{"id":2578,"date":"2026-07-11T02:15:45","date_gmt":"2026-07-11T02:15:45","guid":{"rendered":"https:\/\/rdp.in\/gpu-mart\/product\/nvidia-gb300-nvl72-supercluster\/"},"modified":"2026-07-13T01:45:44","modified_gmt":"2026-07-13T01:45:44","slug":"nvidia-gb300-nvl72-supercluster","status":"publish","type":"product","link":"https:\/\/rdp.in\/gpu-mart\/product\/nvidia-gb300-nvl72-supercluster\/","title":{"rendered":"NVIDIA GB300 NVL72 Supercluster"},"content":{"rendered":"<p><strong>A complete NVIDIA GB300 NVL72 Supercluster in a single, turnkey containerised node.<\/strong> Eight liquid-cooled GB300 NVL72 racks arrive pre-integrated as one production-ready AI factory node \u2014 576 NVIDIA Blackwell Ultra B300 GPUs and 288 Grace CPUs operating as a single coherent accelerator, delivered to your site and commissioned as a plug-and-play unit rather than assembled on the floor over months.<\/p>\n<h3>Key highlights<\/h3>\n<ul>\n<li>8\u00d7 GB300 NVL72 racks in one containerised node \u2014 <strong>576 Blackwell Ultra B300 GPUs + 288 Grace CPUs<\/strong> as one system.<\/li>\n<li><strong>~165.6 TB HBM3e<\/strong> plus <strong>~320 TB<\/strong> fast system memory as node-level unified memory.<\/li>\n<li><strong>1,040 TB\/s<\/strong> aggregate NVLink bandwidth \u2014 near-zero-bottleneck all-to-all GPU communication.<\/li>\n<li><strong>~11.5 EFLOPS<\/strong> peak (FP4 sparse) and <strong>~8.8 EFLOPS<\/strong> dense per node for trillion-parameter reasoning.<\/li>\n<li>Warm-water <strong>direct liquid cooling<\/strong> with in-row CDU rated up to <strong>1.8 MW<\/strong> thermal dissipation.<\/li>\n<li>Redundant power: <strong>64\u00d7 33 kW<\/strong> shelves with integrated busbars and comprehensive BMS\/safety.<\/li>\n<li>High-speed data spine: <strong>ConnectX-8 (800 Gb\/s)<\/strong> + BlueField-3 DPUs; Quantum-X800 InfiniBand or Spectrum-X Ethernet options.<\/li>\n<li>Complete stack: <strong>NVOS<\/strong>, full <strong>NVIDIA AI Enterprise<\/strong> (576 GPU subscriptions), Mission Control &amp; DOCA \u2014 containerised and plug-and-play.<\/li>\n<\/ul>\n<h3>AI workload fit<\/h3>\n<ul>\n<li>Large-scale \/ foundation-model pretraining at trillion-parameter scale.<\/li>\n<li>Post-training alignment and fine-tuning (SFT \/ RLHF) on frontier models.<\/li>\n<li>Real-time, test-time-scaling inference and serving.<\/li>\n<li>Agentic AI and multi-step reasoning workloads.<\/li>\n<li>Generative AI (text, image, video), NLP &amp; speech model families.<\/li>\n<li>HPC + AI convergence and sovereign \/ national-scale AI.<\/li>\n<\/ul>\n<h3>AI workload positioning<\/h3>\n<p>This is a rack-scale-to-node building block for an AI factory. The balance of unified HBM3e capacity, 1,040 TB\/s NVLink, and a high-speed ConnectX-8 \/ BlueField-3 data spine lets 576 GPUs train and serve models that will not fit on a single rack, while warm-water DLC and redundant 33 kW power shelves sustain the density in continuous production. It sits at the top of the DRACO tier \u2014 above a single GB300 NVL72 rack \u2014 and scales out to 16-, 32- and 64-rack superclusters.<\/p>\n<h3>Industry use cases<\/h3>\n<ul>\n<li><strong>Sovereign &amp; public sector:<\/strong> data-resident national AI, on-soil foundation-model programs.<\/li>\n<li><strong>Neocloud \/ AI cloud:<\/strong> multi-tenant training-and-inference capacity as a deployable node.<\/li>\n<li><strong>Research &amp; higher education:<\/strong> frontier-scale training for labs and institutions.<\/li>\n<li><strong>Enterprise AI factories &amp; GCCs:<\/strong> in-house model development and agentic platforms.<\/li>\n<li><strong>Defence, telecom &amp; government:<\/strong> secure, air-gap-capable large-scale AI.