WESTMINSTER, Colo., Aug. 4, 2026 — Liqid, a global leader in software-defined memory and GPU pooling infrastructure, today announced its role as a strategic technology partner in Abaco, the memory-centric computing system designed by Micron, a subcontractor to the U.S. Department of Energy’s (DOE) Pacific Northwest National Laboratory (PNNL). Liqid is providing the scale-up memory pool that allows PNNL to expose hundreds of terabytes of coherent active memory to data-intensive AI for Science workloads.
Solving the Memory Wall for Scientific AI
For decades, memory architectures have limited applications such as computational chemistry, molecular biology, and AI-driven simulation by restricting access to the large memory capacity they require. The Abaco system pioneers CXL-enabled memory pooling and sharing, shattering the capacity limits of individual servers to support massive-scale AI workloads.
Liqid delivers the industry’s first and only fully disaggregated, software-defined memory pooling solution. In a Liqid rack-scale configuration, up to 160TB of DRAM can be dedicated to a single host or dynamically allocated and shared across as many as 16 server nodes. Real-time allocation of terabytes of memory enables zero stranded capacity and massive performance scaling when needed.
What Liqid Contributes to PNNL
- Liqid EX-5410C Memory Platform: CXL 2.0 memory expansion system supporting up to 40TB of DRAM per chassis and scaling to a unified memory pool of more than 160TB for memory-intensive AI, HPC, and database workloads.
- Liqid Matrix Software: The orchestration and management layer that turns pooled memory into a programmable resource. Liqid Matrix delivers the industry’s only unified interface for real-time deployment, management, and orchestration of composable GPU, memory, and storage – and integrates with Kubernetes, Slurm, Ansible, and other tooling already used in scientific computing environments.
- Liqid CXL 2.0 Fabric: A dedicated CXL 2.0 switch and host bus adapters (HBAs) deliver ultra-low-latency, high-bandwidth connectivity between hosts and the memory pool.
Memory Sharing and Benchmark Results
Liqid and Micron have also collaborated to enable the industry’s first architecture that supports true memory-sharing capabilities. Micron’s famfs Linux kernel enhancement enables advanced memory sharing across systems with software cache coherence, laying the foundation for future performance gains through larger shared memory pools, reduced data movement, and more efficient access to massive datasets. With famfs, disaggregated memory pools from the Liqid system become exposed to the host operating system as a file system, requiring no application layer changes to harness the latency and bandwidth advantages of pooled memory.
Using the famfs Linux kernel in the Abaco system with Micron high-capacity RDIMMs, Micron lab benchmarks using representative workloads demonstrated performance improvements. Results showed up to 30x faster processing time for graph analytics workloads and up to 7x more tokens per second for certain KV cache workloads, validating the power of CXL memory pooling for AI and scientific computing, with additional optimization and performance gains expected as the platform enables new features.
“PNNL is exactly the kind of environment that proves what composable CXL memory can do at the frontier of science,” said Sumit Puri, Liqid Founder and CTO. “PNNL and Micron have built something singular: a memory-centric system designed for the workloads that could define the next decade of AI for Science. Liqid is honored to extend that platform with the scale, orchestration, and composability that turn pooled memory into a programmable resource. This is what tokens per dollar and tokens per watt look like in a national laboratory environment.”
Why It Matters for AI for Science
PNNL will leverage Abaco to advance AI for Science, with AI-integrated computational chemistry as the first user-facing workload. By extending the lab’s pooled memory with Liqid’s CXL memory fabric, the system can keep larger structures, larger parameter spaces, and heavier intermediate data resident in active memory, directly attacking the memory bottleneck that has historically constrained scientific AI.
PNNL’s testbed is available to DOE-funded researchers from national laboratories and academia through the Advanced Memory to Support Artificial Intelligence for Science (AMAIS) initiative, funded by the DOE Advanced Scientific Computing Research program within the DOE Office of Science.
Liqid is the leader in software-defined memory and GPU pooling infrastructure, delivering flexible, high-performance, and efficient on-premises datacenter and edge solutions for next-generation AI inference, KV cache, in-memory databases, VDI, virtualization, and HPC workloads. Liqid solutions are trusted by organizations across financial services, higher education and research, healthcare, telecommunications, media and entertainment, and government and defense.
Liqid enables customers to manage, configure, reconfigure, and scale essential compute, accelerators (GPU, DPU, TPU, FPGA), memory, storage, and networking into physical bare-metal server systems in seconds. Liqid customers can optimize their IT infrastructure and achieve up to 100% memory and GPU utilization for maximum tokens per watt, tokens per dollar, and tokens per second. Learn more at www.liqid.com.
Sectors:
Academia, Science
Tags:
Liqid
The Big Data Inside Amazon’s New Fire Phone
The new Fire phone that Amazon launched this week looks like your ordinary black smartphone,…
Three Reasons to be Scared of the Internet of Things
We know the Internet of Things forecasts: 50 billion connected devices by 2020. Apparently, there’s…
Wanted: Intelligent Middleware That Simplifies Big Data Analytics
We’ve seen tremendous technological innovation in the data analytics space over the past 10 years….
GPUs Tackle Massive Data of the Hive Mind
LIVE from GTC12 — The flock of birds that weaves seamlessly through the sky, propelled…
Why Hadoop on IBM Power
In the quest to achieve data-driven insight, Hadoop running on Intel X86-based processors has emerged…
Datanami’s Leverage Big Data Summit Wraps Up
Dialog and networking were on the Datanami agenda this week as we kicked off our…
AI Infrastructure Spending Has Surpassed $1 Trillion. Here’s Where the Money Is Going and What’s Next
AI used to be a race to build the best model. Now it’s a race…
Token Optimization in Enterprise AI: How Context Architecture Determines Whether Your AI Program Scales Economically
Enterprise AI budget forecasting is mostly guesswork right now. Most organizations set their initial allocations…
AI Is Creating a New Market for Infrastructure Data. S&P Global Wants In
The AI boom has kicked off a massive wave of data center construction around the…
Why the AI Semantic Layer Is Becoming the Foundation of Enterprise AI
BigDATAwire recently spoke with Cindi Howson, Chief Data and AI Strategy Officer (CAIO) at ThoughtSpot,…
Building Industrial Software with AI: Guardrails for Systems Where Code Touches the Physical World
AI coding assistants have become part of everyday software development. They write functions, generate tests,…
