NVIDIA GB300 Grace Blackwell Ultra Superchip
TrainingNvidia · Blackwell Ultra (Grace Blackwell Superchip) · 2026
Grace Blackwell Ultra platform GPU. GB300 pairs NVIDIA Grace CPUs with Blackwell Ultra GPUs over NVLink-C2C and is deployed rack-scale as GB300 NVL72. vramGb reflects the 288GB HBM3e per Blackwell Ultra GPU published by NVIDIA; per-GPU bandwidth is ~8 TB/s with NVLink 5 (1.8 TB/s bidirectional per GPU) and ~15 PFLOPS dense FP4. This model backs the DataCrunch/Verda "GB300 SXM6 288GB" SKU. Dense FP16/FP8 TFLOPS are not separately broken out on the pages checked and are left unset.
- VRAM
- 288 GB
- Memory
- HBM3e
- FP16
- —
- FP8
- —
From
Verified 9 Jul 2026—
Offerings of this GPU
3 rows
Verda (formerly DataCrunch)
Compute
Quote only
auto-checked 2026-07-09
- Config
- 4× GPU · Bare metal · 288 GB
- Region
- FIIS
- Status
- Available
EU Score98
Verda (formerly DataCrunch)
Compute
Quote only
auto-checked 2026-07-09
- Config
- 1× GPU · Bare metal · 288 GB
- Region
- FIIS
- Status
- Available
EU Score98
Verda (formerly DataCrunch)
Compute
Quote only
auto-checked 2026-07-09
- Config
- 2× GPU · Bare metal · 288 GB
- Region
- FIIS
- Status
- Available
EU Score98
Spec sources
- Official Docswww.nvidia.com/en-us/data-center/gb300-nvl729 Jul 2026
- Official Docsdeveloper.nvidia.com/blog/inside-nvidia-blackwell-ultra-the-chip-powering-the-ai-factory-era9 Jul 2026