The Supermicro DLC-2 resolution reduces knowledge centre energy consumption by as much as 40% in comparison with air-cooled installations. These superior applied sciences allow quicker deployment and decreased time-to-online for cutting-edge liquid-cooled AI infrastructure. Moreover, the full price of possession decreases by as much as 20%. The excellent chilly plate protection of elements permits for decrease fan speeds and fewer required followers, considerably decreasing knowledge centre noise ranges to roughly 50dB.
“With the anticipated demand for liquid-cooled knowledge centres rising to 30% of all installations, we realised that present applied sciences had been inadequate to chill these new AI-optimised techniques,” mentioned Charles Liang, president and CEO of Supermicro. “Supermicro continues to stay dedicated to innovation, inexperienced computing, and enhancing the way forward for AI, by considerably decreasing knowledge centre energy and water consumption, noise, and area. Our newest liquid-cooling innovation, DLC-2, saves knowledge centre electrical energy prices by as much as 40%.”
Supermicro goals to save lots of 20% of knowledge centre prices and apply DLC-2 improvements as a part of knowledge centre constructing block options to make liquid-cooling extra broadly accessible and accessible.
A significant factor of the brand new liquid-cooling structure is a GPU-optimised Supermicro server, which incorporates eight NVIDIA Blackwell GPUs and two Intel Xeon 6 CPUs, all in simply 4U of rack top. This technique is designed to help elevated provide coolant temperatures. This distinctive and optimised design incorporates chilly plates for CPUs, GPUs, reminiscence, PCIe switches, and voltage regulators. This design reduces the necessity for high-speed followers and rear-door warmth exchangers, thereby reducing cooling prices for the information centre.
The brand new Supermicro DLC-2 resolution stack helps the brand new 4U entrance I/O NVIDIA HGX™ B200 8-GPU system, and the in-rack Coolant Distribution Unit (CDU) has an elevated capability of eradicating 250kW of warmth generated per rack. The Supermicro DLC-2 resolution additionally utilises vertical coolant distribution manifolds (CDMs) to take away sizzling liquid and return cooler liquid to the servers for your entire rack. The decreased rack area necessities permits extra servers to be put in, rising computing density per unit of flooring area. The vertical CDM is offered in numerous sizes, exactly matching the variety of servers put in within the rack. The complete DLC-2 resolution stack is totally built-in with Supermicro SuperCloud Composer software program for knowledge centre-level administration and infrastructure orchestration.
The environment friendly liquid circulation and almost full liquid-cooling warmth seize protection, at as much as 98% per server rack, permit for a rise within the inlet liquid temperature at as much as 45°C. The upper inlet temperature eliminates the necessity for chilled water, chiller compressor gear price, and extra energy utilization, saving as much as 40% of knowledge centre water consumption.
Mixed with liquid-cooled server racks and clusters, DLC-2 additionally provides hybrid cooling towers in addition to water towers as a part of knowledge centre constructing blocks. The hybrid cooling towers mix the options of ordinary dry and water towers right into a single design. That is particularly helpful in knowledge centre areas with robust seasonal temperature variation to scale back utilization of assets and prices additional.
Supermicro serves as a complete one-stop resolution supplier with international manufacturing scale, delivering knowledge centre-level resolution design, liquid-cooling applied sciences, networking, cabling, a full knowledge centre administration software program suite, L11 and L12 resolution validation, onsite deployment, {and professional} service and help. With manufacturing services throughout San Jose, Europe, and Asia, Supermicro provides unmatched manufacturing capability for liquid-cooled rack techniques. This ensures well timed supply, decreased whole price of possession (TCO), and constant high quality.
Knowledge Centre Constructing Block Options (DCBBS) simplify and shorten buildouts
Supermicro can be asserting Knowledge Centre Constructing Block Options, the best resolution to beat the immense complexities of outfitting liquid cooled AI factories with all important infrastructure elements, together with servers, storage, networking, rack, liquid cooling, software program, providers, and help. As an growth of Supermicro’s System Constructing Block Options, DCBBS adopts a standardised, but versatile resolution structure, vastly expanded in scope to deal with probably the most demanding AI knowledge centre coaching and inference workloads, enabling simpler knowledge centre planning, buildout, and operation – all whereas decreasing price.
