2X Memory Capacity in 2 Seconds

What would your team build if memory were no longer a constraint? How much better could your applications perform? And what if all of this could be accomplished in a matter of minutes—or even seconds?

For organizations running memory-bound or memory-intensive workloads—like large-scale AI/ML training, in-memory databases, real-time analytics, or high-performance computing—server memory, or DRAM, is often the bottleneck that forces trade-offs in speed, scale, or cost.

But what if we could rethink those limits entirely? Imagine what your architecture, application performance, or customer experience could look like if memory capacity could elastically scale with your needs—no forklift upgrades, no sky-high cloud bills, just more room to run.

With MEXT, this becomes possible.

MEXT Predictive Memory

MEXT Predictive Memory is an AI-powered, patent-pending, software-only solution designed to cost-effectively expand memory capacity. Here’s how it works.

MEXT continually monitors which pages in DRAM are hot, or actively utilized, and which have gone cold. It offloads the cold pages to NVMe Flash drives. MEXT’s AI engine predicts which offloaded pages might be requested by the application, and transparently moves them back into DRAM before the requests are ever made.

The end result is that the system’s Flash essentially appears as additional memory to the OS. In other words, the application sees its total available memory as the combination of its DRAM memory AND MEXT Memory (in an amount equating to the size of the Flash). This enables at least 2X—and sometimes up to 4X—the memory capacity, for very minimal additional costs. These additional costs are pennies compared to what it would have cost to purchase the equivalent memory capacity in the form of pure DRAM.

Simple to Install, Ready to Scale

MEXT installs in minutes, via a simple package that includes a small telemetry kernel module and the MEXT AI engine.

Take a look at this quick demo. Upon installing MEXT, the system’s available memory immediately doubles—from close to 512GB to close to 1TB. And it takes all of 2 seconds to execute.