Industree DDR5 MRDIMM Hits 12800 MT/s, Boosts AI Server Bandwidth 60%

Industree Embedded Technology released a new DDR5 MRDIMM module that targets the memory bandwidth bottleneck in AI servers. This release matters because AI language models struggle with data throughput, and higher memory speed directly improves token processing efficiency. The module aims to solve the 'memory wall' issue that currently limits compute core resources in data-intensive scenarios.

New module targets AI memory wall with 60% bandwidth gain

The new module uses a dual-rank simultaneous access architecture to reduce the load on the memory controller and improve latency. It incorporates two new components called the Multiplexed Register Clock Driver (MRCD) and the Multiplexed Data Buffer (MDB). These components work together to manage data flow more efficiently than standard server memory designs.

Specifications

• Transfer Speed: 12800 MT/s
• Bandwidth Improvement: 60% increase over DDR5 8000 MT/s RDIMM
• Pin Count: 287-pin
• Capacity Options: 32GB to 128GB
• Operating Temperature: 0 to 95 degrees Celsius

The memory module achieves a transfer speed of 12800 MT/s, which represents a 60% increase in bandwidth over standard DDR5 8000 MT/s RDIMM. It features a 287-pin design that fits into existing DDR5 RDIMM slots without requiring a system redesign. The module supports capacities ranging from 32GB to 128GB and operates in temperatures from 0 to 95 degrees Celsius.

Industree states that systems using this MRDIMM have higher system memory bandwidth and compute core resources. The company claims this significantly improves Token-based data processing efficiency when handling the same AI language models. The module also includes eFuse and TVS protection features to ensure reliability in harsh environments.

We looked at DDR5 Memory Prices Surge 320% as earlier while tracking Industree embedded technology launches. The memory is expected to start sampling globally in the fourth quarter of 2024. This product targets AI servers, edge AI inference, and other data-intensive applications where bandwidth is critical.