We take cryptocurrency — so the coins you hold can buy real hardware.Pay in crypto, spend it on real hardware. Free insured shipping worldwide over $399. Plain, unbranded boxes, sent from Germany.Free insured shipping over $399. Your rate is locked for 30 minutes once checkout opens.Rate locked 30 minutes. Every item sealed, serial-checked and covered by a 24-month warranty.Sealed, 24-month warranty.

Saved items Sign in
Compact silver mini PC, cube shape, smooth matte finish

PNY Technologies, Inc.

PNY Technologies, Inc. NVDGXSPARK-PB barebone mini PC 128 GB LPDDR5x 4 TB NVME

SKU CH-3WPC-WF22V

Barebone mini PC with 20-core Arm CPU, 128 GB LPDDR5x unified memory and 4 TB NVME storage delivering up to 1 petaFLOP AI performance

  • CPU Type20 core Arm, 10 Cortex-X925 + 10 Cortex-A725 Arm
  • CPU Cores20-Core
  • Memory installed128GB LPDDR5x, unified system memory
  • Storage Installed4TB NVME M.2 SSD with self-encryption
  • Discrete Graphic CardNVIDIA Blackwell Architecture
  • Max LAN Speed10Gbps

Option Single

  • Up to 1 petaFLOP AI performance at FP4 precision
  • 20-core Arm CPU with 10 Cortex-X925 and 10 Cortex-A725 cores
  • 128 GB unified LPDDR5x memory for coherent data access
  • 4 TB NVME storage with self-encryption
  • ConnectX-7 Smart NIC enables 10 Gbps networking and dual-system scaling to 405 billion parameter models

Compact AI workstation

The PNY NVDGXSPARK-PB is a barebone mini PC built on the NVIDIA GB10 Grace Blackwell Superchip. It delivers up to 1 petaFLOP of AI performance at FP4 precision for researchers and developers working with large language models. The system integrates a 20-core Arm CPU with 10 Cortex-X925 and 10 Cortex-A725 cores alongside a Blackwell architecture GPU. It targets users who need desktop-scale AI compute in a 1.99 by 5.91 by 5.91 inch enclosure weighing 2.65 lbs.

Unified memory and storage limits

The platform provides 128 GB of coherent LPDDR5x unified system memory shared between CPU and GPU. This capacity supports models up to 200 billion parameters on a single unit. The installed 4 TB NVME M.2 SSD features self-encryption for data at rest. Networking reaches 10 Gbps over RJ45 and 802.11be Wi-Fi 7 wireless, while the ConnectX-7 Smart NIC enables linking two systems for models up to 405 billion parameters.

Platform compatibility checks

Verify that your workloads run on the 20-core Arm architecture before ordering. Confirm that 128 GB of unified memory meets your model size requirements, as the memory is not expandable. Ensure your software stack supports NVIDIA Blackwell GPU features including fifth-generation Tensor Core technology. Check that the single HDMI 2.1a output and four rear USB Type-C ports satisfy your display and peripheral needs.

Highlights

  • Up to 1 petaFLOP AI performance at FP4 precision
  • 20-core Arm CPU with 10 Cortex-X925 and 10 Cortex-A725 cores
  • 128 GB unified LPDDR5x memory for coherent data access
  • 4 TB NVME storage with self-encryption
  • ConnectX-7 Smart NIC enables 10 Gbps networking and dual-system scaling to 405 billion parameter models

Specifications

BrandNVIDIA
ModelNVDGXSPARK-PB
CPU Type20 core Arm, 10 Cortex-X925 + 10 Cortex-A725 Arm
CPU Cores20-Core
Memory installed128GB LPDDR5x, unified system memory
Storage Installed4TB NVME M.2 SSD with self-encryption
Discrete Graphic CardNVIDIA Blackwell Architecture
Max LAN Speed10Gbps
Wireless LAN802.11be
WiFi GenerationWi-Fi 7
BluetoothBluetooth 5.3
HDMI1 x HDMI 2.1a
Rear USB4 x USB Type C
RJ451
Other Rear Ports1 x ConnectX-7 Smart NIC
Dimensions1.99" x 5.91" x 5.91"
Weight2.65 lbs.
PackagingNVIDIA DGX Spark
Quick Start Card
Support Card
Power Supply
Shipping weight3.08 kg
Package size241 × 241 × 146 mm

Other versions of this model

1 more in the same range

Same family, different memory, cooler or clock — compare before you commit.

Questions about this item

How do I decide between the 128 GB LPDDR5x unified memory and the 4 TB NVME M.2 SSD for my workload?

The 128 GB LPDDR5x unified memory is shared by the Grace CPU and Blackwell GPU for model inference, while the 4 TB NVME M.2 SSD with self-encryption stores datasets and checkpoints locally.

What does the 1 petaFLOP FP4 figure mean when I run models on this system?

It represents peak AI throughput at FP4 precision from the Blackwell GPU fifth-generation Tensor Cores, enabling up to 200 billion parameter models in the unified 128 GB memory.

How does the ConnectX-7 Smart NIC and 10 Gbps LAN affect multi-system scaling?

The ConnectX-7 Smart NIC and 10 Gbps RJ45 link let you connect two DGX Spark units, combining memory to support models up to 405 billion parameters.

Also in this aisle

People compared these

Same shelf, same checkout — eight coins and a 30-minute rate lock.