SYSTEMPBA-0620
8 × NVIDIA Blackwell GPUs
B200
Use forAI trainingInferenceHPC
NVIDIA
NVIDIA DGX B200
NVIDIA DGX B200 combines eight Blackwell GPUs, 1.44TB of HBM3e and fifth-generation NVLink in a fully integrated 10U platform for frontier-model training and high-throughput inference.
Request a quote →Engineer replies, not a bot.
REPLY WITHIN 24H Authorized sourcing Engineer support
Bulk order or custom build?Talk to an engineer →
Technical
specifications
The details behind your next build.
System
- GPUs
- 8x NVIDIA B200 Tensor Core GPUs
- GPU architecture
- NVIDIA Blackwell
- Total GPU memory
- 1,440 GB HBM3e
- FP4 Tensor Core performance
- 144 PFLOPS sparse / 72 PFLOPS dense
- FP8 Tensor Core performance
- 72 PFLOPS sparse
Interconnect
- GPU fabric
- fifth-generation NVLink
- NVLink bandwidth
- 14.4 TB/s aggregate
- NVSwitches
- 2
- Scale-out networking
- 8x NVIDIA ConnectX-7
- Infrastructure services
- 2x NVIDIA BlueField-3 DPUs
Host
- Processors
- 2x Intel Xeon Platinum 8570
- System memory
- 2 TB, configurable to 4 TB
- OS storage
- 2x 1.92 TB M.2 NVMe in RAID 1
- Internal storage
- 8x 3.84 TB U.2 NVMe
Software
- Platform
- NVIDIA DGX software stack
- Operations
- NVIDIA Mission Control
- Operating system
- NVIDIA DGX OS / Ubuntu
Physical
- Form factor
- 10U rackmount
- Maximum system power
- approximately 14.3 kW
- Dimensions
- 444 x 482.2 x 897.1 mm
- Operating temperature
- 10–35 °C
Every order, covered.
[ 01 ]
Sealed & genuine
Factory-sealed, authorized stock. Warranty valid.
[ 02 ]
Insured, tracked dispatch
Lead time is confirmed on your order, not guessed at.
[ 03 ]
Real engineer support
Talk to people who run this hardware.
[ 04 ]
Returns without drama
30 days. DOA? Replacement ships first.
