Marvell intros CXL 2.0 memory acceleration and expansion gear

Marvell has launched Structera brand CXL 2.0 memory expansion and acceleration products with a DDR4 DRAM recycling capability to enable its reuse as DDR5 memory gets installed.

Raghib Hussain

CXL (Computer eXpress Link) is the extension of the PCIe bus outside a server’s chassis, based on the PCIe 5.0 standard. CXL 2.0 provides for memory pooling between a CXL memory device and more than a single accessing host. It enables cache coherency between a server CPU host and three device types. Marvell reckons that X86-based data centers will find their servers running out of memory when executing Gen AI-type workloads such as  deep learning recommendation models (DLRM) and other machine learning applications. Its Structera products ”enable terabytes of memory to be added to general-purpose servers and address high-capacity memory applications such as in-memory databases.”

Raghib Hussain, Marvell’s president of products and technologies, put out a statement: “Our new Structera CXL product line will be a game-changer in enabling optimal resource utilization and lowering energy consumption for scaling memory-intensive workloads in the cloud,” meaning, we understand, both private and public clouds.

There are two product lines, each supporting inline LZ4 compression/decompression and AES-XS 256-bit encryption. They also have embedded hardware security modules and a secure boot facility. The compression HW adheres to the Google and Meta specifications that have been submitted to the Open Compute Project (OCP).

The  Structera A 2504 memory accelerator, which “integrate server-class processor cores and multiple memory channels with CXL to address high-bandwidth memory applications” for one or more servers. The Structera X 2404 and X 2504 products are “designed for expanding memory capacity per server to address high-capacity memory applications.“ Both are PCIe 5.0 interconnect designs. 

The A 2504 comes with a 16-core Arm Neoverse CPU, up to 4TB of DDR5 memory capacity – using 32 x 128GB DIMMs – and supports up to 200 GBps memory bandwidth.

Marvell Structera A 2504 diagram

Marvell says that adding one Structera A 2504 to a single 64-core DLRM server would increase the number of compute cores by 25 percent (64 vs 80), aggregate memory bandwidth by 50 percent (400 GBps vs 600 GBps), total memory by up to 4TB, and memory bandwidth per core by 25 percent (6.25 GBps vs. 7.5 GBps. It would also improve memory bandwidth power efficiency from 1W to 0.83W per GBpsec. 

Adding two A 2504s would increase compute cores by 50 percent, double aggregate memory bandwidth, increase memory capacity by up to 8TB, increase memory bandwidth per core by 50 percent, and improve memory bandwidth power efficiency to 0.75W per GB/sec. 

The X 2404 and X 2504 memory pool products ditch the Neoverse Arm CPU but are otherwise much the same,. The X2404 supports 4 DDR4 memory modules, with 3 DIMMs per channel, while the X2504 supports a quartet of faster DDR5 chips and with 2 DIMMs per channel. 

Marvell Structera X 2404 diagram. The X 2504 block diagram has DDR5 memory in its bottom layer and a 16 lane PCIe/CXL module in its top right box

Marvell claims that these “are also the first CXL memory-expansion products that can simultaneously support two server CPUs to optimize power, space and memory utilization.”

It makes the point that “the Structera X 2404 enables operators to recycle their DDR4 DIMMs from decommissioned general-purpose servers. Millions of functional DDR4 DIMMs are expected to become e-waste over the next few years as operators replace existing general-purpose servers with new ones utilizing DDR5.”

This means that “using “free” DDR4 DIMMs to expand the capacity in these servers reduces CAPEX by thousands of dollars per general-purpose server4. Repurposing decommissioned DDR4 DIMMs also addresses data center sustainability goals.”

Marvell will develops custom CXL silicon for public cloud operators that is optimized for their unique architectures and workloads. 

The Structera release had supportive statements from AMD, Arm, Intel and Micron.

Structera A near-memory accelerators, Structera X memory-expansion controllers and custom CXL silicon will sample in the fourth quarter of 2024. Get a Structera A product brief here, a Structera X 2404 product brief here and an X2504 product brief here.