Vultr offers affordable access to Nvidia GPUs

Cloud services provider Vultr has launched what it claims is the first GPU virtualization platform for smaller and midsize companies that don’t need the much more powerful and much more expensive options offered by the big cloud players.

When Nvidia introduced its Ampere A100 processor in 2020, it emphasized that it was the first graphics processor to support Multi-Instance GPU, or MIG. This allows for partitioning the GPU into seven virtual GPUs, in much the same way a hypervisor partitions CPU cores.

Now Vultr says it’s the first cloud provider to offer fractional A100 GPU instances to customers through its Vultr Talon platform. The company notes there’s no one size fits all when it comes to customer workloads. Other cloud services providers that offer GPU instances make the full GPU available for a hefty price. Talon is a much smaller instance with a much lower price for customers who just need a snack, not a seven-course meal.

The high cost of GPU instances is often justifiable for the largest enterprise workloads, especially if they require multiple GPUs running in parallel. But many businesses and developers may want to start small and get their feet wet with AI, and the cost of even a single GPU can be prohibitive to getting started and experimenting.

“If you were to try to access a full single card or an eight-card system from someone like AWS, you’re spending at least a couple of thousand a month, and that’s outside the range of budget for a lot of companies,” noted J.J. Kardwell, CEO of Vultr. “There are many workloads in AI in ML that do not require a full card worth of resources,” Kardwell added.

For many researchers and developers, much of their work is testing and iterating, and their usage is very inconsistent. They may test on smaller scale datasets, and then over time, scale it up, Kardwell said. But the big cloud providers don’t offer small bites of GPUs.

In addition to the hardware, Vultr offers the full Nvidia AI enterprise software stack of tools, libraries, and frameworks and the adjacent technology that Nvidia has developed to help users get the most out of the technology. While other providers that offer GPU instances include their own GPU tools, it doesn’t make sense to reinvent the wheel when Nvidia has already done it, Kardwell said.

“Nvidia has built the best-in-class software stack for getting the most out of the GPU hardware. And you’re talking about users being able to access these at a really accessible price point, and to get both the best GPU hardware in the world and the purpose-built optimized software stack,” said Kardwell.

Vultr grew under the radar

If you’ve never heard of Vultr, join the club. It has been around since 2014, but it maintained a very low profile, which you might think is contrary to its interests. Vultr has taken zero venture dollars and operated without a sales or marketing team only until recently. However, getting by on word of mouth alone, it has grown organically to achieve an annualized run rate of more than $125 million, and it has 25 locations around the world.

Vultr offers the standard portfolio of services including cloud computing, cloud storage, and bare metal. Its primary differentiator is cost, and it targets smaller customers.

“We tend to be dramatically less expensive than the hyperscalers, [and we’re] able to meet the fundamental needs of the vast majority of users,” said Kardwell. “The hyperscalers really are focused on meeting the needs of the largest enterprises with the biggest budgets in the world, and the rest of the companies and developers around the world are, frankly, underserved by big tech clouds.”

Kardwell said Vultr’s regular services are somewhere between 30% to 50% less expensive than those of AWS, and for bandwidth-intensive users, it is 1/15th of the cost compared to the top cloud service providers. It achieves this through automation and efficiency, he said.

Initial availability of Talon will be in the company’s New Jersey location before rolling out globally in the coming weeks. The company plans to add more graphics-oriented high-end GPUs for different use cases such as virtual desktops and graphics processing in the coming months.