NVIDIA DGX Spark
Your Personal AI Supercomputer
One PetaFLOP of AI performance on your desk. Run models with up to 200 billion parameters, fine-tune on proprietary data, and develop AI applications -- all without sending a single byte to the cloud.
Or schedule a call at a time that works for you
Desktop Form Factor
1
PetaFLOP FP4
128
GB Unified Memory
200B
Parameter Models
4TB
NVMe Storage
What You Can Do With DGX Spark
A full AI development platform that fits on your desk. No cloud bills, no data exposure, no GPU rental queues.
Run LLMs Locally
Run large language models with up to 200 billion parameters directly on your desk. Llama, Mistral, Gemma, and thousands of open-source models at full speed with zero latency.
Fine-Tune Models
Customize foundation models on your proprietary data. Train domain-specific AI that understands your industry, your terminology, and your workflows.
Eliminate Cloud Costs
Stop paying per-token API fees and hourly GPU rentals. DGX Spark is a one-time investment that pays for itself in months compared to cloud AI spending.
Private & Secure AI
Your data never leaves your facility. Perfect for classified, regulated, or sensitive workloads where cloud AI is not an option.
Scale with NVLink
Connect two DGX Spark units via NVLink to double memory to 256GB and boost compute. Scale your personal AI lab as your needs grow.
AI Development
Build RAG applications, AI agents, computer vision systems, and production AI services. Full CUDA development environment with NVIDIA AI Enterprise.
Who DGX Spark Is Built For
From individual researchers to compliance-driven organizations, DGX Spark puts enterprise AI in reach.
AI Researchers
Experiment with large models, run ablation studies, and prototype new architectures without waiting for cloud GPU availability.
Data Scientists
Develop and test ML pipelines, run feature engineering at scale, and iterate on models faster with local compute that is always available.
Small AI Teams
Give your team a shared AI development platform. One DGX Spark can serve multiple developers for prototyping, testing, and small-scale training.
Defense Contractors
Air-gapped AI for classified environments. Run AI models on sensitive defense data that cannot leave your facility. CMMC-compliant deployment available from PTG.
Healthcare Organizations
HIPAA-compliant AI for patient data analysis, medical imaging, clinical NLP, and research. All processing stays on-premises. PTG configures full HIPAA controls.
AI Developers
Build production AI applications -- RAG systems, chatbots, code assistants, image generators -- with a local development environment that matches production performance.
Technical Specifications
Every detail of the NVIDIA DGX Spark personal AI supercomputer.
| AI Performance | 1 PetaFLOP FP4 |
| GPU | NVIDIA Blackwell GPU |
| CPU | NVIDIA Grace CPU 20 ARM Cores: 10x Cortex-X925 + 10x Cortex-A725 |
| Memory | 128GB Unified LPDDR5X 273 GB/s bandwidth | Shared between CPU and GPU |
| Storage | Up to 4TB NVMe SSD |
| Networking | 1x 10GbE Ethernet |
| Wireless | WiFi 7 + Bluetooth 5.3 |
| Multi-Unit | Connect 2 units via NVLink Double memory to 256GB and increase compute |
| Form Factor | Desktop AI Computer Compact, office-friendly design |
| Power | Standard wall outlet No special electrical requirements |
| Cooling | Fan-cooled, office-quiet No special HVAC requirements |
| Software | NVIDIA AI Enterprise, NGC Catalog, NIM CUDA, cuDNN, PyTorch, TensorFlow included |
The PTG Advantage
DGX Spark is powerful out of the box. With PTG, it becomes a complete AI solution tailored to your organization.
Pre-Configured AI Stack
We set up your preferred frameworks, models, development tools, and workflows before delivery. Unbox it and start building immediately.
CMMC/HIPAA Compliant
Our CMMC-RP certified team configures encryption, access controls, audit logging, and security hardening to meet your compliance framework.
AI Consulting Included
Every DGX Spark purchase includes bundled AI consulting hours. We help you identify use cases, select models, and build your first AI applications.
On-Site Setup & Training
For Raleigh-Durham Triangle clients, we deliver, install, and train your team in person. Remote setup and training available nationwide.
NVIDIA DGX Spark
- One-time purchase, unlimited usage
- Data never leaves your facility
- Zero latency, always available
- CMMC/HIPAA compliant by design
- No per-token or per-hour fees
- Run any model -- no vendor lock-in
Cloud AI Services
- Recurring monthly costs that scale with usage
- Data sent to third-party servers
- Network latency, outages, rate limits
- Compliance complexity with external processors
- Per-token and per-hour billing adds up fast
- Limited to provider's model selection
Frequently Asked Questions
With 128GB of unified memory, DGX Spark can run large language models with up to 200 billion parameters locally. This includes models like Llama 3.1 70B, Mixtral 8x22B, and many other open-source and commercial models. You can run inference, fine-tuning, and RAG applications entirely on your desktop without any cloud dependency.
DGX Spark uses NVIDIA's Grace CPU with unified memory architecture, meaning the CPU and GPU share the same 128GB memory pool. Traditional GPU workstations have separate CPU and GPU memory connected by PCIe, which creates a bottleneck for large AI models. DGX Spark eliminates this bottleneck and delivers 1 PetaFLOP of FP4 performance in a compact desktop form factor that uses a fraction of the power.
Yes. Two DGX Spark units can be connected via NVIDIA NVLink to double the available memory to 256GB and increase compute performance. This allows you to run even larger models or increase throughput for inference workloads. PTG can configure and optimize multi-unit setups for your specific use case.
DGX Spark comes with NVIDIA AI Enterprise software suite and access to NVIDIA NGC catalog of pre-trained models, containers, and AI tools. PTG additionally pre-configures your preferred AI frameworks including PyTorch, TensorFlow, Ollama, and development environments. We also set up NVIDIA NIM microservices for easy model deployment.
No. DGX Spark runs on standard office power from a regular wall outlet and uses quiet fan-based cooling suitable for office and desk environments. It is designed to sit on or under your desk like a desktop workstation, with no special electrical or HVAC requirements.
Absolutely. Our entire team is CMMC-RP certified, and we specialize in compliant AI deployments. DGX Spark is ideal for compliance-sensitive environments because all data processing happens locally -- no data ever leaves your facility. PTG configures DGX Spark with encryption, access controls, audit logging, and security hardening to meet HIPAA, CMMC, and NIST 800-171 requirements.
NVIDIA DGX Spark is available for pre-order through authorized partners like Petronella Technology Group. Contact us to reserve your unit and discuss pre-configuration options. PTG offers bundled packages including setup, training, and ongoing AI consulting support.
Bring AI to Your Desk
NVIDIA DGX Spark puts a PetaFLOP of AI performance in a desktop form factor. No data center required, no cloud dependency, no compromise on capability.
Reserve your DGX Spark today and our CMMC-RP certified team will configure it for your exact use case -- whether that is private LLM inference, compliant AI development, or cutting-edge research.
Or schedule a call at a time that works for you
Petronella Technology Group | 5540 Centerview Dr, Suite 200, Raleigh, NC 27606 | Since 2002