AI is entering a new phase, with teams moving from single-model experiments to large-scale production AI services. Vultr and AMD are addressing that demand by combining the AMD Helios™ rackscale solution, powered by AMD Instinct™ MI455X GPUs, with Vultr’s composable cloud infrastructure.
With the AMD Helios rackscale solution deployed on Vultr, customers can access high-memory, rackscale-accelerated compute for training, inference, fine-tuning, and agentic AI, providing a more open path to scale demanding workloads.
AMD Helios™ rackscale solution: Delivering high-memory, rack-scale AI
The AMD Helios rackscale solution delivers AI builders a silicon, rack-scale design, scale-up and scale-out networking, software, and security foundation. Featuring 72 AMD Instinct MI455X GPUs, built on next-generation AMD CDNA™ architecture, the solution delivers performance and efficiency for demanding AI training, inference, and agentic workloads.
The integrated infrastructure foundation also includes:
- 18 AMD EPYC™ CPUs
- High memory bandwidth: Up to 1.67 PB/s per rack
- Expanded HBM4 memory capacity: Up to 31 TB per rack
- Scale-up: 260 TB/s UALink™ over Ethernet
- Open scale-out: 43 TB/s UEC-aligned Ethernet
- AMD Pensando™ networking designed for lossless, high-throughput AI clusters
- Enhanced AI training, inference optimization, and exceptional throughput with ROCm™ software
- Direct liquid cooling (DLC) options for greater thermal management, rack-level performance, and energy efficiency
With the AMD Helios rackscale solution, production AI teams access scalable GPU capacity, high memory density, low-latency communication, open software support, isolation for multi-tenant workloads, repeatable deployments, and advanced support for low-precision AI datatypes.
It’s a streamlined path to scaling AI with utmost control.
Deploy on Vultr for easy access, flexibility, global reach, and more
Vultr provides a path to AMD-powered AI infrastructure in the cloud, offering more reliable access, deployment flexibility (from bare metal to virtualized instances), cloud-native orchestration, global reach (with 33 cloud data center regions across six continents), customer control, availability, and security.
Deployment models, including the Vultr Console and API, combine the AMD Helios rackscale solution with Vultr’s easy-to-use composable cloud infrastructure and open ecosystem. Where validated, customers can deploy Kubernetes-oriented GPU workloads on the free Vultr Kubernetes Engine (VKE) control plane for repeatable deployment and scaling patterns, simplifying the operational model for cloud-native AI.
Teams can also leverage the self-service Vultr Clusters to create, modify, and manage clusters of AMD Instinct MI455X GPUs on demand. A Slurm or Kubernetes workload scheduler simplifies day-to-day management, while an observability suite enables greater visibility into GPU performance and health. Vultr Clusters, Vultr GPU-enabled images, and AMD ROCm enable teams to move from planning to validation faster while maintaining control and visibility over their AI workflows.
AI builders increasingly need capacity near users while meeting local or regional data sovereignty requirements. Vultr’s global, composable cloud platform enables distributed deployments aligned with varying AI strategies, reducing deployment friction and bringing AMD-powered infrastructure closer to where teams build, serve, and govern AI.
Also featuring storage integration and customer support, Vultr provides a clear path to implementation for large-scale AI and agentic workloads.
Vultr and AMD Helios rackscale solution: A collaboration built for AI’s next era
AI experimentation has given way to long-term investment, and agentic AI is front and center. IDC forecasts a tenfold increase in the number and complexity of enterprise agents over the next five years. Successful workloads will depend on efficiency, openness, and flexibility – requiring a new approach to infrastructure.
More than ever before, teams need to reduce friction and risk to improve performance, control, flexibility, and operational efficiency.
The AMD Helios rackscale solution, powered by AMD Instinct MI455X GPUs, with Vultr’s composable cloud infrastructure, is how. It’s a combination built for the current state of enterprise AI adoption – as well as the next wave of technological breakthroughs.
Reserve capacity
Want to make sure you get access to AMD Instinct MI455X GPUs and the AMD Helios rackscale solution? We’re taking pre-orders for reserved capacity for deployments from 2027 through 2028. Contact our sales team to explore.