Marvell Technology’s new Structera memory controllers do one thing that competing hardware does not: they compress data while it sits in DRAM. That single capability, delivered through a dedicated silicon block the company calls the Compression-Decompression Block, or CDB, is the entire story here. The CDB operates at full bandwidth.
It compresses data as it is written to memory and decompresses it on read. The host CPU never touches the process.
This is not software compression, which drags down system performance. It is a purpose-built hardware solution that works in real time, independent of the processor. Marvell says the compression ratio can reach 3.64 times.
That means a system with 256 gigabytes of DRAM could effectively hold over 900 gigabytes of data, from the perspective of the applications using it. For AI workloads, which the company describes as consuming memory at an unprecedented rate, that kind of density matters.
The broader tech industry is facing widespread shortages of memory and storage. Making every gigabyte count is the practical goal. The two new controllers are called Structera X and Structera A.
Both are part of Marvell’s CXL family. CXL, or Compute Express Link, is an interconnect standard that lets CPUs, memory, and accelerators share data coherently. Most CXL controllers on the market do not offer compression for the data set.
Marvell’s do. That is the difference.
The company submitted its compression specifications to the Open Compute Project. Marvell claims this makes its controller the only CXL memory controller with inline compression available as an in-production feature. That claim, if it holds, gives the company a clear lead in a market where memory capacity is the bottleneck.
Think of it this way. Computer memory is like counter space in a busy kitchen.
When the counter is full, cooking stops. Marvell’s hardware compresses the ingredients as they are put away, then expands them again when the chef needs them. The chef never knows it happened.
The kitchen just has more room. The practical upshot is significant. AI workloads demand high memory capacities.
The need for bandwidth and density is not slowing down. Memory shortages are real and widespread.
Marvell is betting that hardware compression at the memory controller level is the answer, rather than relying on software or on simply adding more DRAM. Adding more DRAM is expensive and physically constrained. Compression is not.
The dedicated silicon solution operates at full bandwidth. That is the key engineering detail.
Software compression would bog the system down. The CDB does not. It compresses and decompresses in real time, at line rate, without involving the host CPU.
That frees up memory resources for the platform. The CPU can focus on computation, not on compression overhead. Marvell unveiled the Structera controllers on June 27 in Santa Clara.
The move comes as AI workloads devour memory at an unprecedented rate. The broader tech industry faces widespread shortages of memory and storage solutions.
Marvell’s answer is a hardware block that compresses data in memory, transparently, at full bandwidth, and to a ratio of up to 3.64 times. That is the development. The rest is detail.
{“@context”: “https://schema.org”, “@type”: “NewsArticle”, “headline”: “Marvell’s New Memory Controllers Compress Data Directly in DRAM”, “description”: “Marvell Technology’s new Structera memory controllers do one thing that competing hardware does not: they compress data while it sits in DRAM. That single capability,…”, “datePublished”: “2026-06-27T13:42:00+00:00T09:30:00+00:00”, “author”: {“@type”: “Person”, “name”: “Ariel Antonio”}, “publisher”: {“@type”: “Organization”, “name”: “InfoPulse Today”, “url”: “https://newsfeed.infopulsetoday.com/”}}
