
NVIDIA DGX Spark
Compact desktop AI supercomputer for prototyping, fine-tuning and running models locally.
- Superchip: NVIDIA GB10 Grace Blackwell
- Memory: Up to 128 GB unified system memory
- AI performance: Up to 1 petaFLOP (FP4)
Deskside AI supercomputer for developers and research teams working on large models.
NVIDIA DGX Station is a deskside AI supercomputer powered by the GB300 Grace Blackwell Ultra Desktop Superchip. It brings 748 GB of coherent memory and up to 20 petaFLOPS of FP4 AI compute to a single workstation, enough for models up to one trillion parameters.
It runs the same NVIDIA AI software as the data center, supports Multi-Instance GPU so several developers can share it, and includes BMC and Redfish management so IT can run it like a server.
A Blackwell Ultra GPU and a 72-core Grace CPU joined by 900 GB/s NVLink-C2C.
252 GB HBM3e plus 496 GB LPDDR5X in one unified memory pool.
Up to 20 PFLOPS of FP4 AI compute for training and inference.
Up to 800 Gb/s networking; link two DGX Stations together.
Up to seven isolated instances to share the system across a team.
Out-of-band BMC telemetry, Redfish API and NVIDIA DCGM.
| GPU | 1× NVIDIA Blackwell Ultra |
|---|---|
| CPU | 1× NVIDIA Grace, 72-core Neoverse V2 |
| NVLink-C2C | 900 GB/s |
| MIG | Up to 7 instances |
| GPU memory | 252 GB HBM3e | 7.1 TB/s |
|---|---|
| CPU memory | 496 GB LPDDR5X | 396 GB/s |
| Total coherent memory | 748 GB |
| FP4 Tensor Core | 20 PFLOPS |
|---|---|
| FP8/FP6 Tensor Core | 10 PFLOPS |
| INT8 Tensor Core | 330 TOPS |
| FP16/BF16 Tensor Core | 5 PFLOPS |
| TF32 Tensor Core | 2.5 PFLOPS |
| FP32 | 80 TFLOPS |
| FP64 / FP64 Tensor Core | 1.3 TFLOPS |
| Networking | ConnectX-8 SuperNIC, up to 800 Gb/s; 2× QSFP112 (400 Gb/s each); 1× 10 GbE; 1× 1 GbE BMC |
|---|---|
| Storage | 4× M.2 Gen 5 slots |
| PCIe slots | 1× PCIe Gen 5 x16, 2× PCIe Gen 5 x16 (x8 electrical) |
| Optional graphics | NVIDIA RTX PRO 6000, 4000 or 2000 Blackwell |
| Decoders | 7 NVDEC, 7 nvJPEG |
| Power | 1,600 W total system power, 20 A circuit |
| Operating system | Ubuntu with NVIDIA AI developer tools |
Tell us your workload and timeline for the NVIDIA DGX Station. We reply within one business day with a configuration, pricing and lead time.