> This page is for version Release 3.0 (default). > For other versions, use one of these documentation indexes: > - Release 3.0 (default): https://docs.nvidia.com/isaaclab/3.0/llms.txt > - Release 2.3: https://docs.nvidia.com/isaaclab/2.3/llms.txt > - Release 2.2: https://docs.nvidia.com/isaaclab/2.2/llms.txt > - Release 2.1: https://docs.nvidia.com/isaaclab/2.1/llms.txt > For clean Markdown of any page, append .md to the page URL. > For a complete documentation index, see https://docs.nvidia.com/isaaclab/llms.txt. > For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.nvidia.com/isaaclab/_mcp/server. # Multi-GPU and Multi-Node Training Scale one reinforcement learning job across the GPUs in a workstation or across several nodes with `train_multigpu`. Isaac Lab starts one training process per GPU, gives each process its own simulation environments, and synchronizes policy updates across the processes. The same launcher model is used in three multi-GPU benchmarks for measuring startup, simulation, and end-to-end training performance. > **Warning** > > Multi-GPU and multi-node training requires Linux and NVIDIA NCCL. Windows is > not supported. > **Warning** > > `train_multigpu` is experimental and may change in a future release. ## Start a multi-GPU training run First, verify that the task trains on one GPU. This separates task or configuration problems from distributed-launch problems: ```bash uv run isaaclab train --task Isaac-Cartpole ``` Then run the same task on every visible GPU: ```bash uv run isaaclab train_multigpu --task Isaac-Cartpole ``` That is the complete transition from a single-GPU run. `train_multigpu` adds `--distributed` and selects the distributed launcher automatically. All other arguments are the same arguments accepted by `train`: ```bash uv run isaaclab train_multigpu \ --task Isaac-Reorient-KukaAllegro \ --num_envs 4096 \ --max_iterations 100 ``` `--num_envs` is the number of environments **on each GPU**. With four GPUs and `--num_envs 4096`, the job collects experience from 16,384 environments in total. All ranks write into one run folder. Each rank trains with the launch seed plus its rank on its own GPU. Rank 0 writes its settings (`params/env.yaml` and `params/agent.yaml`) and videos where a single-GPU run does; every other rank writes its own `params/` and videos to `rank_/`. > **Tip** > > Add `--dry_run` to print the resolved launcher command without starting > training. This is useful when checking GPU counts, rendezvous options, or > forwarded training arguments. ### Choose the GPUs By default, the launcher uses every visible GPU. Set `--num_gpus` when you want a specific worker count: ```bash uv run isaaclab train_multigpu --num_gpus 2 --task Isaac-Cartpole ``` Use `CUDA_VISIBLE_DEVICES` to choose the physical devices. The launcher sees only the devices in that list: ```bash CUDA_VISIBLE_DEVICES=1,3 uv run isaaclab train_multigpu \ --num_gpus 2 --task Isaac-Cartpole ``` Run `nvidia-smi` before launching to confirm that the expected devices are available and have enough free memory. ### Choose an RL library Multi-GPU training supports RSL-RL, RL-Games, and skrl. RSL-RL is the default for most core tasks. | Library | Distributed backend | Command | | -------- | ------------------- | ------------------------------------------------------------------------------------------ | | RSL-RL | PyTorch | `uv run isaaclab train_multigpu --rl_library rsl_rl ...` | | RL-Games | PyTorch | `uv run --extra rl-games isaaclab train_multigpu --rl_library rl_games ...` | | skrl | PyTorch | `uv run --extra skrl isaaclab train_multigpu --rl_library skrl ...` | | skrl | JAX | `uv run --extra skrl-jax isaaclab train_multigpu --rl_library skrl --ml_framework jax ...` | skrl with JAX uses skrl's distributed launcher instead of `torchrun`. Pass an integer `--num_gpus` and use `--coordinator_address` to configure its coordinator: ```bash uv run --extra skrl-jax isaaclab train_multigpu \ --rl_library skrl --ml_framework jax --num_gpus 4 \ --coordinator_address localhost:5000 \ --task Isaac-Cartpole ``` ## Measure scaling with the three multi-GPU benchmarks Use the benchmark commands when the goal is to measure performance rather than train a policy for later use. Isaac Lab provides three multi-GPU workflows: | Benchmark | What it measures | Use it to answer | | ------------------- | -------------------------------------------------------------------- | ----------------------------------------------------- | | `startup_multigpu` | Startup time for rank 0 