More Affordable Split Nvidia GPU Instances on Vultr

To meet the needs of developers who want to experiment with GPU acceleration, Vultr offers split instances of the Nvidia A100 for customers who don’t require the high price of power charges.

Cloud service provider Vultr said it was the first to bring a GPU virtualization platform to small and medium -sized businesses that didn’t require the powerful and more expensive options of large cloud players. In 2020, during the presentation of the Ampere A100 processor, Nvidia specified that this graphics processor was the first to support the Multi-instance GPU, or MIG, which made it possible to split the GPU into seven virtual GPUs, from the same method. hypervisor. partitions of CPU cores. Today, Vultr says it is the first cloud provider to offer split instances of the A100 GPU to its customers through its Vultr Talon platform. According to the vendor, there is no one-size-fits-all solution to meet the diverse workloads of the customer. Other cloud service providers that offer GPU instances only offer a full GPU, at a premium price. Vultr’s Talon instance size is smaller and lower priced, and meets the smaller needs of some customers.

The high cost of GPU instances is often justified when it comes to running the most important enterprise workloads (simulation, modeling, or even AI), especially if they require multiple GPUs running simultaneously. Many companies and developers love to delve into AI, but the cost of a GPU can be a barrier to getting started and experimenting. “An entire instance or eight-card system from a vendor like AWS costs several thousand dollars a month, which many companies can’t afford,” said JJ Kardwell, CEO of Vultr. “Multiple AI and ML workloads can be done without full sample resources,” Kardwell added. “Typically, a lot of the work of researchers and developers is tested and iterated, and the usage is very thorough. They can test on small datasets and then, over time, enlarge,” Kardwell explains. But large cloud providers do not offer split GPUs.

In addition to the hardware, Vultr offers the full Nvidia AI enterprise software stack, including tools, libraries and frameworks, and adjacent technology developed by the Santa Clara vendor to help users get the most out of the technology. “Other vendors offering GPU opportunities include their own GPU tools, but what’s the point of re-inventing the wheel when Nvidia has already done that?”, Mr. Kardwell estimates “Nvidia has created the best in-class software stack to make the most of GPU hardware. Users can access these products at truly affordable prices, and benefit from the world’s best GPU hardware and a software stack specially optimized for it, “said the CEO of Vultr.

Very discreet growth

Made in 2014, Vultr has so far remained very cautious, which may seem contrary to its interests. The provider did not raise any funds from a venture capitalist and until recently operated without a sales or marketing team. But, thanks to word-of-mouth alone, it has grown organically at a recurring annual rate of more than $ 125 million, and it has 25 locations around the world. Vultr offers a classic portfolio of services, including cloud computing, cloud storage, and bare metal. The value of its services is its primary asset, and it targets small customers. “We’re significantly cheaper than hyperscalers, and we can meet the basic needs of most users,” Kardwell said. “Hyperscalers are highly focused on the activities of the largest companies with massive budgets, but clearly other companies and developers around the world are very underserved by large tech clouds,” he added.

According to JJ Kardwell, standard Vultr services are 30% to 50% cheaper than AWS, and for bandwidth -hungry users, its price is 1/15e less than the best cloud service provider. “These results are achieved through automation and optimization,” he added. Talon will first be available at the company’s New Jersey location before launching worldwide in the coming weeks. In the coming months, the vendor also plans to add high-end, graphics-oriented GPUs to meet a variety of use cases, such as virtual desktops and graphics processing.

Leave a Comment