Marvell Structera CXL Accelerators Compress Data 3.64x to Ease Memory Shortages

Memory shortages are forcing data centers to rethink how they handle data, and Marvell is responding with hardware that shrinks information before it hits the RAM. The company introduced the Structera CXL Accelerators to reduce the physical memory footprint required for intensive database and AI workloads. This approach allows servers to store more data in the same space, which matters because buying more DRAM is becoming increasingly expensive and difficult. By compressing data at the hardware level, these accelerators help operators stretch their existing memory capacity without waiting for new chips to arrive.

New hardware controllers stretch DDR5 capacity with inline LZ4 compression

The new lineup includes two distinct controllers: the Structera X and the Structera A. Both devices connect to servers using CXL 2.0 and PCIe 5.0 interfaces, supporting either a 16-lane or dual 8-lane configuration. The Structera X focuses on pure memory expansion, supporting more than 6TB of DDR5 DRAM capacity. The Structera A variant adds processing power by integrating 16 Arm Neoverse V2 cores running at 3.2 GHz. These controllers sit between the host processor and the memory modules, managing data flow and compression transparently.

Specifications

• Compression Ratio: Up to 3.64x
• Compression Algorithm: LZ4
• Encryption: XTS-AES 256-bit
• Memory Bandwidth: 200 GB/s
• Interface: CXL 2.0 / PCIe 5.0

At the core of the technology is a dedicated Compression-Decompression Block that uses the LZ4 algorithm. This hardware block performs inline compression and decompression, meaning the host CPU does not need to dedicate cycles to shrinking data. Marvell claims the Structera CDB matches the compression quality of standard host-side software implementations. The system also includes inline XTS-AES 256-bit encryption and decryption for data security. The accelerators deliver up to 200 GB/s of memory bandwidth to handle the compressed data streams efficiently.

Performance testing shows the hardware can achieve compression ratios of up to 3.64x on specific database workloads. This ratio allows the system to store significantly more database records within the same physical memory limits. The compression happens automatically as data moves through the controller, reducing the load on the main processor. The Structera A model leverages its 16 Arm cores to manage these complex tasks alongside the compression block. This design ensures that the overhead of compression does not bottleneck the overall system performance.

Marvell positioned these accelerators as a direct solution to the intensifying memory shortages affecting the AI and data center markets. The technology offers a way to maintain performance while reducing the dependency on large volumes of physical DRAM. Operators can now deploy more workloads per server rack by utilizing the 3.64x compression capability. The Structera X and A controllers provide a hardware-based alternative to software compression, which typically consumes valuable CPU resources. This shift allows data centers to optimize their infrastructure costs without sacrificing speed or security.