<\/li>\n<\/ul>\n<h3>Performance &amp; how to be sure<\/h3>\n<p>Published node figures: ~11.5 EFLOPS peak (FP4 sparse) \/ ~8.8 EFLOPS dense, with a generational step over the Hopper baseline of ~50\u00d7+ AI-factory output, ~10\u00d7 user responsiveness and ~5\u00d7 throughput-per-watt. Rather than quote a single tokens\/sec number that will not match your models, RDP offers a <strong>&ldquo;benchmark your model&rdquo; POC<\/strong>: bring your training run or inference workload and we will size and validate it against this configuration before you commit.<\/p>\n<h3>Series &amp; upgrade path<\/h3>\n<p>DRACO is the flagship tier. The GB300 family scales: a single <strong>GB300 NVL72<\/strong> rack-scale AI factory \u2192 this <strong>8-rack containerised Supercluster<\/strong> (576 GPUs) \u2192 <strong>16 \/ 32 \/ 64-rack<\/strong> superclusters. Start at the node that matches your model scale and add nodes as demand grows.<\/p>\n<h3>On-prem vs cloud (TCO)<\/h3>\n<p>For sustained frontier training and inference, an owned node removes per-GPU-hour cloud billing, data-egress fees and multi-tenant contention, and keeps data on-soil for DPDP \/ sovereignty requirements. RDP provides predictable INR capital pricing, GST input-credit invoicing (HSN 8471), and financing \/ lease options to compare against a 3-year cloud run.<\/p>\n<h3>Software &amp; day-one readiness<\/h3>\n<p>Ships with <strong>NVOS<\/strong> managing the hardware layer, full <strong>NVIDIA AI Enterprise<\/strong> licensing across all 576 GPUs, and <strong>Mission Control<\/strong> + DOCA fleet orchestration. CUDA, cuDNN, drivers and container runtime are pre-integrated so the node is workload-ready on power-up.<\/p>\n<h3>Power, thermal &amp; acoustics<\/h3>\n<p>Housed in a high-cube container: <strong>64\u00d7 33 kW<\/strong> power shelves with redundant busbars, warm-water direct liquid cooling via an in-row CDU rated to <strong>~1.8 MW<\/strong>. Site needs grid + network hookup and a warm-water loop; the container isolates acoustics and thermal from occupied space.<\/p>\n<h3>Deployment, warranty &amp; support<\/h3>\n<p>Arrives pre-integrated and containerised, bypassing years of traditional data-center construction; plug-and-play for immediate grid and network hookup, with comprehensive training and detailed O&amp;M documentation. Backed by full NVIDIA Enterprise support and RDP pan-India onsite service, GST invoice, and GeM availability. <strong>Lead time: 12&ndash;16 weeks.<\/strong> This is premium, constrained inventory \u2014 allocation is prioritised for strategic partners.<\/p>\n<h3>Why RDP<\/h3>\n<p>RDP Technologies is a Make-in-India OEM with 14+ years and 300,000+ units shipped, delivering sovereign \/ DPDP-ready AI infrastructure with predictable INR pricing, GST invoicing and pan-India onsite SLAs.<\/p>\n<h3>Full node configuration (bill of materials)<\/h3>\n<p>One node = 8\u00d7 GB300 NVL72 racks. Per-node totals:<\/p>\n<ul>\n<li><strong>Compute:<\/strong> 144\u00d7 1U liquid-cooled compute trays \u00b7 288\u00d7 GB300 Grace-Blackwell Ultra superchips \u00b7 576\u00d7 Blackwell Ultra B300 GPUs (288 GB HBM3e each; FP4\/FP6\/FP8 Tensor Cores) \u00b7 288\u00d7 Grace CPUs (72-core Arm Neoverse V2, up to 480 GB LPDDR5X each).<\/li>\n<li><strong>NVLink scale-up:<\/strong> 72\u00d7 5th-gen NVLink switch trays \u00b7 18 NVLink-5 links per GPU \u00b7 1.8 TB\/s per GPU \u00b7 1,040 TB\/s aggregate per node.<\/li>\n<li><strong>In-rack networking:<\/strong> 576\u00d7 ConnectX-8 SuperNICs (dual-port 800 Gb\/s) \u00b7 144\u00d7 BlueField-3 B3240 DPUs \u00b7 16\u00d7 SN2201 OOB management switches.<\/li>\n<li><strong>Scale-out fabric:<\/strong> NVIDIA Quantum-X800 InfiniBand or Spectrum-X Ethernet leaf\/spine at 800 Gb\/s per GPU (selected at design stage), with pre-terminated optical trunks.