Marvell Shows Photonic Fabric Memory Module With 32 TB of GPU RAM

The product name, Photonic Fabric Memory Module, points to Marvell's use of photonic links rather than electrical connections to attach GPUs to memory. TechRadar's headline said the module can link up to 16 GPUs and that the memory capacity reaches 32 TB. The report did not provide further technical specifications, pricing, availability or benchmark results in the material available.

TechRadar's headline also said Nvidia "won't stick around doing nothing," a phrase that points to expected competition in GPU memory and interconnect technology. The material did not include a specific Nvidia response or product plan.

Marvell's module is being positioned around two claims: large memory capacity for GPUs and light-based connections that can scale to multiple GPUs. The report links those claims to lower capital expenditure, though it did not detail the cost structure or compare the module with existing electrical interconnect systems.

Because the report is based on Marvell's presentation as covered by TechRadar, key questions remain about when the Photonic Fabric Memory Module will be available and how it will perform outside the company's own descriptions. The headline's reference to Nvidia makes clear that the competitive context extends beyond Marvell's announcement, but no further details on that competition were provided.

Editor's Summary

Marvell has shown a Photonic Fabric Memory Module that TechRadar says provides 32 TB of RAM for GPUs and uses light to connect up to 16 GPUs. The report frames the design as a way to cut CAPEX, while its headline says Nvidia is unlikely to remain idle. No pricing, availability or independent performance data was included in the available report.