Scaling Enterprise AI with Determined and NVIDIA DGX
Determined’s deep learning training platform helps teams train deep learning models in minutes and hours, not days and weeks. Through techniques like advanced hyperparameter tuning, state-of-the-art distributed training, and intelligent cluster management, Determined lets model developers train models instead of fretting over ML infrastructure.
Teams that leverage NVIDIA DGX systems have already made a huge investment in AI. But, effectively leveraging all of the world class compute, networking, and storage provided by their DGX system remains elusive.
By speeding up model training, Determined allows these teams to increase their overall ROI in deep learning by leveraging tools like smart cluster utilization through AI-native job scheduling, out-of-the-box distributed training, and intuitive experiment tracking. Since Determined is already NVIDIA Validated and in the NGC catalog, getting started has never been easier.
Deep Learning in hours and minutes, not days and weeks.
Ship models faster and do more science!
Our open-source platform - used in conjunction with DGX systems - enables you to train models in minutes and hours, not days and weeks.
Focus on research and models
Instead of arduous tasks like manual hyperparameter tuning, re-running faulty jobs, and worrying about hardware resources.
Distributed training as it was meant to be
Our distributed training implementation outperforms the industry standard, requires no code changes, and is fully integrated with our state of the art training platform.
Share resources, experiments, and data
— and do it safely.
Real-time experiment dashboard
With built-in experiment tracking and visualization, Determined records metrics automatically, makes your ML projects reproducible, and allows your team to collaborate more easily.
Sophisticated checkpointing and resource scheduling
Your researchers will be able to build on the progress of their team and innovate in their domain, instead of fretting over errors and infrastructure.
Expert built for expert teams
We’re dedicated to the science of Deep Learning and are leaders in productionizing DL workflows, and we can’t wait to help your team thrive.
Determined is run at scale on DGX Systems and can take advantage of the following unique capabilities:
Seamless scale: RoCE/IB-optimized multi-node workloads out-of-the-box
Tight integration into the NVIDIA software ecosystem: NCCL, NGC Containers, MIG
Fully containerized: deploys on Kubernetes or bare metal on DGX systems
Trusted: validated by NVIDIA for GPU performance and compatibility
Reengineering drug discovery with Determined
Recursion Pharmaceuticals uses Determined to train more models faster, which enables their drug discovery teams to collaborate and iterate quickly by trying out new ideas multiple times a day. Notably, Determined’s deep learning training platform is installed on hundreds of NVIDIA GPUs to power drug discovery. Recursion’s ML team sped up hyperparameter search by 14x and brought model training time down from three days (on average) to three hours on their on-premise GPUs.
The easiest way to get started with Determined is to visit the NGC Catalog and install the Determined Helm chart in your cluster.
