DRACO 256× GH200 AI Supercomputer

Overview

The DRACO 256× GH200 AI Supercomputer is a turnkey, liquid-cooled HPC + AI supercomputer built on 256 NVIDIA GH200 Grace-Hopper accelerators, delivering ~37 TB HBM3e of aggregate accelerator memory across a multi-rack system with a non-blocking InfiniBand spine. It is engineered for organisations that need both traditional HPC (simulation, FP64) and frontier AI on one machine — on-premises, in INR, on a GST invoice.

Delivered as a single engagement, RDP designs the reference architecture, integrates, cools and burns it in, and hands over one validated supercomputer with one warranty and one support contract — removing multi-vendor integration risk. Its defining strength: tightly-coupled Grace CPU + Hopper GPU with coherent memory for HPC + AI.

Key highlights

• 256× GH200 · ~37 TB HBM3e aggregate — accelerator memory for large simulations and frontier AI.
• NVLink-C2C + non-blocking InfiniBand — NVLink-C2C coherent CPU-GPU links and a non-blocking InfiniBand spine between superchips.
• HPC + AI convergence — strong FP64/HPC throughput alongside mixed-precision AI on one system.
• 256× Grace (ARM, 1 per superchip) + Grace LPDDR5X coherent memory — CPU and memory matched to 256 accelerators.
• 4 PB parallel NVMe parallel filesystem — high-throughput storage for datasets, checkpoints and simulation output.
• Multi-Rack, liquid-cooled, turnkey — delivered, integrated and validated; one engagement, one warranty.
• On-prem data sovereignty — data and models stay in-house; DPDP-friendly, air-gappable.
• Make-in-India OEM — predictable INR pricing, GST tax invoice (HSN 8471), pan-India onsite support, GeM-procurable.

AI workload fit (what it actually runs — honestly)

• HPC + simulation: FP64 scientific computing (CFD, weather, molecular dynamics, finite-element) alongside AI.
• Frontier AI training: distributed training of large models across the 256 accelerators with 3D parallelism.
• Large-scale inference & fine-tuning: serve and fine-tune many large models in parallel.
• Engineering note: this is an NVIDIA Grace-Hopper system on the CUDA software stack. At this scale the network and parallelism strategy decide real performance; we run a scaling test on your codes and models rather than quoting a peak number.

AI workload positioning

This sits at the HPC + foundation-model convergence: a supercomputer that runs both scientific simulation and frontier AI. With ~37 TB HBM3e of accelerator memory on a non-blocking fabric, it is sized to sustain mixed HPC/AI workloads on-prem — the sovereign, owned alternative to renting hyperscale capacity.

Industry use cases

• Government & national labs — a sovereign HPC + AI supercomputer on GeM-procurable infrastructure.
• Weather, climate & energy — large FP64 simulation plus AI surrogates.
• Pharma & life sciences — molecular dynamics, drug discovery and AI in one system.
• Aerospace, automotive & manufacturing — CFD/FEA simulation plus AI design.
• Academia & research — a shared institutional supercomputer for HPC and AI.
• Neocloud / HPC providers — a converged HPC+AI service platform.

Performance — and how to be sure

We don’t publish inflated peak numbers. The honest picture: 256 GH200 accelerators (~37 TB HBM3e) on a non-blocking fabric are sized for converged HPC + AI at scale. Want certainty? Request a free benchmark — your HPC codes and AI models, including a scaling test — on a representative configuration before you commit; we’ll send back real performance and scaling efficiency.

Series & upgrade path

• DRACO (flagship supercomputer tier) — this.
• Scale ladder: from a few hundred accelerators to exascale-class systems (thousands of accelerators).
• Architecture options: NVIDIA Grace-Hopper, AMD Instinct MI300X/MI300A — chosen to fit your codes and software stack.
• When to step up: extend to a larger data hall — co-design the fabric, power and facility with an RDP architect.

On-prem vs cloud — the TCO case

At supercomputer scale, owning is a strategic and economic decision: an always-on hyperscale cloud system dominates any budget over a multi-year horizon, while on-prem removes egress and keeps sovereign data and codes in-house. RDP pricing is fixed in INR with a GST input-credit-eligible invoice — ask for a multi-year TCO and financing model.

Software & day-one readiness

Ships pre-integrated and validated: NVIDIA driver, CUDA, cuDNN, NCCL, the InfiniBand stack, Slurm and/or Kubernetes, with PyTorch and vLLM / Triton / TensorRT-LLM on Ubuntu LTS, with monitoring and a scheduler. Optional managed operations and an HPC/MLOps platform.

Power, cooling & rack integration

A multi-rack liquid-cooled supercomputer with high power density — RDP scopes facility power, CDU/manifold and water, the InfiniBand spine, floor layout and redundancy in the reference-architecture design. (Exact power, BTU, flow and facility figures confirmed in the design package.) Full out-of-band management.

Deployment, warranty & support

• Made to order, integrated, racked, cabled, cooled and burned-in in India; project timeline confirmed at quote.
• Delivered as one system: nodes, leaf/spine switches, cabling, PDUs, cooling, parallel storage, and the full HPC/AI software stack.
• Onsite warranty + AMC with pan-India coverage, system-level SLAs and an escalation path (exact terms confirmed at quote).

Why RDP

14 years of Make-in-India infrastructure and 300,000+ devices shipped. Indian OEM, INR pricing, GST tax invoice (HSN 8471), pan-India onsite engineers, GeM availability, and DPDP / sovereign-AI-ready deployment.

Buy with confidence

This is a turnkey HPC + AI supercomputer, made to order — talk to an RDP solution architect, co-design the architecture for your codes and models, get a multi-year TCO, and run a benchmark before you commit. Request a quote to begin.

Specifications