NVIDIA
B300↗
Blackwell UltraFrontier models. Next-generation reasoning.
- GPU memory
- 288 GB HBM3e
- Bandwidth
- 8 TB/s max
- Deployment
- Dedicated capacity
NVIDIA B300 Big ideas.
Big ideas.
Meet massive
compute.
Exceptional GPU capacity for extraordinary ambitions. From your first training run to the next frontier of AI.
288 GBHBM3e memory / B300
8 TB/sMemory bandwidth / B300, up to
8 GPUsPer HGX B300 system
Explore the architecture Your workload.
The right horsepower.
Choose the architecture that moves you forward.
Powerful alone. Transformative at scale.
NVIDIA
B200↗
BlackwellSerious scale for training and inference.
- GPU memory
- 180 GB HBM3e
- Bandwidth
- 8 TB/s max
- Deployment
- Dedicated capacity
NVIDIA
H200↗
HopperMore memory for your most demanding models.
- GPU memory
- 141 GB HBM3e
- Bandwidth
- 4.8 TB/s
- Deployment
- Dedicated capacity
NVIDIA
H100↗
Hopper · SXMA proven foundation for AI and HPC.
- GPU memory
- 80 GB HBM3
- Bandwidth
- 3.35 TB/s
- Deployment
- Dedicated capacity
Specifications are per GPU. Configuration and availability subject to confirmation.Technical specifications
Think in models.
We think in
infrastructure.
A great GPU is only the beginning. Plan the memory, interconnect, and cluster shape your workload actually needs.
High-bandwidth GPU interconnect
Dedicated, multi-GPU configurations
Architecture matched to your model
HGX B300REFERENCE TOPOLOGY
8 × B300
GPU 00288 GB
GPU 01288 GB
GPU 02288 GB
GPU 03288 GB
GPU 04288 GB
GPU 05288 GB
GPU 06288 GB
GPU 07288 GB
NVLink + NVSwitchALL-TO-ALL
2.3 TBAGGREGATE HBM3e
8 GPUsONE CONNECTED SYSTEM
Less waiting.
More what’s possible.
Compute for the work that can’t stand still.
Train at the frontier.
Build foundation models with the memory and parallel compute that large-scale training demands.
Make inference fly.
Put reasoning and generative models to work with high-bandwidth memory built for demanding inference.
Go beyond the model.
Accelerate simulations, scientific research, and complex calculations with GPU-powered computing.
What will you
build next?
The next big thing starts with
the right compute.