## MEXT Performance Benchmarking

### How MEXT AI-Powered Predictive Memory™ Works

- MEXT continually offloads cold memory pages from DRAM to a 50x lower-cost tier, flash.
- MEXT's AI engine tracks related pages and predicts which pages in flash are likely to get requested by the application.
- MEXT transparently pushes those pages back into DRAM—even before they are requested by the application.
- From the application's point of view, all relevant pages can be found in DRAM—maintaining performance within a smaller DRAM footprint.

### MEXT Performance vs. All DRAM System

With MEXT, customers can achieve comparable application performance using 50% of the DRAM of their original configuration. For instance, when comparing 64 GB DRAM system to a 32 GB DRAM + MEXT system, performance for Redis hits 94% of the all-DRAM configuration, Neo4j hits 95%, and Memcached hits 96%.

- Redis (32 GB DRAM / 32 GB MEXT) 94% Memtier Benchmark
- Neo4j (32 GB DRAM / 32 GB MEXT) 95% WDBench Benchmark
- Memcached (32 GB DRAM / 32 GB MEXT) 96% Memtier Benchmark
- All DRAM (64 GB) 100%

### Performance / $ with MEXT

If we look at the amount of performance achieved per dollar spent, we see that leveraging MEXT yields significant efficiencies.

For instance, if we take the Performance / $ of an all-DRAM system to be a reference level of X, MEXT enables a Performance / $ of 1.7X for Redis, 1.7X for Neo4j, and 1.7X for Memcached.

- All DRAM: 1X
- Redis: 1.7X
- Neo4j: 1.7X
- Memcached: 1.7X