NVIDIA DGX Spark Price, Specifications & Availability
Explore the NVIDIA DGX Spark 128GB unified memory and 4TB SSD configuration. Review its GB10 Grace Blackwell architecture, local AI capabilities, 200Gb/s networking, pricing considerations and procurement availability.
Price and delivery estimates are confirmed individually against current supply and destination.
Executive Summary
NVIDIA DGX Spark is a compact personal AI supercomputer built around the NVIDIA GB10 Grace Blackwell Superchip. The 128GB configuration combines a Blackwell-generation GPU with a 20-core Arm CPU and 128GB of coherent unified system memory. The 4TB model adds self-encrypting NVMe M.2 storage for local development data, model files and project environments.
DGX Spark is designed to bring the NVIDIA AI software ecosystem to an individual developer’s desk, research lab or team workspace. It is not a conventional gaming PC or a multi-GPU rack server: its value is the integration of compute, unified memory, software and high-speed networking in a compact system.
What Is the NVIDIA DGX Spark AI Supercomputer?
NVIDIA DGX Spark is a small-form-factor AI system designed for local AI development, prototyping, inference and fine-tuning workflows. It runs NVIDIA DGX OS and provides access to NVIDIA’s AI software stack, development tools, libraries and supported models.
Its unified memory architecture allows the CPU and GPU to access a shared memory space. This is particularly relevant when developers want to experiment with larger AI models locally, subject to model architecture, quantization, context length and workload requirements.
Who is DGX Spark designed for?
- AI developers building and testing generative AI applications.
- Researchers experimenting with supported large language and vision-language models.
- Software teams developing AI agents and retrieval-augmented generation workflows.
- Organizations that need a local AI development platform without deploying a full data-center rack.
NVIDIA DGX Spark Price (128GB RAM, 4TB SSD)
The purchase price of NVIDIA DGX Spark can vary by region, authorized supply channel, configuration, warranty and support terms, taxes, shipping and current stock. A public retail price in one market should not be treated as a confirmed delivered price for another destination.
For the 128GB unified memory and 4TB NVMe configuration, Sadid Gostaran can provide a current quotation after confirming the destination country, required quantity and delivery preference. No fixed price or stock status is stated here unless confirmed for the specific order.
Send your destination and required quantity to receive current pricing and procurement information.
Get Price via WhatsApp ↗NVIDIA DGX Spark Technical Specifications
The following table summarizes the 128GB / 4TB configuration using NVIDIA’s published product specifications and user documentation.
| Specification | NVIDIA DGX Spark 128GB / 4TB |
|---|---|
| System platform | NVIDIA Grace Blackwell |
| Superchip | NVIDIA GB10 Grace Blackwell Superchip |
| CPU | 20-core Arm CPU: 10 Cortex-X925 + 10 Cortex-A725 |
| GPU architecture | NVIDIA Blackwell |
| Tensor Cores | 5th Generation |
| RT Cores | 4th Generation |
| AI performance | Up to 1 PFLOP FP4 (theoretical, with sparsity) |
| Unified system memory | 128GB LPDDR5x, coherent CPU-GPU memory |
| Memory interface | 256-bit |
| Memory bandwidth | 273 GB/s |
| Storage | 4TB NVMe M.2 with self-encryption |
| High-speed NIC | NVIDIA ConnectX-7, up to 200Gb/s |
| Ethernet | 1× RJ-45, 10GbE |
| Wireless | Wi-Fi 7, Bluetooth 5.4 |
| USB | 4× USB Type-C |
| Display | 1× HDMI 2.1a; DisplayPort over USB-C (DP Alt Mode) |
| Power supply | 240W |
| GB10 TDP | 140W |
| Dimensions | 150 × 150 × 50.5 mm |
| Weight | Approximately 1.2 kg |
| Operating system | NVIDIA DGX OS |
The 1 PFLOP FP4 figure is a theoretical peak figure and uses NVIDIA’s stated sparsity convention; it should not be interpreted as a guaranteed application throughput result. Configuration and regional SKU details should be checked before purchase.
