AI Compute
20 PFLOPS
Up to 20 PFLOPS of FP4 AI compute with sparsity.
NVIDIA DGX Station
Built for Your Next AI Breakthrough.
Bring demanding AI workloads to your desk. Powered by the NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip and 748GB of coherent memory, DGX Station gives developers, researchers and data scientists the capacity to develop, fine-tune and run large AI models locally.
Powerful AI Computing. Right at Your Desk.
Coherent Memory
748GB
252GB of GPU memory and 496GB of CPU memory in one coherent system.
Model Capacity
1 Trillion
Run AI models with up to 1 trillion parameters locally.
DGX Station platform highlights. Compute figures are theoretical peaks; actual performance varies by workload. Model capacity depends on precision, quantization, context length and software, and does not indicate equivalent training or fine-tuning capacity.
Find the Right System for Your AI Workloads.
Compare NVIDIA DGX Spark and DGX Station, from compact desktop development to demanding deskside AI workloads.
Swipe horizontally to compare both systems →
| Platform comparison |
DGX Spark
Compact desktop AI computing
|
DGX Station
Advanced deskside AI computing
|
|---|---|---|
| NVIDIA Superchip |
NVIDIA GB10 Grace Blackwell |
NVIDIA GB300 Grace Blackwell Ultra Desktop |
| AI Performance* | Up to 1 PFLOP FP4 with sparsity | Up to 20 PFLOPS FP4 with sparsity |
| Memory | 128GB Coherent unified LPDDR5X memory | 748GB Coherent CPU and GPU memory 252GB HBM3e GPU + 496GB LPDDR5X CPU |
| Model Capacity** | Up to 200 billion parameters | Up to 1 trillion parameters |
| CPU | 20-core Arm CPU | 72-core NVIDIA Grace CPU |
| GPU Architecture | NVIDIA Blackwell | NVIDIA Blackwell Ultra |
| AI Networking | NVIDIA ConnectX-7 Up to 200Gb/s | NVIDIA ConnectX-8 Up to 800Gb/s |
| Applications |
AI prototyping Fine-tuning and inference Local AI agents Edge application development |
Large-model development Fine-tuning and inference Advanced data science Shared team AI computing |
| Form Factor | Compact desktop system | Deskside workstation |
| Explore More | Explore More |
*Theoretical peak FP4 performance with sparsity.
Actual performance varies by workload.
**Model capacity depends on precision, quantization,
context length and software. These figures do not
indicate equivalent fine-tuning or training capacity.
System design, ports and configurations vary by manufacturer.
Images are illustrative.
Explore DGX Station Systems
Discover available systems for your next AI project. Contact us for configuration details and availability.
FAQs
NVIDIA DGX Station is a deskside AI supercomputer for developers, researchers and data scientists working on AI development, fine-tuning and inference.
It combines the NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip with 748GB of coherent memory.
DGX Station systems are offered by different manufacturers. Check each product listing for its configuration and included software.
NVIDIA states that DGX Station supports models of up to 1 trillion parameters.
Actual capacity and performance depend on model precision, quantisation, context length, batch size and software. This headline capacity should not be interpreted as a guarantee that every model of that size will run at every precision or speed.
Inference, fine-tuning and full training have different memory requirements. Confirm your intended model and workload before choosing a configuration.
No. NVIDIA specifies 252GB of HBM3e GPU memory with 496GB of LPDDR5X CPU memory.
These form a coherent CPU–GPU memory pool. The total is not the same as having 748GB of dedicated GPU VRAM.
Yes. Supported models and datasets can be processed on the local system for AI development, inference and data science.
Internet access may still be required to download models, containers, software packages, updates or licences, and for applications that use external services.
Some workloads may require additional infrastructure. Running locally does not automatically make an application fully offline.
NVIDIA lists Ubuntu with NVIDIA AI Developer Tools for DGX Station. It has also announced DGX Station for Windows, with availability expected in Q4 2026.
The supplied operating system and software depend on the selected model. Do not assume that a Windows announcement means every existing DGX Station supports a Windows installation.
Check the product listing and manufacturer support for your applications, drivers and Arm64 software requirements before ordering.
We Ideal Tech PC Sdn Bhd supplies the DGX Station-based hardware system according to the specifications, accessories, software and warranty coverage stated in the individual product listing.
AI model development, dataset preparation, model training, application development, API integration, workflow configuration, software migration and custom AI deployment services are not included unless they are specifically stated or separately quoted.
Customers are advised to confirm their intended models, software frameworks, technical requirements and deployment environment before placing an order. Our team can assist with product specification and hardware-related enquiries.
No. Use this page as an introduction to the DGX Station platform. For a purchase, compare the individual product listings for storage, expansion options, operating system, included software, accessories and warranty.
Features described for the platform do not automatically mean every optional component or software licence is included in every system.
Have another question about DGX Station? Speak to us here.