NVIDIA introduced TensorRT multi-device inference integration in Dynamo-Triton, enabling model serving across multiple GPUs rather than a single GPU.

NVIDIA’s developer announcement introduces TensorRT multi-device inference integration in NVIDIA Dynamo-Triton. The stated capability is model serving across multiple GPUs.

The supplied excerpt frames the feature around generative AI workloads whose compute and memory requirements increasingly exceed what one GPU can provide. This is a software-serving capability, rather than an announcement of a new GPU, server, or installed cluster.

The excerpt does not specify supported GPU counts, hardware generations, model families, interconnect requirements, or deployment topology. It also provides no throughput, latency, or cost benchmark, so the announcement does not establish a quantified performance improvement.