• Docs >
  • Accelerator: HPU training

Accelerator: HPU training

Audience: Users looking to save money and run large models faster using single or multiple Gaudi devices.

What is an HPU?

Habana® Gaudi® AI Processor (HPU) training processors are built on a heterogeneous architecture with a cluster of fully programmable Tensor Processing Cores (TPC) along with its associated development tools and libraries, and a configurable Matrix Math engine.

The TPC core is a VLIW SIMD processor with an instruction set and hardware tailored to serve training workloads efficiently. The Gaudi memory architecture includes on-die SRAM and local memories in each TPC and, Gaudi is the first DL training processor that has integrated RDMA over Converged Ethernet (RoCE v2) engines on-chip.

On the software side, the PyTorch Habana bridge interfaces between the framework and SynapseAI software stack to enable the execution of deep learning models on the Habana Gaudi device.

Gaudi offers a substantial price/performance advantage – so you get to do more deep learning training while spending less.

For more information, check out Gaudi Architecture and Gaudi Developer Docs.

Run on 1 Gaudi

To enable PyTorch Lightning to utilize the HPU accelerator, simply provide accelerator="hpu" parameter to the Trainer class.

trainer = Trainer(accelerator="hpu", devices=1)

Run on multiple Gaudis

The devices=8 and accelerator="hpu" parameters to the Trainer class enables the Habana accelerator for distributed training with 8 Gaudis. It uses HPUParallelStrategy internally which is based on DDP strategy with the addition of Habana’s collective communication library (HCCL) to support scale-up within a node and scale-out across multiple nodes.

trainer = Trainer(devices=8, accelerator="hpu")

Scale-out on Gaudis

To train a Lightning model using multiple HPU nodes, set the num_nodes parameter with the available nodes in the Trainer class.

trainer = Trainer(accelerator="hpu", devices=8, strategy="hpu_parallel", num_nodes=2)

In addition to this, the following environment variables need to be set to establish communication across nodes. Check out the documentation on Cluster Environment for more details.

  • MASTER_PORT - required; has to be a free port on machine with NODE_RANK 0

  • MASTER_ADDR - required (except for NODE_RANK 0); address of NODE_RANK 0 node

  • WORLD_SIZE - required; how many workers are in the cluster

  • NODE_RANK - required; id of the node in the cluster

The trainer needs to be instantiated on every node participating in the training.

On Node 1:

    python -m some_model_trainer.py (--arg1 ... train script args...)

On Node 2:

    python -m some_model_trainer.py (--arg1 ... train script args...)

Select Gaudis automatically

Lightning can automatically detect the number of Gaudi devices to run on. This setting is enabled by default if the devices argument is missing.

# equivalent
trainer = Trainer(accelerator="hpu")
trainer = Trainer(accelerator="hpu", devices="auto")

How to access HPUs

To use HPUs, you must have access to a system with HPU devices.

Known limitations

© Copyright Copyright (c) 2018-2023, Lightning AI et al...

Built with Sphinx using a theme provided by Read the Docs.