Event

<p>Serving multiple models on a single GPU sounds great until something segfaults.</p> <p>Two approaches dominate for parallel inference: MIG (hardware partitioning) and MPS (software sharing). Both promise efficient GPU sharing.</p> <p>I tested both strategies for video generation workloads in parallel.</p> <p>This talk digs into what actually happened: where things worked, where memory isolation fell apart, which configs crashed, and what survives under load.</p> <p>By the end, you'll know:</p> <ol> <li>How to utilize unused GPU capacity.</li> <li>How to setup MIG and MPS.</li> <li>Memory issues, crashes, and failures.</li> <li>Workload specific configs</li> </ol>