{"id":4271,"date":"2025-09-18T16:42:32","date_gmt":"2025-09-18T16:42:32","guid":{"rendered":"https:\/\/salarydistribution.com\/machine-learning\/2025\/09\/18\/nvidia-blackwell-born-for-extreme-scale-ai-inference\/"},"modified":"2025-09-18T16:42:32","modified_gmt":"2025-09-18T16:42:32","slug":"nvidia-blackwell-born-for-extreme-scale-ai-inference","status":"publish","type":"post","link":"https:\/\/salarydistribution.com\/machine-learning\/2025\/09\/18\/nvidia-blackwell-born-for-extreme-scale-ai-inference\/","title":{"rendered":"NVIDIA Blackwell: Born for Extreme-Scale AI Inference"},"content":{"rendered":"<div>\n<p><!-- OneTrust Cookies Consent Notice start for nvidia.com --><\/p>\n<p><!-- OneTrust Cookies Consent Notice end for nvidia.com --><\/p>\n<p>\t<!-- This site is optimized with the Yoast SEO Premium plugin v25.9 (Yoast SEO v25.9) - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ --><br \/>\n\t<title>NVIDIA Blackwell: Born for Extreme-Scale AI Inference | NVIDIA Blog<\/title><\/p>\n<p>\t<!-- \/ Yoast SEO Premium plugin. --><\/p>\n<p><!-- Stream WordPress user activity plugin v4.1.1 --><\/p>\n<p>\t\t\t\t<!-- Hotjar Tracking Code for NVIDIA --><\/p>\n<p>    <img decoding=\"async\" loading=\"lazy\" src=\"https:\/\/www.nvidia.com\/content\/dam\/1x1-00000000.png\" width=\"1\" height=\"1\" alt=\"country_code\"><\/p>\n<div id=\"page\" class=\"hfeed site\">\n\t<a class=\"skip-link screen-reader-text\" href=\"#content\">Skip to content<\/a><\/p>\n<p>\t<!-- #masthead --><\/p>\n<div class=\"full-width-layout dark\">\n<div class=\"full-width-layout__hero dark no-headline\">\n<div class=\"full-width-layout__hero-content dark no-headline\">\n<div class=\"full-width-layout__hero-content__inner dark no-headline\">\n<p>\n\t\t\t\t\tNVIDIA Blackwell\u2019s scale-up capabilities set the stage to scale out the world\u2019s largest AI factories.\n\t\t\t\t<\/p>\n<\/p><\/div>\n<\/p><\/div>\n<\/p><\/div>\n<div class=\"full-width-layout__sections\">\n<div class=\"full-width-layout__article-copy-section dark\">\n<div class=\"full-width-layout__copy\">\n<p>\u00a0<\/p>\n<p>\u00a0<\/p>\n<p>The NVIDIA Blackwell architecture is the reigning leader of the AI revolution.<\/p>\n<p>Many think of Blackwell as a chip, but it may be better to think of it as a platform powering large-scale AI infrastructure.<\/p>\n<\/div>\n<\/div>\n<div class=\"full-width-layout__article-copy-section dark\">\n<div class=\"full-width-layout__copy\">\n<p>\u00a0<\/p>\n<p>\u00a0<\/p>\n<p>\u00a0<\/p>\n<h2 class=\"full-width-layout__heading\"><strong>Surging Demand and Model Complexity<\/strong><\/h2>\n<p>Blackwell is the core of an entire system architecture designed specifically to power AI factories that produce intelligence using the largest and most complex AI models.<\/p>\n<p>Today\u2019s frontier AI models have hundreds of billions of parameters and are estimated to serve nearly a billion users per week. The next generation of models are expected to have well over a trillion parameters \u2014 and are being trained on tens of trillions of <a href=\"https:\/\/blogs.nvidia.com\/blog\/ai-tokens-explained\/\">tokens<\/a> of data drawn from text, image and video datasets.<\/p>\n<p>Scaling out a data center \u2014 harnessing up to thousands of computers to share the work \u2014 is necessary to meet this demand. But far greater performance and energy efficiency can come from first scaling up: by making a bigger computer.<\/p>\n<p>Blackwell redefines the limits of just how big we can go.