The Open-Source AI Stack
RSS
All hardware

Hardware · x86 unified

NVIDIA DGX Spark (GB10)

NVIDIA

Grace-Blackwell developer appliance with 128 GB coherent unified memory and the full CUDA stack. About 96 GB is addressable as VRAM. Capacity is generous; the 273 GB/s bandwidth is the decode limiter.

CPU · GPU · NPU 273 GB/s memory bus 128 GB Unified memory
x86 unified. Bus width tracks bandwidth (273 GB/s, sets decode speed); the box tracks capacity (128 GB, sets what fits). Unified memory is one pool shared by CPU and accelerator.

Specs

Memory
128 GB Unified LPDDR5X
Bandwidth
273 GB/s
Power
140 W
Form factor
soc
Interconnect
none
Released
2025-10

What it runs (single unit, Q4_K_M, 4K context)

Model Fits? Decode ceiling
Llama 3.1 8B Instruct yes ~54 tok/s
Llama 3.3 70B Instruct yes ~6.7 tok/s
Qwen 2.5 72B Instruct yes ~6.5 tok/s
DeepSeek-V3 no —

Ceiling is the theoretical rooflineruntimeA performance model that bounds throughput by either compute or memory bandwidth, whichever is the limiting resource for an operation's arithmetic intensity. Open full entry ; open the explorer to set quant, context, and runtime and see the realistic range.

Sources