Primemas Unveils Groundbreaking AI Memory Solutions at FMS 2026 Conference

Primemas Unveils Groundbreaking AI Memory Solutions at FMS 2026



Primemas Inc., a pioneer in advanced AI memory systems and chiplet-based silicon technology, made headlines at the Future of Memory and Storage (FMS) 2026 event. The company showcased its innovative CXL memory solutions, including its exclusive end-to-end CXL 3.0 portfolio and its collaboration with Micron on the ambitious Abaco Project.

Shifting Paradigms in Memory Architecture



The Abaco Project sets out to revolutionize how data centers handle memory for extensive scientific AI and high-performance computing (HPC) applications. Developed for the Department of Energy's Pacific Northwest National Laboratory (PNNL), this groundbreaking project promises over 100TB of pooled memory capacity, designed to meet the requirements of next-generation, data-intensive workloads.

As AI datasets continue to grow exponentially, traditional memory architectures face significant limitations. By transforming memory into a dynamic, rack-scale infrastructure layer, the Abaco System Architecture allows advanced CPU and GPU clusters to access vast pools of Micron's DDR5 memory, thus relieving the pressure of local memory constraints.

Innovative CXL Memory Expansion Cards



Central to the Abaco architecture are the Primemas CXL add-in cards, which serve as the core engine for enhancing memory capabilities. Available in two variants, the PMA14 and PMA16, these PCIe-based CXL memory expansion cards can support up to 4TB of DRAM capacity, significantly boosting memory per CXL port in contrast to conventional memory solutions.

Scheduled for shipment to Micron in September, these add-in cards will be pivotal for the multi-phase testing and benchmarking required for the Abaco system’s successful deployment.

SLiM: A Leap Forward in AI Inference



In addition to showcasing the Abaco Project, Primemas also introduced its SLiM (Switchless Pooled Memory) architecture. This innovative system is tailored for AI inference workloads that heavily rely on KV-cache, offering both high capacity and ultra-low latency without the complexity and cost associated with external CXL switches. The 1U space-efficient design aims to streamline access to pooled DRAM across multiple servers, enhancing performance and efficiency.

Industry-Driven Changes



The increasing demands of AI and data-centric applications necessitate new architectures that optimize memory utilization across the data center landscape. Luis Ancajas, Director of CXL Product Management at Micron, highlighted the critical role collaboration plays in advancing memory system innovations. “Working alongside the Department of Energy, PNNL, and strategic partners like Primemas, we are pushing the boundaries of what’s possible in memory technology,” he stated.

Furthermore, Sumit Puri, Co-Founder and CTO at Liqid, remarked on the collaborative synergy of CXL silicon and Micron DDR5 with their memory pooling solutions, emphasizing the practicality of accessing hundreds of terabytes of system memory as a unified resource.

A Future-Oriented Vision



“There is a distinct shift happening towards disaggregated memory models,” asserted Joseph Baco, VP of Business Development at Primemas. The enthusiasm and inquiries about the upcoming deployment of the Abaco Project at PNNL serve as validation of the industry's commitment to overcoming traditional memory limitations.

As the momentum builds, Primemas has laid out an ambitious roadmap towards intelligent memory systems, integrating CPU and FPGA acceleration within their Hublet architecture. This leap is designed to create sophisticated near-memory computing platforms specifically aimed at enhancing AI capabilities and database acceleration.

For those interested in exploring more about the PNNL Abaco deployment or Primemas’ cutting-edge products, the business development team is available for inquiries.

About Primemas



Founded in Santa Clara, California, Primemas focuses on delivering efficient semiconductor solutions through its Hublet architecture, catering to the growing demands within sectors like AI and data analytics. Their innovative memory products not only enhance performance but also promise a more cost-effective development process for emerging technologies. For additional information, visit www.primemas.com

Topics Consumer Technology)

【About Using Articles】

You can freely use the title and article content by linking to the page where the article is posted.
※ Images cannot be used.

【About Links】

Links are free to use.