AI Compute: NVIDIA DGX Spark
The NVIDIA DGX Spark is a compact desktop “AI supercomputer” built on the GB10 Grace Blackwell Superchip. It provides high-capacity on-premises AI compute for EQ Syntropy, with enough unified memory to run large reasoning models locally.
DGX Spark is a desktop developer and evaluation machine, not a hardened industrial unit (office-ambient temperature range and a consumer desktop build, not extended-temperature or industrially rated). It is the AI-compute analogue of the Lab Gateway: suited to prototyping, benchmarking, and high-capacity on-prem evaluation. For hardened, always-on industrial edge deployments, choose AGX Orin or Thor.
When to Choose DGX Spark
- Developer / data-science workstation: a workstation for building, fine-tuning, and benchmarking Syntropy models locally, with 128 GB of unified memory for models up to ~200B parameters (up to ~405B across two linked units).
- High-capacity on-prem AI in a non-industrial setting: a climate-controlled lab, office, or server room where you want data-local inference and large local models without the industrial hardening of an edge unit or the rack-mounting of a GPU server.
- On-prem, no cloud: data and inference stay local, with no per-query fees or usage limits.
To evaluate Syntropy before committing to hardware, the simpler options are EQ-Hosted (cloud) or Customer-Hosted (your existing GPU). For hardened field or edge deployment, use AGX Orin or Thor.
Platform Specifications
| Component | Specification |
|---|---|
| Superchip | NVIDIA GB10 Grace Blackwell |
| AI Performance | Up to 1 PFLOP (FP4) |
| GPU | NVIDIA Blackwell, 5th-gen Tensor Cores |
| CPU | 20-core Arm (10× Cortex-X925 + 10× Cortex-A725), up to 4.0 GHz |
| Memory | 128 GB unified LPDDR5X |
| Storage | 4 TB NVMe M.2 (self-encrypting) |
| Networking | ConnectX-7 200GbE (links two units for larger models) |
| OS | NVIDIA DGX OS (Ubuntu-based) |
| Form Factor | Desktop, 150 × 150 × 50.5 mm (not industrial-rated) |
Network Configuration
Like any gateway, DGX Spark connects to your facility LAN and to the EQ Wave sensor subnet via fiber media converters (the LAN side serves EQ Sight — see Deployment Options). A single sensor can connect directly; for several sensors, add a switch on the isolated sensor subnet — an unmanaged switch is sufficient, since the subnet uses fixed addressing. Its ConnectX-7 networking far exceeds what the sensor fabric requires.
AI Capabilities
With EQ Syntropy enabled, the Blackwell GPU accelerates physics-informed analytics, event detection, and diagnostic models, accessed through EQ Sight. Inference runs locally with no external connectivity required. See EQ Syntropy for AI feature details, and the AI Compute overview for the shared software stack and a platform comparison.
Current Status
EQ is validating the DGX Spark deployment. Contact [email protected] for status and early access.