<\/p>\n<\/div>\n<\/div>\n<div class=\"full-width-layout__full-width-video-section\">\n\t<video class=\"full-width-layout__video js-responsive-video\" autoplay muted loop playsinline data-sources='{\"mobile\":[{\"src\":\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/02-blackwell-ai-params-phone.mp4\",\"type\":\"video\/mp4\"},{\"src\":\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/02-blackwell-ai-params-phone.webm\",\"type\":\"video\/webm\"}],\"desktop\":[{\"src\":\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/02-blackwell-ai-params-desktop.mp4\",\"type\":\"video\/mp4\"},{\"src\":\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/02-blackwell-ai-params-desktop.webm\",\"type\":\"video\/webm\"}],\"tablet\":[{\"src\":\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/02-blackwell-ai-params-tablet.mp4\",\"type\":\"video\/mp4\"},{\"src\":\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/02-blackwell-ai-params-tablet.webm\",\"type\":\"video\/webm\"}]}'>Your browser does not support the video tag.<\/video><\/p>\n<p>\n\t\t\t<span class=\"full-width-layout__media-caption-callout\"><br \/>\n\t\t\tExponential growth of parameters in notable AI models over time. \t\t<\/span><\/p>\n<p>\t\t\t<span class=\"full-width-layout__media-credits\"><br \/>\n\t\t\tData Source: Epoch (2025), with major processing by Our World In Data\t\t<\/span>\n\t<\/p>\n<\/div>\n<div class=\"full-width-layout__article-copy-section dark\">\n<div class=\"full-width-layout__copy\">\n<p>\u00a0<\/p>\n<p>\u00a0<\/p>\n<p>\u00a0<\/p>\n<h2 class=\"full-width-layout__heading\"><strong>Today\u2019s Most Challenging Form of Computing<\/strong><\/h2>\n<p>AI factories are the machines of the next industrial revolution. Their work is AI inference \u2014 the most challenging form of computing known today \u2014 and their product is intelligence.<\/p>\n<p>These factories require infrastructure that can adapt, scale out and maximize every bit of compute resource available.<\/p>\n<p>What does that look like?<\/p>\n<p>A symphony of compute, networking, storage, power and cooling \u2014 with integration at the silicon and systems levels, up and down racks \u2014 orchestrated by software that sees tens of thousands of Blackwell GPUs as one.<\/p>\n<p>The new unit of the data center is <a href=\"https:\/\/www.nvidia.com\/en-us\/data-center\/gb200-nvl72\/\">NVIDIA GB200 NVL72<\/a>, a rack-scale system that acts as a single, massive GPU.<\/p>\n<p>\u00a0<\/p>\n<\/div>\n<\/div>\n<div class=\"full-width-layout__standard-image-section\">\n\t\t\t<img decoding=\"async\" width=\"2048\" height=\"1152\" src=\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/jenson-blackwell-nvl72-scale-scaled.jpg\" class=\"full-width-layout__image\" alt=\"\" loading=\"lazy\">\t<\/p>\n<p>\n\t\t\t<span class=\"full-width-layout__media-caption-callout\"><br \/>\n\t\t\tNVIDIA CEO Jensen Huang shows off the NVIDIA GB200 NVL72 system and the NVIDIA Grace Blackwell superchip during his keynote at CES 2025.\t\t<\/span><\/p>\n<\/div>\n<div class=\"full-width-layout__article-copy-section dark\">\n<div class=\"full-width-layout__copy\">\n<p>\u00a0<\/p>\n<p>\u00a0<\/p>\n<p><img decoding=\"async\" loading=\"lazy\" class=\"full-width-layout__image aligncenter\" src=\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/one-big-die-vector-scaled.jpg\" alt=\"GB300 Die Vector\" width=\"2048\" height=\"2003\" data-wp-editing=\"1\"><\/p>\n<h2 class=\"full-width-layout__heading\"><strong>Birth of a Superchip<\/strong><\/h2>\n<p>At the core, the NVIDIA Grace Blackwell superchip unites two Blackwell GPUs with one NVIDIA Grace CPU.<\/p>\n<p>Fusing them into a unified compute module \u2014 a superchip \u2014 boosts performance by an order of magnitude. To do so requires a new high-speed interconnect technology introduced with the NVIDIA Hopper architecture: <a href=\"https:\/\/www.nvidia.com\/en-us\/data-center\/nvlink\/\">NVIDIA NVLink<\/a> chip-to-chip.<\/p>\n<p>This technology unlocks seamless communication between the CPU and GPUs, enabling them to share memory directly, resulting in lower latency and higher throughput for AI workloads.