NVIDIA DGX Spark official product specifications ↗ | DGX Spark hardware guide ↗
GB10 Grace Blackwell Superchip Architecture
At the heart of DGX Spark is the NVIDIA GB10 Grace Blackwell Superchip, which combines a 20-core Arm CPU and a Blackwell GPU in a tightly integrated platform. NVIDIA’s coherent unified memory design enables CPU and GPU workloads to work with the same system memory pool rather than relying on separate CPU and GPU memory pools in the conventional desktop arrangement.
This architecture is intended to simplify local AI development and enable experimentation with models that benefit from a large shared memory capacity. Actual model capacity and performance depend on precision, quantization, context window, runtime, batch size and available system resources.
128GB Unified Memory and 4TB NVMe SSD
128GB coherent unified memory
The 128GB LPDDR5x memory is shared coherently between the CPU and GPU. This unified design can make it easier to load and work with AI models that would be difficult to fit into the dedicated VRAM of many conventional desktop GPUs. It does not mean that every model with a 128GB file size will run: runtime overhead, model format, context length and other memory allocations also matter.
4TB self-encrypting NVMe storage
The 4TB NVMe M.2 drive provides local capacity for model checkpoints, datasets, development environments and application assets. Self-encrypting storage is a useful security feature, but customers should still establish their own backup, access-control and data-handling policies.
DGX Spark Networking, QSFP Ports and Connectivity
DGX Spark combines a standard 10GbE RJ-45 Ethernet connection with a high-speed NVIDIA ConnectX-7 networking interface rated up to 200Gb/s. NVIDIA’s hardware documentation identifies two QSFP network connectors for the ConnectX-7 interface. These high-speed ports support direct system-to-system connections and multi-system configurations when the correct cables and topology are used.
What are the high-speed ports used for?
- Connecting two DGX Spark systems directly using a supported QSFP cable.
- Building a small multi-system AI development cluster.
- Connecting through a compatible high-speed Ethernet switch for larger supported configurations.
Port capability and supported link speed depend on the selected cable, peer device, switch, firmware and supported topology. Do not assume that every QSFP cable or switch is compatible.
Can You Connect Multiple DGX Spark Systems?
Yes. NVIDIA describes connecting multiple DGX Spark systems using ConnectX-7 networking. Two systems can be connected directly with a supported QSFP cable, without an external switch. NVIDIA also documents a four-system configuration using a compatible 200GbE switch that supports RoCE v2 and the required QSFP56-DD connectivity.
| Cluster size | Networking approach | Key note |
|---|---|---|
| 1 system | Standalone DGX Spark | No cluster interconnect required |
| 2 systems | Direct QSFP cable connection | No external switch required |
| 3 systems | Supported direct / network topology | Follow NVIDIA’s current topology guidance |
| 4 systems | Compatible 200GbE switch with RoCE v2 | Check QSFP56-DD port and throughput requirements |
For a four-system setup, NVIDIA’s product guidance specifies a 200GbE switch with QSFP56-DD ports and RoCE v2 support. Confirm the latest compatibility list and exact cabling requirements before purchasing a switch.
What Can You Do with NVIDIA DGX Spark?
Local LLM development and inference
Developers can experiment with supported language models locally, build inference pipelines and test application behavior without sending every development request to a remote service.
Fine-tuning and model experimentation
DGX Spark can support fine-tuning workflows for compatible models, subject to model size, training method, precision, dataset and memory requirements. NVIDIA has described fine-tuning models of up to 70 billion parameters as a target use case; this is not a guarantee for every model or configuration.
AI agents and multimodal development
The platform can be used to prototype AI agents, retrieval-augmented generation, vision-language applications and other workflows supported by the NVIDIA software ecosystem.