while every GPU starts the same workload. | How does a fully occupied node affect startup? | | `runtime_multigpu` | Simulation throughput for rank 0 while every GPU runs independently. | How does host contention affect environment stepping? | | `training_multigpu` | Global, synchronized training throughput across every rank. | How well does end-to-end learning scale across GPUs? | Run each benchmark with the same launcher options used by `train_multigpu`: ```bash uv run isaaclab benchmark startup_multigpu \ --num_gpus 2 --task Isaac-Cartpole uv run isaaclab benchmark runtime_multigpu \ --num_gpus 2 --task Isaac-Cartpole --num_envs 4096 uv run isaaclab benchmark training_multigpu \ --rl_library rsl_rl --num_gpus 2 \ --task Isaac-Cartpole --num_envs 4096 --max_iterations 100 ``` `training_multigpu` supports RSL-RL, RL-Games, and skrl with PyTorch. It does not support skrl with JAX or SB3. It also rejects `--video` and `--capture_env_sensors`, which do not produce a meaningful aggregate result across ranks. `--check_success` is supported with RSL-RL and RL-Games: the success metric is averaged over the environments of all ranks, so every rank stops at the same iteration. For multi-node benchmarks, pass the same `--nnodes`, `--node_rank`, and rendezvous options described in [Train across multiple nodes](/source/features/multi_gpu#train-across-multiple-nodes) on every node. ### Read multi-GPU benchmark results Only global rank 0 writes a result bundle. The `extra` fields record the rank layout and clarify which measurements the bundle covers: | `extra` field | Meaning | | --------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | `world_size`, `local_world_size`, `num_nodes` | Rank layout of the job. | | `num_envs_per_rank` | Environments hosted by each rank. | | `workload_scope` | `global` for `training_multigpu` because ranks train in lockstep. `rank0` for `startup_multigpu` and `runtime_multigpu` because each rank runs independently and only rank 0 is measured. | | `measurement_scope` | `rank0_process`: timings, learning curves, CPU, and RAM come from rank 0 alone. | | `gpu_measurement_scope` | `rank0_node`: `resources.devices` reports every GPU visible to rank 0. GPU utilization and memory values remain scoped to rank 0's device. | When comparing one GPU with multiple GPUs, keep `--num_envs` constant **per rank**. An N-GPU job processes N times as many environments as the one-GPU job at the same per-rank setting. Compare the global throughput of `training_multigpu` against N times the single-GPU throughput. Do not treat `startup_multigpu` or `runtime_multigpu` as aggregate rates; they measure rank 0 while the other ranks create host contention. ## How multi-GPU training works For PyTorch workflows, `train_multigpu` wraps [torchrun](https://docs.pytorch.org/docs/stable/elastic/run.html). It launches one process per GPU. Each process owns: * one Isaac Lab application and its vectorized environments, * one copy of the policy, * one rollout buffer, and * one GPU selected by its local rank. The processes collect experience independently and synchronize gradients during policy updates with [DistributedDataParallel](https://docs.pytorch.org/docs/stable/notes/ddp.html). Simulation does not move between GPUs, so available host CPU, RAM, and I/O can become limiting factors as the GPU count grows. ![One training process and simulation workload per GPU](/isaaclab/_fern-files/isaaclab.docs.buildwithfern.com/d31682a343628c4ec7cc419f3f1ef703a43fdfb3dab73d46b90243a098b63573/assets/releases/3.0/static/multi-gpu-rl/a3c-light.svg) ![One training process and simulation workload per GPU](/isaaclab/_fern-files/isaaclab.docs.buildwithfern.com/5a629c3cc8a42acd831ce86952cdbdbdd61722d07c6fd76e266f6cb7deccb568/assets/releases/3.0/static/multi-gpu-rl/a3c-dark.svg) ### Read logs from distributed runs Every rank produces similar startup messages, warnings, and model summaries. The launcher shows local rank 0 by default so the console remains readable. Training metrics already come from global rank 0, and a crash on any hidden rank still reports the failing rank and its traceback. Show output from every rank when processes appear to disagree: ```bash uv run isaaclab train_multigpu \ --task Isaac-Cartpole --log_all_ranks ``` For a clean console and complete per-rank logs on disk, use `torchrun` log redirection: ```bash uv run isaaclab train_multigpu \ --task Isaac-Cartpole --tee 3 --log_dir /tmp/isaaclab-rank-logs ``` The log filtering options apply to PyTorch workflows. skrl with JAX writes