← The Polylog AI Intelligence Brief
Morning Edition · Friday, November 13, 2026Published at 1:16 AM EST · New York
AMD says its Mext software keeps up to 99% of server performance with half the DRAM by moving cold pages to NVMe
The headline figure comes from one Spark workload on Azure, and an AMD chip-design test kept 80% of performance with 75% of memory on flash. Customers must buy a license, and the Linux driver is closed source.
AMD is marketing software as a response to the shortage of server memory. Its Mext product runs on Epyc servers and uses a machine-learning model to predict which memory pages a workload will need. It keeps those pages in DRAM and moves the…
Continue the AI Intelligence Brief
Track frontier labs, chips, export controls, model releases, regulation, and AI infrastructure.
- 5 AI intelligence signals a day
- Frontier labs, compute, and chips
- Model releases and AI infrastructure
- Source-grounded analysis with confidence labels
The Global Intelligence Brief stays free.
Part of a tracked trend
Memory Scarcity Drives Server Memory Tiering
As AI accelerators absorb memory supply, chip vendors and operators increasingly replace DRAM capacity with predicted tiering to flash and CXL memory, turning memory placement software into a competitive product.
Townsquare
Start a discussion in Townsquare.
More from this edition
- Jensen Huang resigns as Nvidia chief executive, and finance chief Colette Kress takes interim control
- Google's Gemini 4 Hydrogen scores 54 on the Artificial Analysis index, ahead of the larger Argon, at a lower list price
- Anthropic ships Claude Fable 5.2 to all customers and restricts the same model, as Mythos 5.2, to verified cyber and life-science users
- Anthropic starts its IPO roadshow at $70 to $78 a share, implying a fully diluted valuation of up to $1.404 trillion
- US regulator extends review of Nvidia GPU rental futures to January 8, delaying CME's compute contracts to 2027
- Anthropic reports the first 3D structure of a bacterial vault particle, solved with Claude assisting the lab work
- Map2SLAM preprint wraps a frozen streaming 3D model in a revisable map and cuts trajectory error by more than 40% against uncalibrated baselines
Comments
0No comments yet.