AI · SEMICONDUCTOR · FACT CHECK
SKD WIRETrue
NVIDIA has introduced a sharing technology that allows multiple AI workloads to divide a single high-end GPU among themselves, with control extending to the GPU's memory, according to materials published on the company's developer blog and official website.
The technology, published through NVIDIA's official developer channels including its developer blog and the Multi-Instance GPU page on its Korean website, enables a single expensive GPU to be partitioned for use by several AI workloads simultaneously.
Alongside the compute partitioning, the announcement covers memory management capabilities — described in developer blog materials on CUDA Memory and KV cache handling — allowing control over memory resources when the GPU is shared.
The materials were distributed via NVIDIA's developer blog, with accompanying images hosted on the company's official domains, including the Multi-Instance GPU technology page on NVIDIA's Korean site.
NVIDIA's move reflects a broader push to maximize utilization of costly AI hardware by letting one accelerator serve multiple workloads rather than dedicating a full GPU to each.
The claim matches the company's official first-party announcement and source materials. The verdict is True.
