{ "hf_id": "nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", "meta": { "title": "NVIDIA Nemotron-3-Ultra-550B-A55B", "slug": "nemotron-3-ultra-550b-a55b", "provider": "NVIDIA", "description": "NVIDIA Nemotron 3 Ultra hybrid Transformer-Mamba MoE model for long-context agentic reasoning, coding, and tool use.", "date_added": "2026-06-04", "date_updated": "2026-09-10", "tasks": [ "text" ], "performance_headline": "550B total / 55B active parameters with BF16 and NVFP4 serving paths", "related_recipes": [ "nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16" ], "hardware": { "b200": "verified" } }, "recommended_command": { "hardware": "h200", "strategy": "multi_node_tp", "variant": "default", "node_count": 2, "deploy_type": "multi_node", "env": { "VLLM_MEMORY_PROFILER_ESTIMATE_CUDAGRAPHS": "1", "GLOO_SOCKET_IFNAME": "$IFACE_NAME", "NCCL_SOCKET_IFNAME": "$IFACE_NAME" }, "docker_image": "vllm/vllm-openai:v0.28.0", "head_command": "vllm serve nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16 \\\n --served-model-name nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B \\\n --trust-remote-code \\\n --kv-cache-dtype fp8 \\\n --max-num-seqs 16 \\\n --max-model-len 262144 \\\n --gpu-memory-utilization 0.90 \\\n --max-num-batched-tokens 32768 \\\n --enable-flashinfer-autotune \\\n --async-scheduling \\\n --mamba-backend triton \\\n --mamba-ssm-cache-dtype float32 \\\n --tensor-parallel-size 16 \\\n --nnodes 2 \\\n --node-rank 0 \\\n --master-addr $HEAD_IP \\\n --enable-auto-tool-choice \\\n --tool-call-parser qwen3_coder \\\n --reasoning-parser nemotron_v3 \\\n --speculative-config '{\"method\":\"mtp\",\"num_speculative_tokens\":5}'", "worker_command": "vllm serve nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16 \\\n --served-model-name nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B \\\n --trust-remote-code \\\n --kv-cache-dtype fp8 \\\n --max-num-seqs 16 \\\n --max-model-len 262144 \\\n --gpu-memory-utilization 0.90 \\\n --max-num-batched-tokens 32768 \\\n --enable-flashinfer-autotune \\\n --async-scheduling \\\n --mamba-backend triton \\\n --mamba-ssm-cache-dtype float32 \\\n --tensor-parallel-size 16 \\\n --nnodes 2 \\\n --node-rank 1 \\\n --master-addr $HEAD_IP \\\n --headless \\\n --enable-auto-tool-choice \\\n --tool-call-parser qwen3_coder \\\n --reasoning-parser nemotron_v3 \\\n --speculative-config '{\"method\":\"mtp\",\"num_speculative_tokens\":5}'", "worker_commands": [ "vllm serve nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16 \\\n --served-model-name nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B \\\n --trust-remote-code \\\n --kv-cache-dtype fp8 \\\n --max-num-seqs 16 \\\n --max-model-len 262144 \\\n --gpu-memory-utilization 0.90 \\\n --max-num-batched-tokens 32768 \\\n --enable-flashinfer-autotune \\\n --async-scheduling \\\n --mamba-backend triton \\\n --mamba-ssm-cache-dtype float32 \\\n --tensor-parallel-size 16 \\\n --nnodes 2 \\\n --node-rank 1 \\\n --master-addr $HEAD_IP \\\n --headless \\\n --enable-auto-tool-choice \\\n --tool-call-parser qwen3_coder \\\n --reasoning-parser nemotron_v3 \\\n --speculative-config '{\"method\":\"mtp\",\"num_speculative_tokens\":5}'" ], "head_argv": [ "vllm", "serve", "nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", "--served-model-name", "nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B", "--trust-remote-code", "--kv-cache-dtype", "fp8", "--max-num-seqs", "16", "--max-model-len", "262144", "--gpu-memory-utilization", "0.90", "--max-num-batched-tokens", "32768", "--enable-flashinfer-autotune", "--async-scheduling", "--mamba-backend", "triton", "--mamba-ssm-cache-dtype", "float32", "--tensor-parallel-size", "16", "--nnodes", "2", "--node-rank", "0", "--master-addr", "$HEAD_IP", "--enable-auto-tool-choice", "--tool-call-parser", "qwen3_coder", "--reasoning-parser", "nemotron_v3", "--speculative-config", "{\"method\":\"mtp\",\"num_speculative_tokens\":5}" ], "worker_argv": [ "vllm", "serve", "nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", "--served-model-name", "nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B", "--trust-remote-code", "--kv-cache-dtype", "fp8", "--max-num-seqs", "16", "--max-model-len", "262144", "--gpu-memory-utilization", "0.90", "--max-num-batched-tokens", "32768", "--enable-flashinfer-autotune", "--async-scheduling", "--mamba-backend", "triton", "--mamba-ssm-cache-dtype", "float32", "--tensor-parallel-size", "16", "--nnodes", "2", "--node-rank", "1", "--master-addr", "$HEAD_IP", "--headless", "--enable-auto-tool-choice", "--tool-call-parser", "qwen3_coder", "--reasoning-parser", "nemotron_v3", "--speculative-config", "{\"method\":\"mtp\",\"num_speculative_tokens\":5}" ], "worker_argvs": [ [ "vllm", "serve", "nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", "--served-model-name", "nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B", "--trust-remote-code", "--kv-cache-dtype", "fp8", "--max-num-seqs", "16", "--max-model-len", "262144", "--gpu-memory-utilization", "0.90", "--max-num-batched-tokens", "32768", "--enable-flashinfer-autotune", "--async-scheduling", "--mamba-backend", "triton", "--mamba-ssm-cache-dtype", "float32", "--tensor-parallel-size", "16", "--nnodes", "2", "--node-rank", "1", "--master-addr", "$HEAD_IP", "--headless", "--enable-auto-tool-choice", "--tool-call-parser", "qwen3_coder", "--reasoning-parser", "nemotron_v3", "--speculative-config", "{\"method\":\"mtp\",\"num_speculative_tokens\":5}" ] ], "head_docker_command": "docker run --gpus all \\\n --privileged --ipc=host -p 8000:8000 \\\n -v ~/.cache/huggingface:/root/.cache/huggingface \\\n -e VLLM_MEMORY_PROFILER_ESTIMATE_CUDAGRAPHS=1 \\\n -e GLOO_SOCKET_IFNAME=$IFACE_NAME \\\n -e NCCL_SOCKET_IFNAME=$IFACE_NAME \\\n vllm/vllm-openai:v0.28.0 nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16 \\\n --served-model-name nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B \\\n --trust-remote-code \\\n --kv-cache-dtype fp8 \\\n --max-num-seqs 16 \\\n --max-model-len 262144 \\\n --gpu-memory-utilization 0.90 \\\n --max-num-batched-tokens 32768 \\\n --enable-flashinfer-autotune \\\n --async-scheduling \\\n --mamba-backend triton \\\n --mamba-ssm-cache-dtype float32 \\\n --tensor-parallel-size 16 \\\n --nnodes 2 \\\n --node-rank 0 \\\n --master-addr $HEAD_IP \\\n --enable-auto-tool-choice \\\n --tool-call-parser qwen3_coder \\\n --reasoning-parser nemotron_v3 \\\n --speculative-config '{\"method\":\"mtp\",\"num_speculative_tokens\":5}'", "worker_docker_command": "docker run --gpus all \\\n --privileged --ipc=host -p 8000:8000 \\\n -v ~/.cache/huggingface:/root/.cache/huggingface \\\n -e VLLM_MEMORY_PROFILER_ESTIMATE_CUDAGRAPHS=1 \\\n -e GLOO_SOCKET_IFNAME=$IFACE_NAME \\\n -e NCCL_SOCKET_IFNAME=$IFACE_NAME \\\n vllm/vllm-openai:v0.28.0 nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16 \\\n --served-model-name nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B \\\n --trust-remote-code \\\n --kv-cache-dtype fp8 \\\n --max-num-seqs 16 \\\n --max-model-len 262144 \\\n --gpu-memory-utilization 0.90 \\\n --max-num-batched-tokens 32768 \\\n --enable-flashinfer-autotune \\\n --async-scheduling \\\n --mamba-backend triton \\\n --mamba-ssm-cache-dtype float32 \\\n --tensor-parallel-size 16 \\\n --nnodes 2 \\\n --node-rank 1 \\\n --master-addr $HEAD_IP \\\n --headless \\\n --enable-auto-tool-choice \\\n --tool-call-parser qwen3_coder \\\n --reasoning-parser nemotron_v3 \\\n --speculative-config '{\"method\":\"mtp\",\"num_speculative_tokens\":5}'", "worker_docker_commands": [ "docker run --gpus all \\\n --privileged --ipc=host -p 8000:8000 \\\n -v ~/.cache/huggingface:/root/.cache/huggingface \\\n -e VLLM_MEMORY_PROFILER_ESTIMATE_CUDAGRAPHS=1 \\\n -e GLOO_SOCKET_IFNAME=$IFACE_NAME \\\n -e NCCL_SOCKET_IFNAME=$IFACE_NAME \\\n vllm/vllm-openai:v0.28.0 nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16 \\\n --served-model-name nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B \\\n --trust-remote-code \\\n --kv-cache-dtype fp8 \\\n --max-num-seqs 16 \\\n --max-model-len 262144 \\\n --gpu-memory-utilization 0.90 \\\n --max-num-batched-tokens 32768 \\\n --enable-flashinfer-autotune \\\n --async-scheduling \\\n --mamba-backend triton \\\n --mamba-ssm-cache-dtype float32 \\\n --tensor-parallel-size 16 \\\n --nnodes 2 \\\n --node-rank 1 \\\n --master-addr $HEAD_IP \\\n --headless \\\n --enable-auto-tool-choice \\\n --tool-call-parser qwen3_coder \\\n --reasoning-parser nemotron_v3 \\\n --speculative-config '{\"method\":\"mtp\",\"num_speculative_tokens\":5}'" ], "head_docker_argv": [ "docker", "run", "--gpus", "all", "--privileged", "--ipc=host", "-p", "8000:8000", "-v", "~/.cache/huggingface:/root/.cache/huggingface", "-e", "VLLM_MEMORY_PROFILER_ESTIMATE_CUDAGRAPHS=1", "-e", "GLOO_SOCKET_IFNAME=$IFACE_NAME", "-e", "NCCL_SOCKET_IFNAME=$IFACE_NAME", "vllm/vllm-openai:v0.28.0", "nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", "--served-model-name", "nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B", "--trust-remote-code", "--kv-cache-dtype", "fp8", "--max-num-seqs", "16", "--max-model-len", "262144", "--gpu-memory-utilization", "0.90", "--max-num-batched-tokens", "32768", "--enable-flashinfer-autotune", "--async-scheduling", "--mamba-backend", "triton", "--mamba-ssm-cache-dtype", "float32", "--tensor-parallel-size", "16", "--nnodes", "2", "--node-rank", "0", "--master-addr", "$HEAD_IP", "--enable-auto-tool-choice", "--tool-call-parser", "qwen3_coder", "--reasoning-parser", "nemotron_v3", "--speculative-config", "{\"method\":\"mtp\",\"num_speculative_tokens\":5}" ], "worker_docker_argv": [ "docker", "run", "--gpus", "all", "--privileged", "--ipc=host", "-p", "8000:8000", "-v", "~/.cache/huggingface:/root/.cache/huggingface", "-e", "VLLM_MEMORY_PROFILER_ESTIMATE_CUDAGRAPHS=1", "-e", "GLOO_SOCKET_IFNAME=$IFACE_NAME", "-e", "NCCL_SOCKET_IFNAME=$IFACE_NAME", "vllm/vllm-openai:v0.28.0", "nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", "--served-model-name", "nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B", "--trust-remote-code", "--kv-cache-dtype", "fp8", "--max-num-seqs", "16", "--max-model-len", "262144", "--gpu-memory-utilization", "0.90", "--max-num-batched-tokens", "32768", "--enable-flashinfer-autotune", "--async-scheduling", "--mamba-backend", "triton", "--mamba-ssm-cache-dtype", "float32", "--tensor-parallel-size", "16", "--nnodes", "2", "--node-rank", "1", "--master-addr", "$HEAD_IP", "--headless", "--enable-auto-tool-choice", "--tool-call-parser", "qwen3_coder", "--reasoning-parser", "nemotron_v3", "--speculative-config", "{\"method\":\"mtp\",\"num_speculative_tokens\":5}" ], "worker_docker_argvs": [ [ "docker", "run", "--gpus", "all", "--privileged", "--ipc=host", "-p", "8000:8000", "-v", "~/.cache/huggingface:/root/.cache/huggingface", "-e", "VLLM_MEMORY_PROFILER_ESTIMATE_CUDAGRAPHS=1", "-e", "GLOO_SOCKET_IFNAME=$IFACE_NAME", "-e", "NCCL_SOCKET_IFNAME=$IFACE_NAME", "vllm/vllm-openai:v0.28.0", "nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", "--served-model-name", "nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B", "--trust-remote-code", "--kv-cache-dtype", "fp8", "--max-num-seqs", "16", "--max-model-len", "262144", "--gpu-memory-utilization", "0.90", "--max-num-batched-tokens", "32768", "--enable-flashinfer-autotune", "--async-scheduling", "--mamba-backend", "triton", "--mamba-ssm-cache-dtype", "float32", "--tensor-parallel-size", "16", "--nnodes", "2", "--node-rank", "1", "--master-addr", "$HEAD_IP", "--headless", "--enable-auto-tool-choice", "--tool-call-parser", "qwen3_coder", "--reasoning-parser", "nemotron_v3", "--speculative-config", "{\"method\":\"mtp\",\"num_speculative_tokens\":5}" ] ], "strategy_spec": { "name": "multi_node_tp", "deploy_type": "multi_node", "display_name": "Multi-Node Tensor Parallel", "short_label": "Multi-node TP", "orientation": "latency", "description": "Multi-node TP. Splits the model across GPUs spanning multiple nodes. TP is set to total GPU count across all nodes. Head node serves requests, worker nodes run with --headless. Requires InfiniBand interconnect.\n", "hardware_match": { "min_gpus": 2, "multi_node": true }, "parallel_flag": "--tensor-parallel-size", "env": { "GLOO_SOCKET_IFNAME": "$IFACE_NAME", "NCCL_SOCKET_IFNAME": "$IFACE_NAME" } }, "hardware_profile": { "brand": "NVIDIA", "generation": "hopper", "compute_arch": "sm90", "display_name": "H200", "description": "NVIDIA H200 SXM 141 GB HBM3e · 8-GPU HGX node", "gpu_count": 8, "vram_gb": 1128, "multi_node": false }, "alternatives": { "single_node_tp": "/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16/strategies/single_node_tp.json", "kv_store_centralized_mooncake": "/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16/strategies/kv_store_centralized_mooncake.json", "kv_store_distributed_mooncake": "/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16/strategies/kv_store_distributed_mooncake.json" }, "by_hardware": { "h200": "/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16/hw/h200.json", "h100": "/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16/hw/h100.json", "b200": "/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16/hw/b200.json", "gb200": "/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16/hw/gb200.json", "b300": "/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16/hw/b300.json", "gb300": "/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16/hw/gb300.json", "mi300x": "/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16/hw/mi300x.json", "mi325x": "/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16/hw/mi325x.json", "mi350x": "/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16/hw/mi350x.json", "mi355x": "/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16/hw/mi355x.json" } }, "model": { "model_id": "nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", "min_vllm_version": "0.22.0", "docker_image": "vllm/vllm-openai:v0.28.0", "architecture": "moe", "parameter_count": "550B", "active_parameters": "55B", "context_length": 