ATOM™-Max Server

High Performing Server for Large-Scale AI Inference

Contact Sales

Large-Scale AI Inference
Starts with a Single Server

ATOM™-Max Server is a power-efficient, single-server solution built for large-scale AI inference. It supports up to 8 ATOM™-Max PCIe cards, enabling deployment of hundreds of AI models spanning vision, LLMs, multimodal, and even physical AI workloads. Fully compatible with leading inference frameworks like vLLM, Triton, and Kubernetes, you can seamlessly transition from GPU workflows with familiar tools and guided tutorials.

See Tutorials

1,024 TFLOPS (FP16)

Peak Performance
Maximum FP16 Performance

512GB, 1TB/s

GDDR6 Memory
High-Capacity, High-Bandwidth Memory

~4.4kW

Power Consumption
Built for Optimal Energy Use

4U

Form Factor
Optimized for Data Centers

Compatible Software

OS

Ubuntu, Rhel 9, AlmaLinux, Rocky Linux

Frameworks & Tools

Hugging Face, PyTorch, TensorFlow, Triton

Inference Serving

VLLM, Triton Inference Server, TorchServe

Orchestration

Docker, OpenStack, Kubernetes, Ray

Performance at

Any Scale

Even under heavy demand, the ATOM™-Max delivers stable, high-throughput performance—generating thousands of tokens and processing image frames per second, all from a single system.

Sustainable

AI Infrastructure

ATOM™-Max delivers maximum AI inference performance within limited server room power budgets. Its exceptional power efficiency significantly lowers total cost of ownership (TCO) and enables a more sustainable AI infrastructure.

Full-stack
Software Support

ATOM™-Max is compatible with popular open-source ecosystems, supporting efficient serving, flexible resource management, and monitoring through tools like vLLM, Triton, Kubernetes, and Prometheus—so you can build full end-to-end services with ease.

RBLN SDK Docs

Variety of

Models and Applications

Run hundreds of AI models out of the box—from LLMs and vision AI to multimodal and physical AI. Build tailored services like chatbots, search, summarization, smart CCTV, and image generation.

Develop

As You Always Have

No need to abandon your existing development environment. Start right away with familiar workflows (PyTorch, TensorFlow, etc.) and step-by-step tutorials.

Applications

Enterprise

Streamline enterprise-wide AI adoption—from development to deployment—with scalable AI infrastructure

Construction

Enable proactive safety monitoring on construction sites with AI-powered surveillance

Healthcare

Support the AI healthcare ecosystem, from personalized wellness to precision medicine

Finance

Build next-gen financial services with secure, real-time AI processing of financial data

Manufacturing

Boost manufacturing productivity with Physical AI-powered smart factories

Telecom

Power advanced telecom services and elevate customer experience with reliable large-scale AI operations.

RBLN SDK
Deploy with Confidence from Day One.

Purpose-built for PyTorch.
Tuned for production.

High-QPS vLLM serving.
Ready out of the box.

Full Triton access with dev tools
you’ll actually use.

One-click deployment.
Zero guesswork.

Driver SDK

Core system software and essential tools for NPU excution

Firmware Kernel Driver User Model Driver System Management Tool

NPU SDK

Developer tools for model and service development

Compiler, Runtime, Profiler Hugging Face Leading Serving Frameworks (vLLM, TorchServe, Triton Inference Server, etc.)

Model Zoo

300+ ready-to-use PyTorch and TensorFlow models optimized for Rebellions NPUs

Natural Language Processing Generative AI Speech Processing Computer Vision