<\/p>\n<p>\u00a0<\/p>\n<\/div>\n<\/div>\n<p>\n\t<video class=\"full-width-layout__video js-responsive-video\" autoplay muted loop playsinline data-sources='{\"desktop\":[{\"src\":\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/05-gb100-pick-and-place-desktop.webm\",\"type\":\"video\/webm\"},{\"src\":\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/05-gb100-pick-and-place-desktop.mp4\",\"type\":\"video\/mp4\"}],\"tablet\":[{\"src\":\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/05-gb100-pick-and-place-tablet.webm\",\"type\":\"video\/webm\"},{\"src\":\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/05-gb100-pick-and-place-tablet.mp4\",\"type\":\"video\/mp4\"}],\"mobile\":[{\"src\":\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/05-gb100-pick-and-place-phone.webm\",\"type\":\"video\/webm\"},{\"src\":\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/05-gb100-pick-and-place-phone.mp4\",\"type\":\"video\/mp4\"}]}'>Your browser does not support the video tag.<\/video><\/p>\n<div class=\"full-width-layout__50-50-image-row-section dark\">\n\t<img decoding=\"async\" width=\"1920\" height=\"1080\" src=\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/06A-TSMC-inside-FAB.jpg\" class=\"full-width-layout__50-50-image-row-image\" alt=\"\" loading=\"lazy\"><img decoding=\"async\" width=\"1920\" height=\"1080\" src=\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/06B-lada.jpg\" class=\"full-width-layout__50-50-image-row-image\" alt=\"\" loading=\"lazy\"><\/p>\n<p>\n\t\t\t<span class=\"full-width-layout__media-caption-callout\"><br \/>\n\t\t\tIt takes a symphony of creation, cutting, assembly and inspection to build a superchip.\t\t<\/span><\/p>\n<\/div>\n<div class=\"full-width-layout__article-copy-section dark\">\n<div class=\"full-width-layout__copy\">\n<p>\u00a0<\/p>\n<p>\u00a0<\/p>\n<p>\u00a0<\/p>\n<h2 class=\"full-width-layout__heading\"><b>A New Interconnect for the Superchip Era<\/b><\/h2>\n<p>Scaling this performance across multiple superchips without bottlenecks was impossible with previous networking technology. So NVIDIA created a new kind of interconnect to keep performance bottlenecks from emerging and enable AI at scale.<\/p>\n<\/div>\n<\/div>\n<p>\n\t<video class=\"full-width-layout__video js-responsive-video\" autoplay muted loop playsinline data-sources='{\"desktop\":[{\"src\":\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/07-GB300-hybrid-compute-tray-desktop.webm\",\"type\":\"video\/webm\"},{\"src\":\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/07-GB300-hybrid-compute-tray-desktop.mp4\",\"type\":\"video\/mp4\"}],\"tablet\":[{\"src\":\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/07-GB300-hybrid-compute-tray-tablet.webm\",\"type\":\"video\/webm\"},{\"src\":\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/07-GB300-hybrid-compute-tray-tablet.mp4\",\"type\":\"video\/mp4\"}],\"mobile\":[{\"src\":\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/07-GB300-hybrid-compute-tray-phone.webm\",\"type\":\"video\/webm\"},{\"src\":\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/07-GB300-hybrid-compute-tray-phone.mp4\",\"type\":\"video\/mp4\"}]}'>Your browser does not support the video tag.<\/video><\/p>\n<div class=\"full-width-layout__article-copy-section dark\">\n<div class=\"full-width-layout__copy\">\n<p>\u00a0<\/p>\n<p>\u00a0<\/p>\n<p><img decoding=\"async\" loading=\"lazy\" class=\"aligncenter\" src=\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/08-on-big-die-vector-nvlink-spine-scaled.jpg\" width=\"1920\" height=\"1878\"><\/p>\n<h2 class=\"full-width-layout__heading\"><strong>A Backbone That Clears Bottlenecks<\/strong><\/h2>\n<p>The NVIDIA NVLink Switch spine anchors GB200 NVL72 with a precisely engineered web of over 5,000 high-performance copper cables, connecting 72 GPUs across 18 compute trays to move data at a staggering 130 TB\/s.<\/p>\n<p>That\u2019s fast enough to transfer the entire internet\u2019s peak traffic in less than a second.