A reserve-only Blackwell Ultra GPU for high-memory AI reasoning, inference, and training infrastructure.
- Memory
- 288GB HBM3e per GPU
- Access
- Reservation only
- Lead time
- 3-7 weeks
Dedicated GPU infrastructure
Explore GPUs
Compare on-demand and reserve-only NVIDIA and AMD GPU infrastructure by memory, pricing, and deployment model.
A reserve-only Blackwell GPU for large-model training, inference, and accelerated AI pipelines.
- Memory
- 180GB HBM3e per GPU
- Access
- Reservation only
- Lead time
- 3-7 weeks
Available in 4 weeks through reservation: an AMD CDNA 3 accelerator with 192GB of HBM3 memory for generative AI and HPC workloads.
- Memory
- 192GB HBM3 per GPU
- Access
- Available in 4 weeks - reserve only
- Lead time
- 4 weeks
A high-memory Blackwell professional GPU for teams that need AI compute and advanced graphics capability in the same system.
- Memory
- 96GB GDDR7 per GPU
- Configurations
- 4
A memory-focused Hopper GPU family for large-model inference and data-intensive scientific computing.
- Memory
- 141GB HBM3e per GPU
- Configurations
- 5
A Hopper data-center GPU family for transformer training, large-scale inference, and scientific computing.
- Memory
- 80GB SXM5 or 94GB NVL per GPU
- Configurations
- 4
A data-center GPU that spans generative AI, multimodal workloads, and high-end visual computing.
- Memory
- 48GB GDDR6 per GPU
- Configurations
- 4
An Ampere data-center GPU for established deep-learning, scientific-computing, and analytics stacks.
- Memory
- 80GB HBM2e per GPU
- Configurations
- 4
A Volta data-center GPU for established deep-learning and scientific-computing applications.
- Memory
- 32GB HBM2 per GPU
- Configurations
- 3
Selection / 02
Choose around the workload.
On-demand rates come from the public pricing feed. MI300X is available in 4 weeks through reservation; B200 and B300 reservations have an estimated 3-7 week lead time.
- 01
Memory fit
Include model weights, runtime state, batches, caches, and framework overhead.
- 02
Software path
Match the CUDA, driver, precision, framework, and container requirements.
- 03
Access model
Compare published on-demand configurations with planned reserve-only capacity.