NVIDIA · Blackwell Ultra · Data centre
NVIDIA B300 (Blackwell Ultra)
Blackwell Ultra, the B200's successor, with more HBM3e memory per GPU, sold in HGX B300 systems.
01 / SPECIFICATIONS
What the hardware holds
- Memory
- 270 GB
- Memory bandwidth
- 7.7 TB/s
- FP8 dense
- 4,500 TFLOPS
- FP16 / BF16 dense
- 2,250 TFLOPS
- Variant
- B300 (HGX B300)
- Memory type
- HBM3e
- Form factor
- SXM (HGX B300 8-GPU baseboard)
- Interconnect
- NVLink 5 1.8 TB/s
- Max power
- 1,100 W
Source notes for B300 (HGX B300)
NVIDIA Blackwell Ultra datasheet, 'Individual Blackwell Ultra GPU Specifications', HGX B300 column - per-GPU values printed directly (system: 8 GPUs, 2.1 TB, 62 TB/s). Datasheet lists FP16/BF16 Tensor 4.5 PFLOPS and FP8/FP6 9 PFLOPS with sparsity (footnote 2: dense is 1/2 sparse); dense = 2.25 / 4.5 PFLOPS. FP4 18 PFLOPS sparse | 14 PFLOPS dense. TDP configurable up to 1,100 W. Also PCIe Gen6 256 GB/s. Linked from https://resources.nvidia.com/en-us-blackwell-architecture/blackwell-ultra-datasheet.
Compute figures are dense, without sparsity. NVIDIA source for B300 (HGX B300) Checked 8 Nov 2026.
02 / MEMORY FIT
How many do you need to hold a model?
| Model size | FP16 / BF16 | FP8 | INT4 |
|---|---|---|---|
| 8B parameters | 1 GPU16 GB of weights | 1 GPU8 GB of weights | 1 GPU4 GB of weights |
| 70B parameters | 1 GPU140 GB of weights | 1 GPU70 GB of weights | 1 GPU35 GB of weights |
| 405B parameters | 4 GPUs810 GB of weights | 2 GPUs405 GB of weights | 1 GPU203 GB of weights |
GPUs = parameters × bytes per parameter × 1.2 headroom ÷ 270 GB, rounded up. This covers weights only. Long contexts, large batches, and training need much more memory.
03 / OFFERS
Reviewed on-demand rates
No reviewed rate yet. The sellers below list the B300; check their pricing pages for current terms.
Per GPU-hour divides a machine's hourly price by its GPU count, so single GPUs and 8-GPU nodes compare fairly. The billed price is what one machine costs per hour. Storage, network transfer, and minimum terms can add to the bill.
04 / SELLERS
Who lists the B300
RunpodGPU cloud · Self-serve
CoreWeaveGPU cloud · Sales-led
NebiusGPU cloud · Self-serve
CrusoeGPU cloud · Self-serve
Vast.aiMarketplace · Self-serve
DigitalOceanGPU cloud · Self-serve
Runpod, CoreWeave, Nebius, Crusoe, Vast.ai, DigitalOcean list the B300 without a rate we have reviewed. Ask them for a quote, especially for clusters of 8 or more GPUs.
05 / COMPARE WITH