Announcing Region Expansion of P6-B200 instances on SageMaker Notebook Instances
NVIDIA Blackwell B200 GPUs with 1440 GB memory now power SageMaker notebooks in us-east-1 — fine-tune frontier LLMs interactively.
View original announcement →Visual Summary
graph TD
A{{P6-B200 on SageMaker Notebooks}}:::announced
B((Data Scientists)):::external
C(SageMaker JupyterLab):::compute
D(SageMaker Code Editor):::compute
E([8x NVIDIA Blackwell GPUs]):::feature
F([1440 GB GPU Memory]):::feature
G(Amazon S3):::storage
H([Foundation Model Fine-tuning]):::feature
I(SageMaker Training):::compute
B ==>|"launches"| A
A -->|"runs in"| C
A -->|"runs in"| D
A -->|"powered by"| E
E -->|"provides"| F
F -->|"enables"| H
H -->|"stores artifacts"| G
C -.->|"submits jobs"| I
I -->|"writes to"| G
classDef announced fill:#ff9900,stroke:#ec7211,color:#fff,font-weight:bold
classDef compute fill:#e3f2fd,stroke:#1565c0,color:#1565c0
classDef storage fill:#e8f5e9,stroke:#2e7d32,color:#2e7d32
classDef feature fill:#fff3e0,stroke:#e65100,color:#e65100
classDef external fill:#f5f5f5,stroke:#616161,color:#616161
What's New
Amazon EC2 P6-B200 instances are now generally available on SageMaker Notebook Instances in the AWS US East (N. Virginia) region, expanding access to NVIDIA Blackwell GPU-powered compute for interactive ML development. These instances offer up to 2x better AI training performance compared to P5en instances and are designed for developing and fine-tuning large foundation models directly within JupyterLab or CodeEditor environments. This release extends the P6-B200 footprint into SageMaker's notebook experience, enabling data scientists to experiment with frontier models without leaving their familiar IDE.
How It Works
- P6-B200 instances are backed by 8 NVIDIA Blackwell B200 GPUs per instance, providing 1440 GB of aggregate high-bandwidth GPU memory (HBM3e) for holding very large model weights in-memory during interactive experimentation.
- The instances pair Blackwell GPUs with 5th Generation Intel Xeon (Emerald Rapids) CPUs, balancing GPU-accelerated tensor operations with high-throughput CPU preprocessing pipelines.
- Users launch P6-B200 as the compute backing a SageMaker Notebook Instance, which runs a Jupyter server on the EC2 instance and provides preconfigured kernels with the SageMaker Python SDK, Boto3, deep learning frameworks (PyTorch, TensorFlow), and data science libraries.
- Within SageMaker Studio, the same hardware can back a JupyterLab space or a Code Editor space, both of which use a single EC2 instance and an attached EBS volume for storage, allowing interactive notebook development and VS Code-style editing respectively.
- Model fine-tuning workflows (e.g., LoRA, full fine-tuning of LLMs) can be run interactively in the notebook environment, with results written to S3 or local EBS, and training jobs can be submitted to SageMaker Training from the same notebook.
- The SageMaker Distribution image pre-installed on notebook instances includes popular ML packages, reducing environment setup time when working with large foundation models on the P6-B200 hardware.
Why It's Important
- The 1440 GB of GPU memory per instance allows practitioners to load very large models (multi-hundred-billion parameter LLMs, mixture-of-experts models) entirely into GPU memory for interactive fine-tuning without requiring multi-node distributed setups.
- Delivering up to 2x training performance over P5en instances means iteration cycles during experimentation are significantly faster, reducing the time from hypothesis to validated result for research and applied ML teams.
- Bringing P6-B200 into the SageMaker Notebook Instance experience lowers the barrier to entry for using cutting-edge Blackwell hardware — users can access it through a familiar Jupyter interface without orchestrating complex infrastructure.
