NVIDIA B200 vs B300
Compare NVIDIA Blackwell B200 and B300 (Blackwell Ultra) side by side: memory and FP4 throughput. Both run as reserved capacity.
- GPU memory
- 192GB HBM3e
- Memory bandwidth
- 8 TB/s
- GPU memory
- up to 288GB HBM3e
- Memory bandwidth
- 8 TB/s
Technical specifications
B200 NVIDIA B200 | B300 NVIDIA B300 | |
|---|---|---|
| Architecture | Blackwell | Blackwell |
| GPU memory | 192GB HBM3e | up to 288GB HBM3e |
| Memory bandwidth | 8 TB/s | 8 TB/s |
| NVLink | 1.8 TB/s | 1.8 TB/s |
| FP8 (Tensor) | 9 PFLOPS | 9 PFLOPS |
| FP4 (Tensor) | 18 PFLOPS | 18 PFLOPS |
| Max TDP | 1,000W | 1,100W |
| GPUs per node | 8 (HGX B200) | 8 (HGX B300) |
*Peak performance with sparsity, per NVIDIA official specs. Final specs may vary by node configuration.
Pricing & availability
What's Blackwell best for?
Frontier-scale pretraining
FP4 and FP8 throughput plus 1.8 TB/s NVLink make trillion-parameter model pretraining practical in a single Blackwell cluster.
Long-context & reasoning inference
B300 (Blackwell Ultra) holds massive KV caches in up to 288GB HBM3e, and FP4 acceleration serves agentic and reasoning workloads at high concurrency.
Generative video & multi-modal
Memory- and compute-heavy workloads (text-to-video, 3D generation, large multi-modal training) fit on a single HGX Blackwell node.
Why AI teams run GPUs on VESSL Cloud
From A100 to Blackwell Ultra
Run A100 and H100 on demand. Reserve H200 and Blackwell GPUs for larger runs, so each workload gets the GPU it needs.
Issues caught early
Every VM Cluster node is health-checked continuously, with its status refreshed about once a minute. Workspaces and Jobs alert you when a run crosses a resource threshold or hits a runtime issue.
From one node to a cluster
Train and serve on up to 8 GPUs on one node, and move to a multi-node VM Cluster when one node is not enough.
Enterprise-ready
SOC 2 Type II compliance, with dedicated support for production AI.
Frequently asked questions
How do I get access to NVIDIA B200 and B300 GPUs?
Blackwell capacity is allocated on request. Talk to our team for current availability and pricing.
What's the difference between the B200 and B300?
The B300 (Blackwell Ultra) increases memory to up to 288GB HBM3e (vs 192GB on the B200) and delivers roughly 1.5× the B200's dense FP4 compute, built for the largest models and high-concurrency inference.
How much memory do Blackwell GPUs have?
The B200 offers 192GB HBM3e per GPU at up to 8 TB/s memory bandwidth and 1.8 TB/s NVLink. The B300 (Blackwell Ultra) has up to 288GB HBM3e per GPU. Talk to our team for current node configurations and availability.
Can I reserve a full Blackwell cluster?
Yes. We provision HGX B200 and B300 nodes (8 GPUs each), from a single node to multi-node clusters. B200 nodes are joined by InfiniBand, and B300 nodes by InfiniBand or RoCE, depending on the cluster. How many nodes you can reserve depends on current supply, so talk to our team for availability.
Is Blackwell better for training or inference?
Both. FP4/FP8 acceleration and large HBM3e capacity make Blackwell ideal for frontier-scale training and high-throughput, low-latency inference.
Where AI models
get their GPUs
- Self-serve on-demand GPUs
- Scale to multi-node clusters
- Reserved terms on H200 and Blackwell
- Dedicated support