<\/p>\n<p>\u00a0<\/p>\n<\/div>\n<\/div>\n<div class=\"full-width-layout__standard-image-section\">\n\t\t\t<img decoding=\"async\" width=\"1920\" height=\"1080\" src=\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/10-amphnol-spine.jpg\" class=\"full-width-layout__image\" alt=\"\" loading=\"lazy\">\t<\/p>\n<p>\n\t\t\t<span class=\"full-width-layout__media-caption-callout\"><br \/>\n\t\t\tTwo miles of copper wire is precisely cut, measured, assembled and tested to create the blisteringly fast NVIDIA NVLink Switch spine.\t\t<\/span><\/p>\n<\/div>\n<div class=\"full-width-layout__standard-image-section\">\n\t\t\t<img decoding=\"async\" width=\"1920\" height=\"1080\" src=\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/11-engineer-holding-appliance.jpg\" class=\"full-width-layout__image\" alt=\"\" loading=\"lazy\">\t<\/p>\n<p>\n\t\t\t<span class=\"full-width-layout__media-caption-callout\"><br \/>\n\t\t\tThe spine cartridge is inspected before installation.\t\t<\/span><\/p>\n<\/div>\n<div class=\"full-width-layout__standard-video-section\">\n\t<video class=\"full-width-layout__video js-responsive-video\" autoplay muted loop playsinline data-sources='{\"mobile\":[{\"src\":\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/09-nvlink-spine-pan-animation-phone.webm\",\"type\":\"video\/webm\"},{\"src\":\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/09-nvlink-spine-pan-animation-phone.mp4\",\"type\":\"video\/mp4\"}],\"tablet\":[{\"src\":\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/09-nvlink-spine-pan-animation-tablet.webm\",\"type\":\"video\/webm\"},{\"src\":\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/09-nvlink-spine-pan-animation-tablet.mp4\",\"type\":\"video\/mp4\"}],\"desktop\":[{\"src\":\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/09-nvlink-spine-pan-animation-desktop.mp4\",\"type\":\"video\/mp4\"},{\"src\":\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/09-nvlink-spine-pan-animation-desktop.webm\",\"type\":\"video\/webm\"}]}'>Your browser does not support the video tag.<\/video><\/p>\n<p>\n\t\t\t<span class=\"full-width-layout__media-caption-callout\"><br \/>\n\t\t\tThe spine, powered up, can move an entire internet\u2019s worth of data in less than a second.\t\t<\/span><\/p>\n<\/div>\n<div class=\"full-width-layout__article-copy-section dark\">\n<div class=\"full-width-layout__copy\">\n<p>\u00a0<\/p>\n<p>\u00a0<\/p>\n<p>\u00a0<\/p>\n<h2 class=\"full-width-layout__heading\"><strong>Building One Giant GPU for Inference<\/strong><\/h2>\n<p>The integration of all this advanced hardware and software, compute and networking enables GB200 NVL72 systems to unlock new possibilities for AI at scale.<\/p>\n<p>Each rack weighs one-and-a-half tons \u2014 featuring more than 600,000 parts, two miles of wire and millions of lines of code converged.<\/p>\n<p>It acts as one giant virtual GPU, making factory-scale AI inference possible, where every nanosecond and watt matters.<\/p>\n<p>\u00a0<\/p>\n<\/div>\n<\/div>\n<div class=\"full-width-layout__standard-image-section\">\n\t\t\t<img decoding=\"async\" width=\"1920\" height=\"1080\" src=\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/12-full-rack-install.jpg\" class=\"full-width-layout__image\" alt=\"\" loading=\"lazy\">\n\t<\/div>\n<p>\n\t<video class=\"full-width-layout__video js-responsive-video\" autoplay muted loop playsinline data-sources='{\"tablet\":[{\"src\":\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/13-datacenter-timelapse-tablet.mp4\",\"type\":\"video\/mp4\"},{\"src\":\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/13-datacenter-timelapse-tablet.webm\",\"type\":\"video\/webm\"}],\"mobile\":[{\"src\":\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/13-datacenter-timelapse-phone.webm\",\"type\":\"video\/webm\"},{\"src\":\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/13-datacenter-timelapse-phone.mp4\",\"type\":\"video\/mp4\"}],\"desktop\":[{\"src\":\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/13-datacenter-timelapse-desktop.mp4\",\"type\":\"video\/mp4\"},{\"src\":\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/13-datacenter-timelapse-desktop.webm\",\"type\":\"video\/webm\"}]}'>Your browser does not support the video tag.