DRACO 64× GB300 NVL72 AI SuperCluster

Overview

The DRACO 64× GB300 NVL72 AI SuperCluster is RDP’s flagship sovereign AI factory — a turnkey, liquid-cooled multi-rack data hall of 4608 Grace-Blackwell Ultra GPUs across 64 unified NVLink-domain racks, joined by a non-blocking spine-leaf InfiniBand fabric, delivering ~1.3 PB HBM3e of aggregate GPU memory. It is the largest system RDP builds, sized to train the largest foundation models a nation or enterprise will run — on-premises, in INR, on a GST invoice.

Engineered for sovereign-AI programmes and national-scale operators, it is delivered as a single engagement: RDP designs the reference architecture, integrates and burns it in, and hands over one validated AI factory with one warranty and one support contract — removing multi-vendor integration risk at the largest scale.

Key highlights

• 64× GB300 NVL72 · ~1.3 PB HBM3e aggregate — 4608 Grace-Blackwell Ultra GPUs, RDP’s largest single-engagement AI system.
• 64 unified NVLink domains + non-blocking InfiniBand spine — each NVL72 rack is one 72-GPU accelerator; 64 racks scale over a full-bisection fabric.
• 2,304× NVIDIA Grace (ARM) (coherent) + Grace LPDDR5X coherent memory — Grace CPUs coherently attached to the Blackwell Ultra GPUs.
• 64 PB parallel NVMe parallel filesystem — data-hall-scale training data and checkpoint storage.
• Multi-Rack data hall, liquid-cooled, turnkey — delivered, integrated and validated; one engagement, one warranty.
• On-prem data sovereignty — training data and weights 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.
• Scale path — extend the data hall with additional SuperPODs on the same fabric.

AI workload fit (what it actually runs — honestly)

• Frontier training: distributed pre-training of the largest foundation models across 4608 GPUs with 3D parallelism.
• National-scale fine-tuning & serving: fine-tune and serve many large models in parallel for an entire organisation or nation.
• RAG, multimodal & agentic platforms: the most demanding production AI platforms at sovereign scale.
• Engineering note: at this scale the data hall is the computer — 64 unified NVLink domains (72 Blackwell Ultra GPUs each, 288 GB per GPU) over a non-blocking InfiniBand spine. We co-design and validate real scaling for your workload; we do not quote a peak FLOPS number as a substitute for an engineered architecture.

AI workload positioning

This is the top of RDP’s range: a complete, sovereign foundation-model factory. With ~1.3 PB HBM3e of GPU memory on a non-blocking fabric, it is sized to sustain the largest training runs and nation-scale serving on-prem — the owned, sovereign alternative to a hyperscale cloud region.

Industry use cases

• Government & sovereign AI — a national foundation-model factory on GeM-procurable infrastructure.
• Neocloud / hyperscale operators — a flagship GPU-cloud region built and validated end-to-end.
• Large conglomerates & BFSI — private frontier-model programmes under data-residency rules.
• Healthcare & life sciences — national research and imaging programmes, data in-house.
• National labs & defence — a sovereign frontier AI supercomputer.
• Telecom & public sector — nation-scale foundation-model platforms.

Performance — and how to be sure

We don’t publish inflated peak numbers. The honest picture: 4608 Grace-Blackwell Ultra GPUs (~1.3 PB HBM3e) on a non-blocking fabric are sized for the largest foundation-model training and nation-scale serving. Want certainty? Request a free scaling benchmark of your model and dataset on a representative configuration before you commit; we’ll send back real tokens/sec, scaling efficiency and projected time-to-train.

Series & upgrade path

• DRACO (flagship supercluster tier) — this, the top of the range.
• Scale ladder: from single NVL72 racks and SXM pods up to this multi-rack GB300 data hall.
• When to step up: add SuperPODs to the data hall — co-design the multi-pod fabric, power and facility with an RDP architect.

On-prem vs cloud — the TCO case

At sovereign data-hall scale, owning is a strategic national decision: an always-on hyperscale cloud region dominates any budget over a multi-year horizon, while on-prem keeps sovereign data, weights and capability in-country. 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, container registry, PyTorch and vLLM / Triton / TensorRT-LLM on Ubuntu LTS, with monitoring and a scheduler. Optional managed operations and a full MLOps platform.

Power, cooling & rack integration

A liquid-cooled multi-rack data hall with the highest power density in the range — RDP co-designs facility power, CDU/manifold and water, the InfiniBand spine, floor layout and redundancy. (Exact power, BTU, flow and facility figures confirmed in the design package.) Full out-of-band management across the data hall.

Deployment, warranty & support

• Made to order, integrated, racked, cabled, cooled and burned-in in India; project timeline confirmed at quote.
• Delivered as one engagement: NVL72 racks, leaf/spine fabric, cabling, PDUs, cooling, parallel storage, and the full cluster software stack.
• Onsite warranty + AMC with pan-India coverage, data-hall SLAs and a dedicated 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 RDP’s flagship sovereign AI factory, made to order — talk to an RDP solution architect, co-design the reference architecture, fabric and data hall, get a multi-year TCO and financing plan, and run a scaling benchmark before you commit. Request a quote to begin.

Specifications