Khoutam, did you notice the recent deliberate language change...

Khoutam, did you notice the recent deliberate language change from "tuneable retention" to "managed retention"? They were referring to it in the LinkedIn video as 4DS memory's "managed retention feature".

"Managed Retention Memory" is Microsoft's term for their take on what is needed to efficiently solve the AI bottleneck.

To me, its a clear acknowledgement from 4DS of what they have in their hands... Something that meets this spec.

We know 20nm success is critical, but the timing could be phenomenal. Within a few months of Microsoft saying "this is what the industry needs" and 4DS could very well deliver the solution.


heres an example of the value that power savings can create for a company like Microsoft: three mile island (nuclear power plant in US) is being re-opened and Microsoft has already bought all the power output for the next 20 years... There is US$3bn in taxes alone, so imagine how much it is in total.

Then imagine a product (4DS) coming along that slashes their energy consumption AND improves the performance of their offering...

How much would you pay to have access to that product?!?

This is why they've got their researchers trying to find a solution... energy consumption is a MASSIVE issue.







******

Microsoft republished their Jan paper this month, gave it a new title:

Storage Class Memory is Dead, All Hail Managed-Retention Memory: Rethinking Memory for the AI Era

Abstract
"AI clusters today are one of the major uses of High Bandwidth Memory (HBM). However, HBM is suboptimal for AI workloads for several reasons. Analysis shows HBM is overprovisioned on write performance, but underprovisioned on density and read bandwidth, and also has significant energy per bit overheads. It is also expensive, with lower yield than DRAM due to manufacturing complexity. We propose a new memory class: Managed-Retention Memory (MRM), which is more optimized to store key data structures for AI inference workloads. We believe that MRM may finally provide a path to viability for technologies that were originally proposed to support Storage Class Memory (SCM). These technologies traditionally offered long-term persistence (10+ years) but provided poor IO performance and/or endurance. MRM makes different trade-offs, and by understanding the workload IO patterns, MRM foregoes long-term data retention and write performance for better potential performance on the metrics important for these workloads."

https://www.microsoft.com/en-us/research/publication/storage-class-memory-is-dead-all-hail-managed-retention-memory-rethinking-memory-for-the-ai-era/