<\/video><\/p>\n<div class=\"full-width-layout__standard-image-section\">\n\t\t\t<img decoding=\"async\" width=\"1355\" height=\"2048\" src=\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/14-blackwell-rack-flat-r2-scaled.jpg\" class=\"full-width-layout__image\" alt=\"\" loading=\"lazy\">\n\t<\/div>\n<div class=\"full-width-layout__article-copy-section dark\">\n<div class=\"full-width-layout__copy\">\n<p>\u00a0<\/p>\n<p>\u00a0<\/p>\n<p>\u00a0<\/p>\n<h2 class=\"full-width-layout__heading\"><strong>GB200 NVL72 Everywhere<\/strong><\/h2>\n<p>NVIDIA then deconstructed GB200 NVL72 so that partners and customers can configure and build their own NVL72 systems.<\/p>\n<p>Each NVL72 system is a two-ton, 1.2-million-part supercomputer. NVL72 systems are manufactured across more than 150 factories worldwide with 200 technology partners.<\/p>\n<\/div>\n<\/div>\n<div class=\"full-width-layout__standard-image-section\">\n\t\t\t<img decoding=\"async\" width=\"1920\" height=\"1080\" src=\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/15-mgx-ecosystem.jpg\" class=\"full-width-layout__image\" alt=\"\" loading=\"lazy\">\n\t<\/div>\n<div class=\"full-width-layout__full-width-image-section\">\n<p>\t\t\t<span class=\"full-width-layout__media-credits\"><br \/>\n\t\t\tFrom cloud giants to system builders, partners worldwide are producing NVIDIA Blackwell NVL72 systems.\t\t<\/span>\n\t<\/p>\n<\/div>\n<div class=\"full-width-layout__article-copy-section dark\">\n<div class=\"full-width-layout__copy\">\n<p>\u00a0<\/p>\n<p>\u00a0<\/p>\n<p>\u00a0<\/p>\n<h2 class=\"full-width-layout__heading\"><strong>Time to Scale Out<\/strong><\/h2>\n<p>Tens of thousands of Blackwell NVL72 systems converge to create <a href=\"https:\/\/blogs.nvidia.com\/blog\/ai-factory\/\">AI factories<\/a>.<\/p>\n<p>Working together isn\u2019t enough. They must work as one.<\/p>\n<p><a href=\"https:\/\/www.nvidia.com\/en-us\/networking\/spectrumx\/\">NVIDIA Spectrum-X Ethernet<\/a> and <a href=\"https:\/\/www.nvidia.com\/en-us\/networking\/products\/infiniband\/quantum-x800\/\">Quantum-X800 InfiniBand<\/a> switches make this unified effort possible at the data center level.<\/p>\n<p>Each GPU in an NVL72 system is connected directly to the factory\u2019s data network, and to every other GPU in the system. GB200 NVL72 systems offer 400 Gbps of Ethernet or InfiniBand interconnect using<a href=\"https:\/\/www.nvidia.com\/en-us\/networking\/ethernet-adapters\/\"> NVIDIA ConnectX-7 NICs<\/a>.<\/p>\n<\/div>\n<\/div>\n<div class=\"full-width-layout__full-width-image-section\">\n\t\t\t<img decoding=\"async\" width=\"2048\" height=\"960\" src=\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/18-time-to-scale-out-scaled.jpg\" class=\"full-width-layout__image\" alt=\"\" loading=\"lazy\">\t<\/p>\n<p>\n\t\t\t<span class=\"full-width-layout__media-caption-callout\"><br \/>\n\t\t\tNVIDIA Quantum-X800 Switch, NVLink Switch, and Spectrum-X Ethernet unify one or many NVL72 systems to function as one.\t\t<\/span><\/p>\n<\/div>\n<div class=\"full-width-layout__50-50-image-row-section dark\">\n\t<img decoding=\"async\" width=\"2048\" height=\"1152\" src=\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/19A-data-center-beauty-shot-scaled.jpg\" class=\"full-width-layout__50-50-image-row-image\" alt=\"\" loading=\"lazy\"><img decoding=\"async\" width=\"2048\" height=\"1152\" src=\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/19B-data-center-beauty-shot-scaled.jpg\" class=\"full-width-layout__50-50-image-row-image\" alt=\"\" loading=\"lazy\"><\/div>\n<div class=\"full-width-layout__article-copy-section dark\">\n<div class=\"full-width-layout__copy\">\n<p>\u00a0<\/p>\n<p>\u00a0<\/p>\n<p><img decoding=\"async\" class=\"full-width-layout__image\" src=\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/20-bluefield-vector-die-scaled.jpg\"><\/p>\n<h2 class=\"full-width-layout__heading\"><strong>Opening Lines of Communication<\/strong><\/h2>\n<p>Scaling out AI factories requires many tools, each in service of one thing: unrestricted, parallel communication for every AI workload in the factory.