Samsung has introduced a new type of memory called LPDDR5X-PIM, which builds processing directly into the memory chip rather than relying entirely on a separate processor. The approach, known as processing-in-memory, is meant to cut the time and energy spent shuttling data back and forth during demanding computing tasks.
In tests using a large language model with about 8 billion parameters, the company reported that the technology ran roughly 2.3 times faster and delivered around three times the token throughput compared with conventional memory. Samsung said the design also offers a large jump in effective bandwidth.
The bottleneck the technology targets is a familiar one in computing, where moving data between memory and the processor can slow everything down and drain power. By handling some calculations inside the memory itself, the chip aims to ease that constraint for artificial intelligence workloads.
Mobile devices and other power-sensitive systems could stand to benefit, since running AI models locally often strains both speed and battery life. Samsung positioned the product as a way to make on-device AI more practical without leaning as heavily on cloud servers.
The announcement adds to a broader industry push to rethink how memory and processing work together as AI demands grow. Whether processing-in-memory becomes widely adopted will depend on how it fits into existing hardware and how software is written to take advantage of it.
