{"id":1833,"date":"2026-06-28T07:28:57","date_gmt":"2026-06-28T07:28:57","guid":{"rendered":"https:\/\/rdp.in\/gpu-mart\/?post_type=product&#038;p=1833"},"modified":"2026-07-05T09:21:18","modified_gmt":"2026-07-05T09:21:18","slug":"draco-16-pb-parallel-fs-nvme-ai-storage","status":"publish","type":"product","link":"https:\/\/rdp.in\/gpu-mart\/product\/draco-16-pb-parallel-fs-nvme-ai-storage\/","title":{"rendered":"DRACO 16 PB Parallel-FS NVMe AI Storage"},"content":{"rendered":"<p>The DRACO 16 PB Parallel-FS NVMe AI Storage is RDP&#8217;s flagship parallel-filesystem NVMe storage system \u2014 built to feed the largest GPU superclusters and supercomputers during frontier AI training. It delivers 16 PB usable (NVMe) at 9.6 TB\/s read with GPUDirect Storage across a single clustered namespace, so datasets, checkpoints and weights stream to thousands of GPUs in parallel without the storage ever becoming the bottleneck \u2014 on-premises, in INR, on a GST invoice.<\/p>\n<p>Engineered for the most demanding AI\/HPC data pipelines, it presents one parallel namespace over a 400G fabric and scales capacity and bandwidth together by adding nodes \u2014 delivered racked, configured and validated as one system with one warranty and one support contract.<\/p>\n<h3>Key highlights<\/h3>\n<ul>\n<li><strong>16 PB usable (NVMe) \u00b7 9.6 TB\/s read<\/strong> \u2014 frontier-scale bandwidth sized to feed thousands of GPUs.<\/li>\n<li><strong>Parallel filesystem + GPUDirect Storage<\/strong> \u2014 a single namespace; data streams from NVMe straight to GPU memory across the supercluster.<\/li>\n<li><strong>Parallel FS (Lustre\/GPFS-class) + NFS\/S3, GPUDirect Storage<\/strong> \u2014 parallel and standard protocols so existing pipelines and schedulers just work.<\/li>\n<li><strong>1,536\u00d7 15.36 TB NVMe (64 nodes)<\/strong> \u2014 dense enterprise NVMe with end-to-end data integrity across 64 nodes.<\/li>\n<li><strong>256\u00d7 400G InfiniBand\/Ethernet<\/strong> \u2014 very high aggregate bandwidth to the GPU fabric; bandwidth grows with capacity.<\/li>\n<li><strong>640M random read<\/strong> \u2014 high aggregate random-read IOPS for metadata- and small-file-heavy datasets.<\/li>\n<li><strong>On-prem data sovereignty<\/strong> \u2014 datasets and weights stay in-house; DPDP-friendly, air-gappable.<\/li>\n<li><strong>Make-in-India OEM<\/strong> \u2014 predictable INR pricing, GST tax invoice (HSN 8471), pan-India onsite support, GeM-procurable.<\/li>\n<\/ul>\n<h3>Where it fits<\/h3>\n<ul>\n<li><strong>Frontier training data lake (primary):<\/strong> streams datasets to a GPU supercluster in parallel without starving any node.<\/li>\n<li><strong>Checkpoints &amp; weights:<\/strong> fast parallel write\/read of the largest checkpoints during frontier runs.<\/li>\n<li><strong>RAG &amp; vector stores:<\/strong> low-latency storage for the largest embeddings and indexes.<\/li>\n<li><strong>HPC scratch:<\/strong> high-throughput scratch for the largest simulations alongside AI.<\/li>\n<\/ul>\n<h3>How it works<\/h3>\n<p>A clustered parallel filesystem stripes data across 64 NVMe nodes and presents one namespace over 256\u00d7 400G InfiniBand\/Ethernet. With GPUDirect Storage, reads bypass the CPU and land directly in GPU memory across the supercluster. Capacity and bandwidth scale together as nodes are added. <em>Honest note: real throughput depends on dataset shape, file sizes and the client fabric \u2014 we validate it on your data, not just a peak number.<\/em><\/p>\n<h3>Industry use cases<\/h3>\n<ul>\n<li><strong>Government &amp; national labs<\/strong> \u2014 sovereign data lakes for national AI and HPC.<\/li>\n<li><strong>AI\/ML &amp; foundation-model teams<\/strong> \u2014 feed GPU superclusters at full speed during frontier training.<\/li>\n<li><strong>BFSI &amp; healthcare<\/strong> \u2014 high-throughput, data-resident storage under compliance.<\/li>\n<li><strong>Media &amp; design<\/strong> \u2014 fast scratch and asset storage for large generative pipelines.<\/li>\n<li><strong>Energy &amp; manufacturing<\/strong> \u2014 the largest simulation and sensor datasets for AI\/HPC.<\/li>\n<li><strong>Neocloud \/ AI providers<\/strong> \u2014 the high-bandwidth storage tier for a GPU cloud at scale.<\/li>\n<\/ul>\n<h3>Performance \u2014 and how to be sure<\/h3>\n<p>We don&#8217;t publish inflated peak numbers. The honest picture: 16 PB usable (NVMe) at 9.6 TB\/s read and 640M random read is sized to keep a GPU supercluster fed. <strong>Want certainty? Request a free benchmark with your datasets and training loop on this exact configuration before you buy<\/strong>; we&#8217;ll send back real sustained throughput, IOPS and GPU-utilisation under your workload.