QUASAR 100 TB All-Flash NVMe AI Storage

Overview

The QUASAR 100 TB All-Flash NVMe AI Storage is an all-flash NVMe storage system built to keep GPUs fed during AI training and inference. It delivers 100 TB usable (NVMe) at 80 GB/s read with GPUDirect Storage, so training data, checkpoints and model weights stream to the GPUs without the storage becoming the bottleneck — on-premises, in INR, on a GST invoice.

Engineered for AI/ML data pipelines, it pairs dense NVMe with a parallel fs and high-speed networking, delivered racked, configured and validated as a single system with one warranty and one support contract.

Key highlights

• 100 TB usable (NVMe) · 80 GB/s read — high-bandwidth flash sized to feed GPU clusters during training.
• GPUDirect Storage — data streams from NVMe straight to GPU memory, bypassing the CPU bounce for maximum throughput.
• Parallel FS + NFS/S3, GPUDirect Storage — parallel/standard protocols so existing pipelines and frameworks just work.
• 24× 7.68 TB NVMe — dense, enterprise NVMe with end-to-end data integrity.
• 2× 200G InfiniBand/Ethernet — high-speed networking to the GPU fabric; no I/O wall.
• 5M random read — high random-read IOPS for many small files and metadata-heavy AI datasets.
• On-prem data sovereignty — datasets 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.

Where it fits

• Training data lake (primary): the high-bandwidth tier that streams datasets to GPU servers and clusters without starving them.
• Checkpoints & weights: fast write/read of large checkpoints during long training runs.
• RAG & vector stores: low-latency storage for embeddings, indexes and retrieval corpora.
• Inference assets: model weights and caches served at line rate to inference nodes.

How it works

An all-flash NVMe array exposes a Parallel FS + NFS/S3, GPUDirect Storage namespace over 2× 200G InfiniBand/Ethernet. With GPUDirect Storage, reads bypass the CPU and land directly in GPU memory, so the GPUs spend time computing, not waiting on I/O. The system scales by adding nodes; capacity and bandwidth grow together. Honest note: real throughput depends on dataset shape, file sizes and the client fabric — we validate it on your data rather than quoting only a peak number.

Industry use cases

• AI/ML platforms — feed GPU clusters with training data at full speed.
• Government & research — sovereign data lakes for national AI and HPC.
• BFSI & healthcare — high-throughput, data-resident storage under compliance.
• Media & design — fast scratch and asset storage for generative and rendering pipelines.
• Manufacturing & energy — sensor/simulation datasets for AI and HPC.
• Neocloud / AI providers — a high-bandwidth storage tier for a GPU cloud.

Performance — and how to be sure

We don’t publish inflated peak numbers. The honest picture: 100 TB usable (NVMe) at 80 GB/s read and 5M random read is sized to keep GPU clusters fed. Want certainty? Request a free benchmark with your datasets and training loop on this exact configuration before you buy; we’ll send back real sustained throughput, IOPS and GPU-utilisation under your workload.

Series & scale path

• QUASAR (performance storage tier) — this.
• Capacity ladder: 50 TB edge → 100/250 TB → PB-scale parallel-FS → object storage at exabyte scale.
• When to step up: grow capacity and bandwidth by adding nodes, or move to a PB-scale parallel filesystem — talk to an architect.

On-prem vs cloud — the TCO case

For AI data at scale, owning beats renting: cloud storage egress and per-GB costs add up fast against large, frequently-read datasets, and on-prem keeps data resident and the GPUs fed without network limits. RDP pricing is fixed in INR with a GST input-credit-eligible invoice — ask for a 3-year TCO comparison including egress savings.

Software & integration

Integrates with your stack: Parallel FS + NFS/S3, GPUDirect Storage, NVIDIA GPUDirect Storage, and standard NFS/S3 clients, with monitoring and quota/multi-tenant management. Works with PyTorch/TensorFlow data loaders, Slurm/Kubernetes and common MLOps tooling.

Power, cooling & rack integration

A 2U air-cooled system with redundant PSUs — specify rack power and cooling. (Exact power draw, BTU, airflow and rack-depth figures confirmed on the build sheet.) Full out-of-band management and drive hot-swap.

Deployment, warranty & support

• Made to order, built, racked and burned-in in India; lead time confirmed at quote.
• In the box: storage system, drives, rails, cabling, quick-start, and the configured filesystem.
• Onsite warranty + AMC with pan-India coverage, drive-replacement SLA and an RMA/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 an AI storage system, made to order — talk to an RDP solution architect, size capacity, bandwidth and fabric for your GPU cluster, get a 3-year TCO, and benchmark your own datasets before you commit. Request a quote to begin.

Specifications