Datacenter Cooling Simplified
|
Maxwell Labs is positioned to offer datacenter customers a 15% cost savings on their operating costs by reducing the average operating temperatures of their servers. The one-step product offering adds additional cost-savings on top of this through advanced monitoring services based on a secure, state-of-the-art IoT-based device. The Datacenter Efficiency Optimizer Service (DCEOS) scales to both efficiency-focused datacenters (cost-savings via temperature reduction), and performance-focused datacenters (increased capability via frequency boosting). In both contexts, a 6-month service agreement results in a one-time optimization service & install of the IoT devices, and at the same time upgrades the highest power components with thermal compounds built from bleeding edge materials - for free. |
|
Advanced Monitoring
Our thermal compounds are tightly integrated into our sensor technology. Because these compounds are form-factor agnostic, they can be retrofitted to any system configuration or hardware. Combined with secure IoT-based monitoring, the solution reduces cost of cooling, improves performance, and lessens the need for redundancy.
Customers receive notifications of elevated temperatures, anomalous hot-spot formation and server behavior, and even advanced notice of impending failures - allowing them to schedule maintenance weeks in advance and reduce server downtime, or address runaway thermal events before they degrade performance of nearby servers. But perhaps the most impressive benefit of our monitoring services is the security it offers to our customers. Unlike traditional application monitoring services or predictive maintenance offerings, our approach does not require any software installation or cloud service connectivity. Our stand-alone device decouples the benefits of monitoring from your hardware and network. This is in stark contrast to other similar services which require bloated software be running 24x7 on every server in your datacenter, introducing security vulnerabilities and interfering with application performance. Instead, we utilize state-of-the-art machine learning algorithms embedded into our devices to infer application identity, detect anomalous behavior, and even predict when hardware will fail. |
|
Cost Savings that Scale
Over 40% of the operating budgets for datacenters is spent on power for cooling, with an equal amount of energy spent on powering the servers themselves.
High-impact innovations using materials for radiative transfer of waste heat are now possible using metamaterials and functionalized nanomaterials.
Maxwell Labs has developed core technologies around these emerging capabilities to reduce datacenter cooling requirements by over 15% for CPU and 20% for GPUs & ASICs - leading directly to immediate cost-savings after installation. The theoretical limits of radiative cooling technologies are not yet known, but next generation products are estimated to realize efficiency improvements of over 30%. The minimum net annual cost savings we provide our customers is ~$170 per standard 2U server, and over ~$650 per dense-accelerator server (8 GPUs).
For small-scale datacenters with a single rack (42U), this equates to over $3500 USD annual cost savings per rack.
For mid-range datacenters (100 racks) the net annual cost savings is over $350,000 USD using our service.
Contact us at contact@maxwell-labs.com or visit maxwell-labs.com for more information.
High-impact innovations using materials for radiative transfer of waste heat are now possible using metamaterials and functionalized nanomaterials.
Maxwell Labs has developed core technologies around these emerging capabilities to reduce datacenter cooling requirements by over 15% for CPU and 20% for GPUs & ASICs - leading directly to immediate cost-savings after installation. The theoretical limits of radiative cooling technologies are not yet known, but next generation products are estimated to realize efficiency improvements of over 30%. The minimum net annual cost savings we provide our customers is ~$170 per standard 2U server, and over ~$650 per dense-accelerator server (8 GPUs).
For small-scale datacenters with a single rack (42U), this equates to over $3500 USD annual cost savings per rack.
For mid-range datacenters (100 racks) the net annual cost savings is over $350,000 USD using our service.
Contact us at contact@maxwell-labs.com or visit maxwell-labs.com for more information.
Fine-Grain Solid-State Optimization
No Water, No Pumps, No moving parts.
Mesh-based interfaces and advanced materials enable targeted cooling across every component within a server - from next-gen accelerators, ASICs and FPGAs, to memory, chipsets and VRMs. |
Intelligent Dynamics
Thermal transfer and heat pumping are controlled by state-of-the-art AI that learns workload hot-spots and predicts where and when to cool. A composable, stackable and highly modular design allows many units to be woven together into complex topologies - working together to keep datacenters cooler and take high-performance applications to peak performance. |
Novel Hybrid Sensor Layers
The key innovation by Maxwell Labs Engineers is the capability for heatsinks to simultaneously sense and respond to hot spots through a thin-film material known as a Thermal Transfer Layer (TTL). The TTL serves as the "eyes" of the algorithm, and simultaneously transfers heat away from a heat source.
|
AI-Powered Monitoring and Cooling Solutions
Our product line of Intelligent Fine-Grain Heat Routing (IFG-HR) devices are based on unique materials and state-of-the-art deep neural networks trained through Reinforcement Learning (RL) on customer-specific simulation environments.
The thermal video to the right illustrates the process of training a IFG-HR device on a grid of hot spots. The environment used for training the algorithm utilizes real-world sensor data combined with data generated by simulations. The grid on which the algorithm plays can be configured to represent arbitrary granularity, topologies and stacking of configurations, allowing for customer-specific customization of cooling capability, component monitoring and performance optimization. |
|