<\/li>\n<li><strong>Storage:<\/strong> 1,152\u00d7 E1.S 3.84 TB NVMe (PCIe Gen5) + 144\u00d7 M.2 1.92 TB boot in-tray; external NVIDIA Enterprise RA-certified storage sized to workload.<\/li>\n<li><strong>Power:<\/strong> 64\u00d7 33 kW power shelves \u00b7 48V DC busbars \u00b7 IT load ~1.06\u20131.08 MW nominal, ~1.24 MW peak (EDPp), busway provisioned to ~1.54 MW.<\/li>\n<li><strong>Cooling:<\/strong> warm-water direct liquid cooling \u00b7 in-row CDU up to 1.8 MW with N+1 pumps \u00b7 blind-mate manifolds and leak detection.<\/li>\n<li><strong>Enclosure &amp; safety:<\/strong> high-cube containerised module housing all 8 racks + CDU \u00b7 clean-agent fire detection\/suppression \u00b7 environmental monitoring\/BMS \u00b7 physical access control.<\/li>\n<li><strong>Software:<\/strong> NVIDIA AI Enterprise (576 GPU subscriptions) \u00b7 Mission Control \u00b7 NVOS \u00b7 DOCA services on BlueField-3.<\/li>\n<li><strong>Programme services:<\/strong> logistics, staging &amp; commissioning \u00b7 site integration and liquid-loop commissioning \u00b7 burn-in &amp; acceptance testing \u00b7 operator training with as-built O&amp;M documentation.<\/li>\n<\/ul>\n<p><strong>Sustained performance:<\/strong> ~8.6 EFLOPS FP4 sustained per node (peak ~11.5 EFLOPS sparse \/ ~8.8 EFLOPS dense).<\/p>\n<h3>Buy with confidence<\/h3>\n<p>Use <strong>Request a Quote<\/strong> to reach a named RDP solution architect for site assessment, benchmark-your-model POC, financing options and a delivery plan. No obligation.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>8-rack containerised node \u00b7 576\u00d7 Blackwell Ultra B300 \u00b7 288\u00d7 Grace \u00b7 ~165.6 TB HBM3e \u00b7 warm-water DLC<\/p>\n","protected":false},"featured_media":1925,"comment_status":"open","ping_status":"closed","template":"","meta":{"_yoast_wpseo_title":"","_yoast_wpseo_metadesc":"","rank_math_title":"NVIDIA GB300 NVL72 Supercluster \u2014 8-Rack Containerised AI Factory (576 B300) | RDP GPU Mart","rank_math_description":"Turnkey NVIDIA GB300 NVL72 Supercluster: 8 liquid-cooled NVL72 racks, 576 Blackwell Ultra B300 GPUs + 288 Grace CPUs, ~11.5 EFLOPS, containerised and plug-and-play. Request a quote from RDP GPU Mart.","_hermes_jsonld":""},"product_brand":[],"product_cat":[16],"product_tag":[],"class_list":["post-2578","product","type-product","status-publish","has-post-thumbnail","product_cat-ai-superclusters","pa_form-factor-multi-rack","pa_gpu-model-nvidia-gb300-grace-blackwell-ultra","pa_industry-defence-aerospace","pa_industry-enterprise-gccs","pa_industry-neocloud","pa_industry-public-sector-sovereign-ai","pa_industry-research-higher-education","pa_industry-telecom-5g","pa_series-draco","pa_use-case-agentic-ai","pa_use-case-fine-tuning","pa_use-case-generative-ai","pa_use-case-hpc-ai","pa_use-case-inference","pa_use-case-llm-training","pa_use-case-nlp-speech","pa_use-case-rag","pa_use-case-sovereign-ai","pa_workload-fit-frontier-llm-training-trillion-parameter-reasoning","first","instock","featured","taxable","shipping-taxable","product-type-external"],"_links":{"self":[{"href":"https:\/\/rdp.in\/gpu-mart\/wp-json\/wp\/v2\/product\/2578","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=2578"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/rdp.in\/gpu-mart\/wp-json\/wp\/v2\/media\/1925"}],"wp:attachment":[{"href":"https:\/\/rdp.in\/gpu-mart\/wp-json\/wp\/v2\/media?parent=2578"}],"wp:term":[{"taxonomy":"product_brand","embeddable":true,"href":"https:\/\/rdp.in\/gpu-mart\/wp-json\/wp\/v2\/product_brand?post=2578"},{"taxonomy":"product_cat","embeddable":true,"href":"https:\/\/rdp.in\/gpu-mart\/wp-json\/wp\/v2\/product_cat?post=2578"},{"taxonomy":"product_tag","embeddable":true,"href":"https:\/\/rdp.in\/gpu-mart\/wp-json\/wp\/v2\/product_tag?post=2578"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}