every rank to the console. ## Train across multiple nodes Every node must have the same Isaac Lab checkout, dependencies, task configuration, and access to training assets. The nodes must also be able to reach one another on the rendezvous port. Choose one node as the rendezvous host. For a two-node PyTorch job with four GPUs per node, run the following on the first node: ```bash uv run isaaclab train_multigpu \ --nnodes 2 --node_rank 0 --num_gpus 4 \ --master_addr 10.0.0.10 --master_port 29500 \ --task Isaac-Cartpole ``` Run the same command on the second node with its own rank: ```bash uv run isaaclab train_multigpu \ --nnodes 2 --node_rank 1 --num_gpus 4 \ --master_addr 10.0.0.10 --master_port 29500 \ --task Isaac-Cartpole ``` The total world size is `nnodes * num_gpus`: eight ranks in this example. You can also use `--rdzv_backend`, `--rdzv_endpoint`, and `--rdzv_id` for an elastic `torchrun` rendezvous. Add `--dry_run` first to verify the command on each node. Each node names the run folder from its own launch time. When the nodes share storage, pass the same `--run_timestamp` value (format `YYYY-MM-DD_HH-MM-SS`) to `train_multigpu` on every node so that all ranks write into one run folder. For skrl with JAX, pass `--nnodes`, `--node_rank`, an integer `--num_gpus`, and the same `--coordinator_address` on every node. Do not pass the PyTorch rendezvous options to a JAX launch. Multi-node scaling depends heavily on the network between nodes. A multi-node job can be slower than a single-node job when gradient synchronization dominates the training iteration. ## Troubleshoot distributed training Start with the smallest useful diagnosis: 1. Confirm that the same task, backend, and training arguments work with `isaaclab train` on one GPU. 2. Add `--dry_run` and check the selected GPU and node counts. 3. Retry at world sizes 2, 3, and 4. A failure at only one world size often points to the communication transport rather than the task. 4. Check GPU placement and interconnects with `nvidia-smi topo -m`. 5. Set `NCCL_DEBUG=INFO` to see which NCCL transport was selected. 6. Apply one workaround at a time and verify that the failure returns when the workaround is removed. ### NCCL hangs and errors A run that stops without a traceback while every participating GPU remains at 100% utilization is usually stalled in an NCCL collective. The following workarounds address known system-specific transport problems: | Symptom | Try | | -------------------------------------------------------------------------------------- | ----------------------------------------------------------------- | | World size 2 hangs on a PCIe system without NVLink. | `NCCL_P2P_DISABLE=1` or `NCCL_P2P_LEVEL=LOC` | | `illegal memory access` appears in `ProcessGroupNCCL`. | `NCCL_SHM_DISABLE=1` | | A rendered job fails because CUDA and the renderer enumerate GPUs in different orders. | `CUDA_DEVICE_ORDER=PCI_BUS_ID` | | A rendered job times out across NUMA nodes during `BROADCAST` or `ALLREDUCE`. | `NCCL_CUMEM_HOST_ENABLE=0`; if needed, try `NCCL_CUMEM_ENABLE=0`. | | Communicator initialization or transport failures persist. | `NCCL_IB_DISABLE=1` or `NCCL_ALGO=Ring`. | For example, test the first workaround without changing a shared configuration: ```bash NCCL_P2P_DISABLE=1 uv run isaaclab train_multigpu \ --num_gpus 2 --task Isaac-Cartpole ``` These variables can reduce communication performance and should be scoped to the affected machine. Set either `NCCL_P2P_DISABLE=1` or `NCCL_P2P_LEVEL=LOC`, not both. Each prevents direct P2P communication and can reduce bandwidth by forcing NCCL to select another transport. Do not commit a workaround into a task or launcher unless it is required by every supported system. Use `nvidia-smi --query-gpu=name,pci.bus_id` to inspect GPU bus IDs before setting `CUDA_DEVICE_ORDER=PCI_BUS_ID`. Run a minimal NCCL collective at the world size that hangs. Save this as `nccl_probe.py`: ```python import os import torch import torch.distributed as dist local_rank = int(os.environ["LOCAL_RANK"]) torch.cuda.set_device(local_rank) dist.init_process_group("nccl") tensor = torch.ones(1024, device=f"cuda:{local_rank}") dist.all_reduce(tensor) torch.cuda.synchronize() print(f"rank {dist.get_rank()} ok", flush=True) dist.destroy_process_group() ``` Launch the probe with the same rank count: ```bash uv run python -m torch.distributed.run --nproc_per_node 2 nccl_probe.py ``` If this probe also hangs, the problem is in NCCL or the system topology, not in Isaac Lab, the task, or the RL library.