<\/p>\n<p><a href=\"https:\/\/www.nvidia.com\/content\/dam\/en-zz\/Solutions\/Data-Center\/documents\/datasheet-nvidia-bluefield-3-dpu.pdf\">NVIDIA BlueField-3 DPUs<\/a> do their part to boost AI performance by offloading and accelerating the non-AI tasks that keep the factory running: the symphony of networking, storage and security.<\/p>\n<\/div>\n<\/div>\n<div class=\"full-width-layout__standard-image-section\">\n\t\t\t<img decoding=\"async\" width=\"2048\" height=\"1152\" src=\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/21-coreweave-aifactory-scaled.jpg\" class=\"full-width-layout__image\" alt=\"\" loading=\"lazy\">\t<\/p>\n<p>\n\t\t\t<span class=\"full-width-layout__media-caption-callout\"><br \/>\n\t\t\tNVIDIA GB200 NVL72 powers an AI factory by CoreWeave, an NVIDIA Cloud Partner.\t\t<\/span><\/p>\n<\/div>\n<div class=\"full-width-layout__standard-image-section\">\n\t\t\t<img decoding=\"async\" width=\"1920\" height=\"1080\" src=\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/22-cooling-tubes-data-center.jpg\" class=\"full-width-layout__image\" alt=\"\" loading=\"lazy\">\n\t<\/div>\n<div class=\"full-width-layout__article-copy-section dark\">\n<div class=\"full-width-layout__copy\">\n<p>\u00a0<\/p>\n<p>\u00a0<\/p>\n<p>\u00a0<\/p>\n<h2 class=\"full-width-layout__heading\"><strong>The AI Factory Operating System<\/strong><\/h2>\n<p>The data center is now the computer. <a href=\"https:\/\/www.nvidia.com\/en-us\/ai\/dynamo\/\">NVIDIA Dynamo<\/a> is its operating system.<\/p>\n<p>Dynamo orchestrates and coordinates AI inference requests across a large fleet of GPUs to ensure that AI factories run at the lowest possible cost to maximize productivity and revenue.<\/p>\n<p>It can add, remove and shift GPUs across workloads in response to surges in customer use, and route queries to the GPUs best fit for the job.<\/p>\n<p>\u00a0<\/p>\n<\/div>\n<\/div>\n<div class=\"full-width-layout__full-width-image-section\">\n\t\t\t<img decoding=\"async\" width=\"2048\" height=\"1094\" src=\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/23-xAI-Collossus-scaled.jpg\" class=\"full-width-layout__image\" alt=\"\" loading=\"lazy\">\t<\/p>\n<p>\n\t\t\t<span class=\"full-width-layout__media-caption-callout\"><br \/>\n\t\t\tColossus, xAI\u2019s AI supercomputer. Created in 122 days, it houses over 200,000 NVIDIA GPUs \u2014 an example of a full-stack, scale-out architecture.\t\t<\/span><\/p>\n<\/div>\n<div class=\"full-width-layout__article-copy-section dark\">\n<div class=\"full-width-layout__copy\">\n<p>\u00a0<\/p>\n<p>\u00a0<\/p>\n<p>Blackwell is more than a chip. It\u2019s the engine of AI factories.<\/p>\n<p>The world\u2019s largest-planned computing clusters are being built on the Blackwell and Blackwell Ultra architectures \u2014 with approximately 1,000 racks of NVIDIA GB300 systems produced each week.<\/p>\n<p>\u00a0<\/p>\n<p>\u00a0<\/p>\n<p>\u00a0<\/p>\n<\/div>\n<\/div>\n<\/div>\n<div class=\"full-width-layout__news-section\">\n<p>Related News<\/p>\n<div class=\"full-width-layout__news\">\n<article class=\"full-width-layout__news-post-tile  post-84815 post type-post status-publish format-standard has-post-thumbnail hentry category-enterprise category-generative-ai tag-artificial-intelligence tag-customer-stories tag-deep-learning-institute tag-economic-development tag-education tag-events tag-healthcare-life-sciences tag-high-performance-computing tag-industrial-manufacturing tag-isaac tag-jetson tag-nvidia-dgx tag-nvidia-in-europe tag-nvidia-nemo tag-nvidia-nim