<\/p>\n<h3>Series &amp; scale path<\/h3>\n<ul>\n<li><strong>DRACO<\/strong> (flagship storage tier) \u2014 <em>this, the largest in the line<\/em>.<\/li>\n<li><strong>Tiering:<\/strong> this NVMe parallel-FS hot tier + an object capacity tier for durable datasets\/archives.<\/li>\n<li><strong>When to step up:<\/strong> add nodes for more capacity and bandwidth \u2014 talk to an architect about the tiering and fabric.<\/li>\n<\/ul>\n<h3>On-prem vs cloud \u2014 the TCO case<\/h3>\n<p>For frontier-scale AI data, owning beats renting: cloud egress and per-GB costs dominate against the largest, most-read datasets, and on-prem keeps data resident and the supercluster fed without network limits. RDP pricing is fixed in INR with a GST input-credit-eligible invoice \u2014 ask for a <strong>multi-year TCO comparison<\/strong> including egress savings.<\/p>\n<h3>Software &amp; integration<\/h3>\n<p>Integrates with your stack: Parallel FS (Lustre\/GPFS-class) + 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.<\/p>\n<h3>Power, cooling &amp; rack integration<\/h3>\n<p>A Multi-rack (64-node) liquid-cooled system with redundant PSUs \u2014 plan rack power, cooling and fabric. <em>(Exact power draw, BTU, flow and footprint figures confirmed on the build sheet.)<\/em> Full out-of-band management and drive hot-swap.<\/p>\n<h3>Deployment, warranty &amp; support<\/h3>\n<ul>\n<li><strong>Made to order<\/strong>, built, racked, cabled and burned-in in India; project timeline confirmed at quote.<\/li>\n<li><strong>In the box:<\/strong> storage nodes, drives, rails, cabling, quick-start, and the configured parallel filesystem.<\/li>\n<li><strong>Onsite warranty + AMC<\/strong> with pan-India coverage, drive-replacement SLA and an RMA\/escalation path <em>(exact terms confirmed at quote)<\/em>.<\/li>\n<\/ul>\n<h3>Why RDP<\/h3>\n<p>14 years of Make-in-India infrastructure and <strong>300,000+ devices shipped<\/strong>. Indian OEM, INR pricing, GST tax invoice (HSN 8471), pan-India onsite engineers, GeM availability, and DPDP \/ sovereign-AI-ready deployment.<\/p>\n<h3>Buy with confidence<\/h3>\n<p>This is RDP&#8217;s flagship AI storage system, made to order \u2014 <strong>talk to an RDP solution architect<\/strong>, size capacity, bandwidth and fabric for your supercluster, get a multi-year TCO, and <strong>benchmark your own datasets before you commit.<\/strong> Request a quote to begin.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>16 PB usable (NVMe) \u00b7 9.6 TB\/s read \u00b7 Parallel FS (Lustre\/GPFS-class) \u00b7 256\u00d7 400G InfiniBand\/Ethernet \u00b7 Multi-rack (64-node)<\/p>\n","protected":false},"featured_media":1942,"comment_status":"open","ping_status":"closed","template":"","meta":{"_yoast_wpseo_title":"","_yoast_wpseo_metadesc":"","rank_math_title":"DRACO 16 PB Parallel-FS NVMe AI Storage \u2014 9.6 TB\/s read, flagship parallel-FS, GPUDirect | RDP GPU Mart","rank_math_description":"RDP flagship on-prem AI storage \u2014 16 PB usable (NVMe), 9.6 TB\/s read, parallel filesystem, GPUDirect Storage. Feed a GPU supercluster during frontier training. Make-in-India, GST invoice. Request a quote.","_hermes_jsonld":""},"product_brand":[],"product_cat":[25],"product_tag":[],"class_list":["post-1833","product","type-product","status-publish","has-post-thumbnail","product_cat-storage-systems","pa_form-factor-multi-rack","pa_series-draco","first","instock","taxable","shipping-taxable","product-type-external"],"_links":{"self":[{"href":"https:\/\/rdp.in\/gpu-mart\/wp-json\/wp\/v2\/product\/1833","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/rdp.in\/gpu-mart\/wp-json\/wp\/v2\/product"}],"about":[{"href":"https:\/\/rdp.in\/gpu-mart\/wp-json\/wp\/v2\/types\/product"}],"replies":[{"embeddable":true,"href":"https:\/\/rdp.in\/gpu-mart\/wp-json\/wp\/v2\/comments?post=1833"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/rdp.in\/gpu-mart\/wp-json\/wp\/v2\/media\/1942"}],"wp:attachment":[{"href":"https:\/\/rdp.in\/gpu-mart\/wp-json\/wp\/v2\/media?parent=1833"}],"wp:term":[{"taxonomy":"product_brand","embeddable":true,"href":"https:\/\/rdp.in\/gpu-mart\/wp-json\/wp\/v2\/product_brand?post=1833"},{"taxonomy":"product_cat","embeddable":true,"href":"https:\/\/rdp.in\/gpu-mart\/wp-json\/wp\/v2\/product_cat?post=1833"},{"taxonomy":"product_tag","embeddable":true,"href":"https:\/\/rdp.in\/gpu-mart\/wp-json\/wp\/v2\/product_tag?post=1833"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}