Description
Desktop AI supercomputer | 1 petaFLOP FP4 performance | 128 GB unified memory | NVIDIA Grace Blackwell architecture | Prototype, fine-tune and deploy large AI models locally
Quick Specs
- Achieve up to 1 petaFLOP of AI compute performance at FP4 precision with the NVIDIA Grace Blackwell architecture — data-centre power on your desk.
- Run large AI models of up to 200 billion parameters locally, backed by 128 GB of unified LPDDR5x system memory and 273 GB/s memory bandwidth.
- Store massive datasets and model checkpoints on a 4 TB self-encrypting NVMe M.2 drive — fast, capacious, and secure.
- Connect two DGX Spark units via the ConnectX-7 Smart NIC to tackle AI models of up to 405 billion parameters in a collaborative desktop cluster.
- Work wirelessly and freely with Wi-Fi 7 and Bluetooth 5.3, while the compact 15 × 15 × 5.05 cm form factor keeps your workspace uncluttered.
A new class of computer for AI builders and researchers
The PNY NVIDIA DGX Spark GB10 redefines what a desktop workstation can do. Designed for developers, data scientists, and AI researchers who need real supercomputer capability without a server room, it delivers up to 1 petaFLOP of AI compute at FP4 precision from a compact, champagne-gold enclosure that weighs just 1.2 kg. Powered by the NVIDIA Grace Blackwell architecture — the same generation driving the world's most advanced data centres — the DGX Spark lets you prototype, fine-tune, and run inference on large language models and generative AI workloads entirely on your desktop, then deploy seamlessly to the data centre or cloud when you are ready to scale.
128 GB of unified memory: no model too large for your desk
Tired of hitting memory walls mid-training and waiting on cloud queues to continue your work? The DGX Spark's 128 GB of LPDDR5x unified system memory — shared coherently between the CPU and GPU — removes that bottleneck entirely. With a 256-bit memory interface delivering 273 GB/s of bandwidth, you can run AI models of up to 200 billion parameters locally in a single system. Link two DGX Spark units together via NVIDIA ConnectX networking and the combined capacity extends to AI models of up to 405 billion parameters, giving your team a compact, high-performance AI cluster without the infrastructure overhead.
Grace Blackwell silicon: 5th-generation Tensor Cores at your fingertips
At the heart of the DGX Spark is the NVIDIA GB10 Superchip, combining a 20-core Arm CPU — ten Cortex-X925 performance cores paired with ten Cortex-A725 efficiency cores — with a Blackwell-generation GPU featuring 5th-generation Tensor Cores and 4th-generation RT Cores. This architecture is purpose-built for the compute demands of modern AI: transformer training, large-scale inference, retrieval-augmented generation, and multi-modal workloads all benefit from the raw arithmetic throughput that previously required a rack of servers. The preinstalled NVIDIA AI software stack — including frameworks, libraries, and pretrained models — means you can be productive from the moment you power on.
Enterprise-grade connectivity in a mini form factor
Despite its compact footprint, the DGX Spark is exceptionally well connected. Four USB Type-C ports (three available for peripherals, one powering the unit), a single HDMI 2.1a output supporting up to 8K displays at 120 Hz with HDR, a 10 GbE RJ-45 port for reliable wired networking, and a ConnectX-7 Smart NIC capable of up to 100 GbE ensure that data moves as fast as your models demand. Wi-Fi 7 and Bluetooth 5.3 keep the unit cable-light when you need flexibility. The 4 TB self-encrypting NVMe M.2 drive safeguards your proprietary models and datasets at rest, meeting the security expectations of professional AI development environments.
Features
- NVIDIA Grace Blackwell architecture delivers up to 1 petaFLOP of FP4 AI compute performance from a desktop-sized unit.
- 128 GB of unified LPDDR5x system memory with 273 GB/s bandwidth supports AI models of up to 200 billion parameters in a single system.
- 5th-generation Tensor Cores and 4th-generation RT Cores accelerate transformer training, inference, and generative AI workloads.
- ConnectX-7 Smart NIC enables dual-unit networking for AI models of up to 405 billion parameters across two connected DGX Spark systems.
- 4 TB self-encrypting NVMe M.2 storage keeps proprietary model data and datasets secure and immediately accessible.
- Four USB Type-C ports, HDMI 2.1a (8K/120 Hz HDR), 10 GbE RJ-45, Wi-Fi 7, and Bluetooth 5.3 for comprehensive desktop connectivity.
- Preinstalled DGX OS with the full NVIDIA AI software stack — frameworks, libraries, and pretrained models included out of the box.
- Compact mini form factor (15 × 15 × 5.05 cm, 1.2 kg) in a distinctive champagne-gold finish.
Specifications
- Architecture: NVIDIA Grace Blackwell
- GPU: Blackwell Architecture
- CPU: 20 core Arm, 10 Cortex-X925 + 10 Cortex-A725
- CUDA Cores: Blackwell Generation
- Tensor Cores: 5th Generation
- RT Cores: 4th Generation
- Tensor Performance: 1 PFLOP
- System Memory: 128GB LPDDR5x, unified system memory
- Memory Interface: 256-bit
- Memory Bandwidth: 273 GB/s
- OS: DGX OS
- Storage: 4 TB NVME.M2 with self-encryption
- USB: 4x USB Type-C
- Ethernet: 10 GbE 1x RJ-45 connector
- NIC: ConnectX-7 Smart NIC
- Wi-Fi: Wi-Fi 7
- Bluetooth: BT 5.3
- Audio output: HDMI® multichannel audio output
- Power Consumption: TBD
- Display Connectors: 1x HDMI®™ 2.1a
- NVENC | NVDEC: 1x | 1x
- Product Dimensions: 15 x 15 x 5 cm
- Product Weight: 1200g
What's in the Box
- NVIDIA DGX Spark – Gold/Champagne x1
- Power Supply x1
- Quick Start Card x1
- Support Card x1
Why PNY
PNY has built its reputation over decades as a trusted provider of professional NVIDIA hardware, supplying developers, researchers, and enterprise teams with the GPU and AI compute technology they need to push the boundaries of possibility. As an authorised NVIDIA partner, PNY brings world-class AI supercomputing hardware to South African professionals and creators with the confidence of genuine products, dedicated technical support, and a 12-month limited warranty.
