AMD said on June 15 it has acquired data centre memory optimisation startup Mext.

According to SiliconANGLE, AMD plans to respond to AI data centre memory supply constraints and raise customer system efficiency to lower operating costs.

Financial terms were not disclosed. AMD stressed that Mext’s technology can speed deployment of large-scale AI workloads and improve utilisation of existing infrastructure.

Mext specialises in “AI-native memory tiering” technology that moves infrequently used data out of expensive DRAM and stores it in NAND flash storage. Flash storage costs are dozens of times cheaper than general memory. A key element of Mext’s technology is its predictive memory engine.

The predictive memory engine continuously analyses memory access patterns and preloads data expected to be used soon from flash into DRAM, enabling access without delays.

AMD sees Mext’s technology as contributing to solving data centre efficiency issues and improving performance. Through the technology, it expects usable memory capacity on existing infrastructure to increase and the need to buy additional DRAM to fall.

AMD plans to integrate Mext’s technology across its data centre portfolio. Dan McNamara (댄 맥나마라), senior vice president of AMD’s Compute and Enterprise unit, said, “We will integrate Mext’s technology across AMD’s data centre portfolio to help enterprise customers extract greater value from infrastructure investment and accelerate AI deployment.”