262144, "base_args": [ "--served-model-name", "nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B", "--trust-remote-code", "--kv-cache-dtype", "fp8", "--max-num-seqs", "16", "--max-model-len", "262144", "--gpu-memory-utilization", "0.90", "--max-num-batched-tokens", "32768", "--enable-flashinfer-autotune", "--async-scheduling", "--mamba-backend", "triton", "--mamba-ssm-cache-dtype", "float32" ], "base_env": { "VLLM_MEMORY_PROFILER_ESTIMATE_CUDAGRAPHS": "1" }, "install": { "pip": { "command": "uv venv\nsource .venv/bin/activate\nuv pip install -U vllm --torch-backend auto" }, "docker": { "command": "docker pull vllm/vllm-openai:v0.28.0" } } }, "features": { "tool_calling": { "description": "Qwen3 Coder tool-call parser with automatic tool choice", "args": [ "--enable-auto-tool-choice", "--tool-call-parser", "qwen3_coder" ] }, "reasoning": { "description": "Nemotron v3 reasoning parser", "args": [ "--reasoning-parser", "nemotron_v3" ] }, "spec_decoding": { "description": "Multi-Token Prediction with 5 speculative tokens", "args": [ "--speculative-config", "{\"method\":\"mtp\",\"num_speculative_tokens\":5}" ] }, "long_context": { "description": "Raise the context window from the default 256K to the model's full 1M tokens. Needs a large KV-cache budget — lower --max-model-len if you are memory-constrained.", "args": [ "--max-model-len", "1048576" ], "env": { "VLLM_ALLOW_LONG_MAX_MODEL_LEN": "1" } } }, "opt_in_features": [ "long_context" ], "variants": { "default": { "precision": "bf16", "vram_minimum_gb": 1320, "description": "BF16 weights" }, "fp4": { "model_id": "nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-NVFP4", "precision": "nvfp4", "vram_minimum_gb": 330, "description": "NVFP4 weights", "json": "/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-NVFP4.json" } }, "kv_offload_support": { "offloading_cpu": "verified", "offloading_fs": "verified" }, "hardware_overrides": {}, "guide": "## Overview\n\nNVIDIA Nemotron 3 Ultra is a 550B total / 55B active hybrid Transformer-Mamba MoE\nmodel built for long-running agentic reasoning, coding, research, and tool use.\n\n## Launch command\n\n```bash\nexport VLLM_MEMORY_PROFILER_ESTIMATE_CUDAGRAPHS=1\n\nvllm serve nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-NVFP4 \\\n --served-model-name nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B \\\n --trust-remote-code \\\n --tensor-parallel-size 8 \\\n --kv-cache-dtype fp8 \\\n --max-num-seqs 16 \\\n --max-model-len 262144 \\\n --gpu-memory-utilization 0.90 \\\n --max-num-batched-tokens 32768 \\\n --enable-flashinfer-autotune \\\n --async-scheduling \\\n --speculative-config '{\"method\":\"mtp\",\"num_speculative_tokens\":5}' \\\n --mamba-backend triton \\\n --mamba-ssm-cache-dtype float32 \\\n --reasoning-parser nemotron_v3 \\\n --enable-auto-tool-choice \\\n --tool-call-parser qwen3_coder\n```\n\n## References\n\n- [BF16 model](https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16)\n- [NVFP4 model](https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-NVFP4)\n", "source": { "source": "vllm", "repo": "vllm-project/vllm", "ref": "main", "sha": "564ac0de1b228568cafcbfbe8ccaccc82d6023eb", "path": "recipes/models/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16.yaml", "edit_url": "https://github.com/vllm-project/vllm/edit/main/recipes/models/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16.yaml", "overlays": [] } }