NVIDIA DGX Spark vs Cloud GPU Resources
DGX Spark and cloud GPU instances solve different procurement problems. DGX Spark is a physical local development system with an upfront purchase cost, while cloud GPU resources are typically billed according to instance type, usage duration, storage and data transfer.
| Evaluation factor | DGX Spark | Cloud GPU |
|---|---|---|
| Access model | Local, dedicated hardware | Remote, service-based access |
| Cost model | Upfront purchase plus operating costs | Usage-based or reserved capacity |
| Data workflow | Local development and data handling | Data transfer and cloud storage planning |
| Scaling | Physical systems and supported cluster design | Provider-dependent instance scaling |
| Best fit | Persistent local AI development | Variable or burst workloads, subject to cloud terms |
The right option depends on utilization, model requirements, data governance, latency, availability, support and total cost of ownership.
NVIDIA DGX Spark Availability and Lead Time
Availability can vary by region, supplier allocation and the exact 128GB / 4TB configuration. Published availability in one market does not guarantee local inventory or a specific delivery date in another country.
For an up-to-date supply check, provide the destination country, quantity and required delivery window. Any estimated lead time should remain subject to supplier confirmation until the order is accepted.
Contact Sadid Gostaran for a current availability check and a configuration-specific quotation.
Check Availability via WhatsApp ↗NVIDIA DGX Spark FAQs
What is the price of NVIDIA DGX Spark 128GB with 4TB SSD?
Pricing varies by market, supply channel, destination and commercial terms. Contact Sadid Gostaran via WhatsApp for a current quotation.
Does NVIDIA DGX Spark have 128GB of GPU VRAM?
No. It has 128GB of coherent unified LPDDR5x system memory shared by the CPU and GPU. It is not a conventional discrete GPU with a separate 128GB VRAM pool.
Does DGX Spark include a 4TB SSD?
The 4TB configuration includes a self-encrypting NVMe M.2 storage drive. Confirm the exact SKU when ordering.
How powerful is the NVIDIA DGX Spark?
NVIDIA specifies up to 1 PFLOP of theoretical FP4 AI performance. Real application performance depends on model, software, precision and workload.
Can two DGX Spark systems be connected directly?
Yes. NVIDIA documents direct connection of two systems using a supported QSFP cable, without an external switch.
Can four DGX Spark systems be clustered?
NVIDIA describes a four-system configuration using a compatible 200GbE switch with QSFP56-DD ports and RoCE v2 support. Confirm current topology and compatibility requirements.
Is DGX Spark suitable for training large AI models?
It supports local AI development and fine-tuning workflows for compatible models. Training feasibility depends on model size, method, precision, dataset and memory requirements; it is not a substitute for every large-scale data-center training cluster.
How can I request a quotation?
Message Sadid Gostaran on WhatsApp with your destination country, quantity and preferred delivery window.
Is NVIDIA DGX Spark the Right AI System for You?
NVIDIA DGX Spark brings the Grace Blackwell platform, 128GB of coherent unified memory and up to 4TB of self-encrypting NVMe storage into a compact desktop AI system. It is particularly relevant for developers and teams that want a local environment for AI development, experimentation and supported fine-tuning workflows.
Before purchasing, compare the required model workload, memory needs, network topology, software compatibility, warranty and destination-specific supply terms. For current pricing and lead-time information, contact Sadid Gostaran directly.
Request an NVIDIA DGX Spark Quotation
Ask about the 128GB unified memory and 4TB SSD configuration, current availability, delivery estimates and procurement options.
+98 912 427 1734MESSAGE ON WHATSAPP ↗ WhatsApp inquiries only · Sadid GostaranAvailable in Iran
Looking for the NVIDIA DGX Spark product page for the Iranian market? Visit the Persian product page for local product information and purchasing options.
View Persian Product PageExplore more: NVIDIA product directory · AI infrastructure in Iran