GoKawiilAMD expands MEXT memory-offload tech, cutting DRAM needs by up to 75%
AMD acquired MEXT, a predictive memory engine that offloads cold DRAM pages to flash storage and uses an AI model to prefetch data before it's needed. AMD says internal tests, such as with Synopsys PrimeTime, show 95% of full-DRAM performance using only 50% of physical DRAM, and the company plans to scale deployment from 145 internal servers to 2,000 next year, alongside a rollout partnership with HPE.
GoKawiil's interpretation of the reporting above, not reported fact.
With DRAM prices surging due to AI demand, a technology that lets servers run on far less physical memory while retaining most performance could meaningfully reduce enterprise hardware costs, if AMD's internal benchmarks hold up in real-world deployments. The HPE partnership and 'MEXT Ready' EPYC servers suggest AMD is positioning this as a broader infrastructure strategy rather than an isolated internal optimization, though full customer rollout timing remains unannounced.
- AMD says MEXT can cut DRAM requirements 20-75% while retaining up to 99% of performance.
- Deployment is scaling from 145 internal servers to a planned 2,000 next year, with HPE as a rollout partner.
- Phase 2, bringing MEXT licenses and RAM View software to customers, has no announced start date.
Source: tomshardware.com — Jake Roach, 2026-11-11
Published there as: “AMD shares how it can retain 95% of server performance with just half the DRAM, sidestepping memory shortages with MEXT”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.