tag-physical-ai tag-riva tag-robotics tag-simulation-and-design tag-social-impact tag-sovereign-ai tag-tensorrt loop-item-1 for-pagenum-1\" id=\"related-news-post-84815\" role=\"article\" aria-labelledby=\"related-news-post-title-84815\" aria-describedby=\"related-news-post-desc-84815\">\n\t<a href=\"https:\/\/blogs.nvidia.com\/blog\/uk-partner-ecosystem-ai-makers\/\" aria-hidden=\"true\"><br \/>\n\t\t<img decoding=\"async\" width=\"960\" height=\"510\" src=\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/uk-ecosystem-featured-1280x680-1-960x510.jpg\" class=\"attachment-medium size-medium\" alt=\"\" aria-hidden=\"true\" loading=\"lazy\">\t<\/a><\/p>\n<p>\t<a id=\"related-news-post-title-84815\" class=\"full-width-layout__news-post-title\" href=\"https:\/\/blogs.nvidia.com\/blog\/uk-partner-ecosystem-ai-makers\/\"><br \/>\n\t\tThe AI Makers: NVIDIA Partners in UK Advance Physical and Agentic AI, Robotics, Life Sciences and More\t<\/a><\/p>\n<div id=\"related-news-post-desc-84815\" class=\"full-width-layout__news-post-excerpt\tdark\">\n\t\tThe U.K. is driving investments in sovereign AI, using the technology to advance industries like manufacturing, life sciences and more. During NVIDIA founder and CEO Jensen Huang\u2019s visit to the&#8230;    <a class=\"read-more\" aria-hidden=\"true\" href=\"https:\/\/blogs.nvidia.com\/blog\/uk-partner-ecosystem-ai-makers\/\">\t\tRead Article        <span data-icon=\"y\"><\/span>    <\/a>\t<\/div>\n<\/article>\n<article class=\"full-width-layout__news-post-tile  post-84628 post type-post status-publish format-standard has-post-thumbnail hentry category-enterprise category-hardware tag-nvidia-blackwell-platform tag-tensorrt loop-item-1 for-pagenum-1\" id=\"related-news-post-84628\" role=\"article\" aria-labelledby=\"related-news-post-title-84628\" aria-describedby=\"related-news-post-desc-84628\">\n\t<a href=\"https:\/\/blogs.nvidia.com\/blog\/mlperf-inference-blackwell-ultra\/\" aria-hidden=\"true\"><br \/>\n\t\t<img decoding=\"async\" width=\"960\" height=\"510\" src=\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/grace-corp-blog-gb300-nvl72-1280x680-r1-2-960x510.png\" class=\"attachment-medium size-medium\" alt=\"\" aria-hidden=\"true\" loading=\"lazy\">\t<\/a><\/p>\n<p>\t<a id=\"related-news-post-title-84628\" class=\"full-width-layout__news-post-title\" href=\"https:\/\/blogs.nvidia.com\/blog\/mlperf-inference-blackwell-ultra\/\"><br \/>\n\t\tNVIDIA Blackwell Ultra Sets the Bar in New MLPerf Inference Benchmark\t<\/a><\/p>\n<div id=\"related-news-post-desc-84628\" class=\"full-width-layout__news-post-excerpt\tdark\">\n\t\tInference performance is critical, as it directly influences the economics of an AI factory. The higher the throughput of AI factory infrastructure, the more tokens it can produce at a&#8230;    <a class=\"read-more\" aria-hidden=\"true\" href=\"https:\/\/blogs.nvidia.com\/blog\/mlperf-inference-blackwell-ultra\/\">\t\tRead Article        <span data-icon=\"y\"><\/span>    <\/a>\t<\/div>\n<\/article>\n<article class=\"full-width-layout__news-post-tile  post-84471 post type-post status-publish format-standard has-post-thumbnail hentry category-enterprise category-supercomputing tag-artificial-intelligence tag-climate tag-events tag-hardware tag-healthcare-life-sciences tag-high-performance-computing tag-nvidia-in-europe tag-nvidia-quantum-2 tag-quantum-computing tag-science tag-scientific-visualization tag-simulation-and-design tag-social-impact tag-supercomputing tag-synthetic-data-generation loop-item-1 for-pagenum-1\" id=\"related-news-post-84471\" role=\"article\" aria-labelledby=\"related-news-post-title-84471\" aria-describedby=\"related-news-post-desc-84471\">\n\t<a href=\"https:\/\/blogs.nvidia.com\/blog\/jupiter-exascale-supercomputer-live\/\" aria-hidden=\"true\"><br \/>\n\t\t<img decoding=\"async\" width=\"960\" height=\"510\" src=\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/jupiter-featured-still-1280x680-1-960x510.jpg\" class=\"attachment-medium