Marvell and Penguin announced a strategic partnership to address AI memory demands, jointly launching a DDR4 CXL memory expansion card and the industry's first multi-rack optical I/O memory pool, the MemoryAI Photonic Appliance. The solution enables multiple XPUs to share HBM-class memory bandwidth over optical I/O at distances of several meters, easing the memory wall facing AI computing and improving DRAM utilization. Built on the Marvell Photonic Fabric platform with silicon photonics memory module technology, the system scales to 16TB of DDR5 RDIMM and 578GB of HBM3E, delivers 57.6 Tbps of total device bandwidth, and supports HBM direct access, DDR5 direct access, and HBM-DDR5 caching modes for intelligent memory allocation. The two companies plan to deliver samples of the MemoryAI Photonic Appliance to select customers in H2 2027.
Translation(model translation, not the source's own text)
Marvell and Penguin jointly launch industry's first multi-rack optical I/O memory pool, delivering HBM-class bandwidth
IT之家 November 17 news: Marvell and Penguin announced yesterday local time a strategic collaboration targeting AI memory demands, jointly launching a DDR4 CXL memory expansion card and the industry's first multi-rack optical I/O memory pool, the MemoryAI Photonic Appliance.
This solution enables multiple XPUs to share HBM-class memory bandwidth over distances of several meters via optical I/O, alleviating the memory wall problem facing AI computing and improving the system's DRAM utilization efficiency.
The system is based on the Marvell Photonic Fabric platform and uses silicon photonics memory module technology, scalable to 16TB DDR5 RDIMM and 578GB HBM3E, with total device bandwidth reaching 57.6 Tbps, supporting HBM direct access, DDR5 direct access, and HBM-DDR5 caching mode, enabling intelligent memory allocation.
The two companies plan to deliver samples of the MemoryAI Photonic Appliance to select customers in 2027H2.
The translation is machine-generated and individual terms may be off. Nothing is rewritten or cut; the original is authoritative.
Published by IT之家
This site only summarises and aggregates; copyright stays with the original author. Headlines and summaries follow the source — where they differ, the original wins.