Skip to main content

Ko ar to var palaist?

No cloud subscription. Your data stays on your hardware.

Large Language Models

Image Generation

ML Training & Fine-tuning

Private AI / GDPR Compliance

Video AI

Software Development & Research

Tehniskās specifikācijas

Skaitļošana
GPU modelis NVIDIA Grace Blackwell
Arhitektūra Grace Blackwell
Tensor veiktspēja Up to 1 PFLOP FP4
Tensor kodoli 5th Generation
RT kodoli 4th Generation
CUDA kodoli Blackwell Generation
Atmiņa un krātuve
Vienotā atmiņa 128 GB LPDDR5x
Atmiņas joslas platums 273 GB/s
Atmiņas interfeiss 256-bit
Krātuve 4 TB NVMe M.2 (self-encrypting)
Savienojamība
Tīkla karte ConnectX-7 @ 200 Gbps
Ethernet 1x RJ-45, 10 GbE
Wi-Fi Wi-Fi 7
Bluetooth BT 5.4
USB porti 4x USB Type-C
Sistēma
CPU 20-core Arm (10x Cortex-X925 + 10x Cortex-A725)
Enerģijas patēriņš 240W
Formas faktors Desktop
Programmatūras komplekts NVIDIA AI Enterprise (PyTorch, TensorFlow, JAX, RAPIDS, TensorRT)
Lietošanas gadījumi Large Language Models, Image Generation, ML Training & Fine-tuning, Private AI / GDPR Compliance, Video AI, Software Development & Research
Ietvari PyTorch, TensorFlow, JAX, RAPIDS, TensorRT, ONNX, Docker

NVIDIA DGX Spark Details

NVIDIA DGX Spark

Supercomputer-class AI, on your desk.

Built on the Grace Blackwell architecture — the same generation powering the world's largest AI data centres — DGX Spark delivers data-centre AI in a desktop form factor, drawing the power of a small appliance.

1 PFLOP
FP4 AI performance
240 W
Total power draw
128 GB
Unified LPDDR5x
Coherent Unified Memory

One address space. No PCIe bottleneck.

The CPU and GPU share 128 GB of LPDDR5x at full bandwidth. Large language models and multimodal workloads that demand two GPU nodes on conventional hardware run in a single coherent memory space here — no copies, no partitioning.

Ready Out of the Box

Plug in and train.

Ships with NVIDIA AI Enterprise pre-installed. PyTorch, TensorFlow, TensorRT, and RAPIDS are ready the moment you power on. The ConnectX-7 NIC links nodes at 200 Gbps without an external switch — and you can scale out to DGX Cloud when the job grows.

PyTorch TensorFlow TensorRT RAPIDS 200 Gbps ConnectX-7
Grace Blackwell · Desktop AI Supercomputer