size-medium\" alt=\"\" aria-hidden=\"true\" loading=\"lazy\">\t<\/a><\/p>\n<p>\t<a id=\"related-news-post-title-84471\" class=\"full-width-layout__news-post-title\" href=\"https:\/\/blogs.nvidia.com\/blog\/jupiter-exascale-supercomputer-live\/\"><br \/>\n\t\tNow Live: Europe\u2019s First Exascale Supercomputer, JUPITER, Accelerates Climate Research, Neuroscience, Quantum Simulation\t<\/a><\/p>\n<div id=\"related-news-post-desc-84471\" class=\"full-width-layout__news-post-excerpt\tdark\">\n\t\tThe J\u00fclich Supercomputing Centre\u2019s JUPITER \u2014 Europe\u2019s first exascale supercomputer \u2014 is officially live&#8230;.    <a class=\"read-more\" aria-hidden=\"true\" href=\"https:\/\/blogs.nvidia.com\/blog\/jupiter-exascale-supercomputer-live\/\">\t\tRead Article        <span data-icon=\"y\"><\/span>    <\/a>\t<\/div>\n<\/article>\n<article class=\"full-width-layout__news-post-tile  post-84453 post type-post status-publish format-standard has-post-thumbnail hentry category-enterprise category-generative-ai category-how-to tag-ai-on tag-artificial-intelligence tag-customer-stories tag-data-science tag-education tag-financial-services tag-industrial-manufacturing tag-nemotron tag-nvidia-blueprints tag-nvidia-nemo tag-nvidia-nim tag-omniverse tag-telecommunications loop-item-1 for-pagenum-1\" id=\"related-news-post-84453\" role=\"article\" aria-labelledby=\"related-news-post-title-84453\" aria-describedby=\"related-news-post-desc-84453\">\n\t<a href=\"https:\/\/blogs.nvidia.com\/blog\/ways-ai-agents-are-raising-team-performance\/\" aria-hidden=\"true\"><br \/>\n\t\t<img decoding=\"async\" width=\"960\" height=\"510\" src=\"https:\/\/blogs.nvidia.com\/wp-content\/uploads\/2025\/09\/ai-on-6-ways-featured-1280x680-2-960x510.jpg\" class=\"attachment-medium size-medium\" alt=\"\" aria-hidden=\"true\" loading=\"lazy\">\t<\/a><\/p>\n<p>\t<a id=\"related-news-post-title-84453\" class=\"full-width-layout__news-post-title\" href=\"https:\/\/blogs.nvidia.com\/blog\/ways-ai-agents-are-raising-team-performance\/\"><br \/>\n\t\tAI On: 6 Ways AI Agents Are Raising Team Performance \u2014 and How to Measure It\t<\/a><\/p>\n<div id=\"related-news-post-desc-84453\" class=\"full-width-layout__news-post-excerpt\tdark\">\n\t\tAI agents are expected to be involved in most business tasks within three years, with effective human-agent collaboration projected to increase human engagement in high-value tasks by 65%&#8230;.    <a class=\"read-more\" aria-hidden=\"true\" href=\"https:\/\/blogs.nvidia.com\/blog\/ways-ai-agents-are-raising-team-performance\/\">\t\tRead Article        <span data-icon=\"y\"><\/span>    <\/a>\t<\/div>\n<\/article><\/div>\n<\/p><\/div>\n<\/p><\/div>\n<p><!-- #colophon --><\/p>\n<\/div>\n<p><!-- #page --><\/p>\n<p><!-- #has-highlight-and-share -->\t\t<\/p>\n<\/div>\n","protected":false},"excerpt":{"rendered":"<p>https:\/\/blogs.nvidia.com\/blog\/blackwell-ai-inference\/<\/p>\n","protected":false},"author":0,"featured_media":4272,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[3],"tags":[],"_links":{"self":[{"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/posts\/4271"}],"collection":[{"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/types\/post"}],"replies":[{"embeddable":true,"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/comments?post=4271"}],"version-history":[{"count":0,"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/posts\/4271\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/media\/4272"}],"wp:attachment":[{"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/media?parent=4271"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/categories?post=4271"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/salarydistribution.com\/machine-learning\/wp-json\/wp\/v2\/tags?post=4271"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}