DGMY has moved digital human production onto the high-performance cluster. Real-time characters, voice and scene updates share the pool that also runs large-model training, instead of waiting on a separate render farm.

One pool, two kinds of work

Training jobs take the long reservations. Interactive digital human sessions take shorter slices when a visitor or a broadcast needs a live character. The scheduler keeps those slices from cutting into a checkpoint.

Closer to the audience

Because the character is rendered next to the rest of the compute, a scene change does not have to cross a second data center. That matters for kiosks, training rooms and live presentations where the delay is obvious.

The character stack is covered in the Digital Human solution, and the shared pool in High-Performance Computing (HPC).