Enterprise success in modern AI, especially inference, reasoning, and agents, requires dense, high-performance, scale-up infrastructure before building out horizontally.
LIQID leads the way in building and enabling the most power- and cost-efficient Scale Up and Scale Out infrastructure solutions.
Enterprises are no longer limited to server instances that max out at 8 GPUs. With Liqid, you can Scale-Up your AI infrastructure by adding up to 30 GPUs to an industry-standard server and you can simultaneously scale memory to support even the largest models.
Liqid’s open-architecture enables leveraging any east-west Scale-Out fabric, as your needs grow, allowing you to cluster multiple systems together to achieve the perforrmance you require without having to move to costly AI Factories that were designed for large operators.
Dynamically pool and scale from as little as one up to 30-GPUs, including 600W devices such as the NVIDIA H200 or new RTX Pro 6000, Intel Gaudi 3, Grok, or d-Matrix without clustering and bandwidth-robbing interconnects.
Deliver massive GPU + Memory to a single CPU host, creating the highest performance while reducing hardware, power, and cooling.
Deploy GPUs, memory, and storage in seconds via Liqid Matrix Software to match your workload needs and precisely assign the necessary resources in real time so you can get to market faster.
LIQID’s RDMA Peer-to-Peer technology delivers 30% increase in performance through direct GPU-to-GPU and GPU-to-NMVe storage data exchange without CPU or host memory involvement; critical for minimizing latency and maximizing bandwidth for AI and HPC workloads.
Built for seamless scale-out using open standards-based building blocks that can be extended across racks and systems, enabling scale of AI factories.
Designed using open, standards-based technology means no lock-in so enterprises can scale out using existing networking hardware and interoperable protocols.
Built-in northbound APIs integrate seamlessly with orchestration tools such as Kubernetes, VMware and OpenShift, job schedulers like Slurm, and automation frameworks such as Ansible to build AI Factories and pool and share resources across the entire datacenter.
Supports Ethernet (RoCE v2, v3) and InfiniBand scale-out fabrics now, with future-proof alignment for emerging fabric technologies, like Nvidia Fusion, Ultra Accelerator Link (UAL), or Ultra Ethernet Connection (UEC) as they become available.
Scale beyond 30 devices and implement multi-tenancy across departments, enabling each to adopt its own inference models or fine-tune LLMs as well as offer high availability (HA) and recovery, distributed management, and enhanced GPU and memory sharing.
LIQID Matrix is the first and only single UI for GPU and Memory composability, enabling simplified scale-up and a software-defined foundation for scale-out.
Power off slots and virtually replace failed devices, like GPUs and DIMMs, non-disruptively via software to minimize downtime and maximize operational efficiency.
Command and control critical AI resources in real-time to meet the demand of diverse and dynamic workloads and achieve 100% and balanced utilization.
Compose from LIQID Matrix GUI, or leverage built-in REST APIs that seamlessly connect with Kubernetes, Slurm, OpenShift, VMware, Ansible, and others.
Speak with one of our product experts and learn how Liqid can drive greater agility and value from your datacenter.
Schedule a demo
Liqid’s innovative software-defined composable infrastructure platform enables users to manage, scale out, and configure physical bare-metal server systems in seconds.
