- XCENA introduced MX1 computational memory with thousands of RISC-V cores at FMS 2025
- MX1 offers near-data processing reducing CPU-memory overhead and enabling petabyte-scale SSD backed expansion
- Product roadmap includes MX1P this year and MX1S in 2026 supporting CXL 3.2
At the recent FMS 2025 event (formerly Flash Memory Summit but now called Future of Memory and Storage to better suit its expanded focus), South Korean startup XCENA took the wraps off its first product, MX1 Computational Memory.
MX1 is built on the PCIe Gen6 and the Compute Express Link 3.2 standard. In putting compute directly next to DRAM, the chip is able to reduce the overhead of moving data back and forth between processors and memory.
Known as near-data processing, this approach could well influence how servers are designed in the years ahead.
Thousands of RISC-V cores
XCENA says it has crammed “thousands” of in-house RISC-V cores into the MX1 in order to handle workloads such as vector database operations, analytics, and memory-heavy queries.
SSD-backed memory expansion enables petabyte-scale capacity while adding compression and reliability features.
ServeTheHome reports there will two different models available. The MX1P is expected later this year (XCENA says working samples will be made available to select partners starting in October), while the MX1S, with dual PCIe Gen6 x8 links and additional features, is slated for release in 2026.
Both will take advantage of the wider bandwidth and flexibility offered by the CXL 3.2 standard.
The product won the “Most Innovative Memory Technology” award at FMS 2025, making it the company’s second straight recognition at the event, after being named “Most Innovative Startup” in 2024.
“Computational memory represents an emerging architectural approach that aims to accelerate performance and efficiency, particularly for data-intensive tasks. It does this by minimizing data movement between processing and memory components,” said Jay Kramer, Chair of the Awards Program and President of Network Storage Advisors.
“We are proud to recognize XCENA with MX1, the world’s first computational memory controller supporting CXL 3.X and enabling high-throughput acceleration for data intensive workloads like AI and analytics.”
In the hope of driving interest in its product, XCENA, which was founded in 2022 as MetisX, is offering a software development kit that includes drivers, runtime libraries, and tools.
The stack is designed to fit standard environments so developers can evaluate and deploy the MX1 in applications ranging from AI inference to in-memory analytics.
You might also like
https://cdn.mos.cms.futurecdn.net/dHJLxqoQLn6phMf6FuPmvY.jpg
Source link
waynewilliams@onmail.com (Wayne Williams)
