huge demand for Nvidia GPUs that in turn led to global shortages that didn’t start to ease until several months ago.
One server offered with the new GPU service is powered by an in-demand Nvidia H100 GPU and an Intel 8568Y CPU with 48 hyperthreaded cores running at 2.3GHz to 4.0GHz. The GPU H100 Virtual Server v2.Mega Extra-Large system also includes 128GB of memory, multipath-enabled NVMe encrypted storage and 25 gigabyte network.
The other server, the GPU A30 Virtual Server v2. ++ Extra Large, comes with an A30 GPU from Nvidia, Intel’s 6526Y CPU with 24 hyperthreaded cores, and the same NVMe storage and 25 GiB network.
“Rackspace’s GPUaaS will give customers on-demand access to powerful accelerated resources optimized for AI, machine learning, data analytics and graphics rendering workloads,” Brian Lillie, president of Private Cloud for Rackspace, said in a statement.
The GPU systems are running in Rackspace’s new SJC3 data center in San Jose, which Rackspace executives said will offer faster access to the infrastructure, lower latency and greater scalability and is based on the latest OpenStack technology. OpenStack was created in 2010 in a collaboration between Rackspace and NASA and the cloud company reaffirmed its commitment to the technology with its One OpenStack strategy and the launch in August of the OpenStack Enterprise cloud solution.