“Supermicro’s DCBBS permits shoppers to simply assemble knowledge centre infrastructure with the quickest time-to-market and time-to-online benefit, deploying as rapidly as three months,” mentioned Charles Liang, president and CEO of Supermicro. “With our whole resolution protection, together with designing knowledge centre layouts and community topologies, energy and battery backup-units, DCBBS simplifies and accelerates AI knowledge centre buildouts resulting in decreased prices and improved high quality.”
DCBBS provides packages of pre-validated knowledge centre-level scalable items, together with a 256-node AI Manufacturing unit DCBBS scalable unit, designed to alleviate the burden of extended knowledge centre design by offering a streamlined bundle of flooring plans, rack elevations, invoice of supplies, and extra. Supermicro offers complete first-party providers to make sure mission success, ranging from session to on-site deployment and continued on-site help. DCBBS is customisable on the system-level, rack cluster-level, and knowledge centre-level to satisfy just about any mission necessities.
Together with our DLC-2 know-how, DCBBS additionally helps clients save as much as 40% energy, decreasing 60% knowledge centre footprint, and reducing 40% water consumption, all of which ends up in 20% decrease TCO.
AI Manufacturing unit Knowledge Centre-Degree Scalable Unit
The necessity for AI infrastructure continues to scale: AI coaching clusters require clusters of hundreds of GPUs to develop basis fashions. AI inference functions are additionally leveraging extra test-time compute capability by operating a number of inference passes with a combination of fashions to ship higher intelligence. Supermicro’s AI Manufacturing unit DCBBS bundle totally equips knowledge centres to sort out these rising AI computational necessities.
Options from Supermicro embody as much as 256 Liquid Cooled 4U Supermicro NVIDIA HGX™ system nodes, every system geared up with 8 NVIDIA Blackwell GPUs (2,048 GPUs in whole), interconnected with as much as 800Gb/s NVIDIA Quantum-X800 InfiniBand or NVIDIA Spectrum™ X Ethernet networking platform. The compute material is supported by elastically scalable tiered storage with high-performance PCIe Gen5 NVMe, TCO optimised Knowledge Lake nodes, and resilient administration system nodes for steady uninterrupted operation.
System-Degree, Rack-Degree, and Knowledge Centre-Degree Customisation
Supermicro incorporates a modular constructing block strategy, composed of three hierarchical ranges: the system-level, rack-level, and knowledge centre stage, giving clients unparalleled design choices in figuring out a system-level invoice of supplies, right down to choosing particular person elements, together with CPUs, GPUs, DIMMs, drives, and NICs. System-level customisation ensures the flexibility to satisfy specialised {hardware} necessities for a selected knowledge centre workloads and functions and permits for granular fine-tuning of knowledge centre assets.
Supermicro aids in designing rack enclosure elevation layouts to make sure optimisation for thermals and cabling, giving clients the flexibility to pick the kind of rack enclosure, together with 42U, 48U, and 52U configurations.
After the preliminary session with the shopper, Supermicro delivers a mission proposal tailor-made to a given knowledge centre energy finances, efficiency goal, or different necessities.
Supermicro DLC-2
With liquid-cooled knowledge centres rising from lower than 1% of the market to an anticipated 30% inside a yr, Supermicro is driving the industry-wide adoption of DLC by serving to clients obtain the problem of needing to construct new liquid-cooled knowledge centres that may extra effectively take away warmth.
DLC offers unmatched effectivity by capturing warmth instantly from the person chips, together with AI GPUs operating at 1000W TDP and past. Liquid cooling infrastructure is deliberate and deployed at knowledge centre scale, together with the piping and facility-side liquid cooling tower for warmth dissipation. Supermicro leads the {industry} in offering a complete resolution for direct-to-chip liquid cooling infrastructure, consisting of DLC techniques, in-rack or in-row coolant distribution items, coolant distribution manifolds, cooling in direction of, and extra.
