GTC 2020: Accelerating Hyperparameter Tuning with Container-Level GPU Virtualization
After clicking “Watch Now” you will be prompted to login or join.
Click “Watch Now” to login or join the NVIDIA Developer Program.
Accelerating Hyperparameter Tuning with Container-Level GPU Virtualization
Jeongkyu Shin, Lablup Inc. | Joongi Kim, Lablup Inc.
Many people think that hyperparameter tuning requires a large number of GPUs to get optimal results quickly. That's generally true, but to what extent? We'll present what we've learned about finding a sweet spot to balance both costs and accuracy by exploiting partitioned GPUs with Backend.AI's container-level GPU virtualization. Our benchmark includes distributed mnist, cifar10 transfer learning and TGS salt identification cases using AutoML with network morphism and ENAS tuner with NNI running on NGC-optimized containers on Backend.AI. You'll get a tangible guide to plan and deploy your GPU infrastructure capacity in a more cost-effective way.