Vultr named NVIDIA Exemplar Cloud after Blackwell tests
Vultr has been named an NVIDIA Exemplar Cloud following benchmark testing on NVIDIA Blackwell GPUs.
The recognition puts Vultr among the first cloud providers to meet performance standards based on NVIDIA reference designs for the Blackwell platform.
Testing was carried out on a 512-node NVIDIA HGX B200 cluster. The benchmarks covered 11 AI training models, including Llama 3.1, Grok-1, Qwen3, DeepSeek-v3-TorchTitan and NVIDIA Nemotron variants.
The Exemplar Cloud initiative has been meticulously devised to provide cloud-computing clientele with a standardised methodology for evaluating and comparing artificial intelligence infrastructure across a diverse range of service providers.
NVIDIA inaugurated this programme specifically to quantify operational performance against established industry benchmarks, particularly as organisations undergo the rigorous process of assessing potential suppliers for large-scale AI training requirements.
Fundamental discrepancies in cloud architecture can significantly influence both technical performance and the comprehensive total cost of ownership. As commercial enterprises and software developers continue to escalate their financial investment in sophisticated artificial intelligence systems, the necessity for benchmark transparency has become increasingly paramount.
This transparency is essential for identifying robust infrastructure capable of sustaining demanding, real-world workloads within an evolving technological landscape.
According to Vultr, the Blackwell testing covered models ranging from smaller configurations such as Llama 3.1 8B to much larger systems including Llama 3.1 405B, Grok-1 314B and DeepSeek-v3-TorchTitan 671B.
Vultr provides cloud infrastructure services including Cloud Compute, Cloud GPU, Bare Metal and Cloud Storage. It also offers Kubernetes support, GPU-enabled images, container repositories and model workflow tools for customers building AI applications.
Market context
This particular designation arrives at a juncture where cloud-service providers are competing intensely to demonstrate that their bespoke systems possess the capability to manage progressively substantial artificial intelligence training assignments with predictable performance outputs and financial expenditures.
For prospective purchasers, the implementation of standardised benchmarks has emerged as a primary mechanism for evaluating whether a specific platform can effectively accommodate large-scale models prior to the commitment of significant capital investment.
Vultr characterises itself as the pre-eminent privately held cloud infrastructure corporation globally and asserts that it provides comprehensive services to hundreds of thousands of active clients across a vast network spanning one hundred and eighty-five countries.
In financing announced in December 2024, Vultr said it was valued at USD $3.5 billion. The company was founded by David Aninowsky and says it was self-funded for more than a decade before that investment.
J.J. Kardwell, Chief Executive Officer of Vultr, commented on the benchmark result.
"Our NVIDIA Exemplar Cloud validation confirms Vultr's ability to deliver industry-leading performance for today's most demanding AI workloads," said J.J. Kardwell, Chief Executive Officer of Vultr.
"This recognition reflects the strength of our infrastructure design and our commitment to providing enterprises with reliable, production-ready AI compute," Kardwell said.