- Support for multi-modal reasoning models and mixture-of-experts architectures directly in a notebook environment enables teams building enterprise copilots, content generation pipelines, and agentic AI applications to prototype end-to-end workflows interactively.
- The availability in US East (N. Virginia), AWS's largest and most feature-rich region, ensures that the majority of enterprise customers with workloads already anchored there can adopt P6-B200 without data residency or latency concerns.
How It's Different
- Compared to P5en instances (NVIDIA H100), P6-B200 delivers up to 2x AI training and inference performance, driven by the architectural advances in NVIDIA's Blackwell generation over Hopper.
- P6-B200 provides 1440 GB of GPU memory across 8 GPUs versus P5en's configuration, enabling single-instance fine-tuning of models that previously required multi-node setups with P5en.
- Unlike the P6e UltraServer line (GB200/GB300 NVL72), which targets frontier-scale multi-trillion-parameter training across 72 GPUs in a single NVLink domain, P6-B200 is positioned as a more accessible single-node option suited for medium-to-large-scale training and interactive development.
- The integration into SageMaker Notebook Instances differentiates P6-B200 from raw EC2 usage by providing managed Jupyter infrastructure, pre-built ML kernels, and direct integration with SageMaker services (Training, Hosting, JumpStart) out of the box.
- P6-B300 (Blackwell Ultra) instances also exist in the P6 family and offer higher performance for large-scale training, but P6-B200 represents the more cost-accessible entry point into Blackwell-class compute on SageMaker notebooks.
When to Prefer It
- Use P6-B200 notebook instances when interactively fine-tuning large LLMs (e.g., 70B–200B parameter models) where the full model or optimizer states need to reside in GPU memory without multi-node coordination overhead.
- Prefer P6-B200 when prototyping or experimenting with mixture-of-experts (MoE) or multi-modal reasoning models that have large memory footprints but don't yet require the full scale of a P6e UltraServer cluster.
- Choose P6-B200 on SageMaker Notebook Instances when your team needs a managed, IDE-integrated environment (JupyterLab or CodeEditor) with Blackwell GPU access, avoiding the operational overhead of self-managed EC2 instances.
- Ideal for generative AI application development — such as enterprise copilots or text/image/video content generation — where rapid iteration in a notebook is needed before promoting code to a full SageMaker Training job.
- Select P6-B200 when your workloads are already in US East (N. Virginia) and you need the highest single-instance GPU performance available on SageMaker notebook instances in that region today.
- Consider P6-B200 for teams evaluating a step-up from P5en who want meaningfully better throughput without moving to the significantly larger and more expensive P6e UltraServer form factor.
Availability
- Status: Generally Available (GA) on SageMaker Notebook Instances.
- Supported Region: AWS US East (N. Virginia) — this announcement specifically represents a region expansion, implying prior availability in at least one other region.
- Supported Environments: SageMaker Notebook Instances (classic), SageMaker Studio JupyterLab spaces, and SageMaker Studio Code Editor spaces.
- Pricing Model: Standard EC2 P6-B200 on-demand or reserved instance pricing applies; SageMaker Notebook Instance pricing is based on the underlying EC2 instance hours consumed — refer to the Amazon SageMaker Pricing page for current rates.
- Limitations: Availability is currently confirmed only for US East (N. Virginia); other regions are not yet announced for this specific SageMaker notebook integration. Capacity may be constrained given the high demand for Blackwell-class hardware.
- Prerequisites: Users should refer to the SageMaker developer guides for JupyterLab and CodeEditor setup; standard SageMaker IAM permissions and service quotas for P6 instance types apply.
Related Resources
- https://aws.amazon.com/ec2/instance-types/p6/
- https://docs.aws.amazon.com/sagemaker/latest/dg/studio-updated-jl.html
- https://docs.aws.amazon.com/sagemaker/latest/dg/code-editor.html
- https://docs.aws.amazon.com/sagemaker/latest/dg/studio-updated.html
- https://docs.aws.amazon.com/sagemaker/latest/dg/nbi.html
