On this pageOfficial specificationsOfficial price changedWhat community results suggestCheck your software and connectivity needsSources
DT

Written and reviewed by DeviceterraPractical guidance · Updated November 11, 2026

THE SHORT ANSWER

DGX Spark is a developer workstation for experimenting with larger models and NVIDIA’s stack. Its 128 GB capacity is the draw; do not assume it will outperform a high-end discrete GPU on every model or workload.

THE FULL GUIDE

Here is what matters, why it matters and how to check it on your own setup.

01

Official specifications

NVIDIA lists DGX Spark with a 20-core Arm CPU, Blackwell GPU, 128 GB coherent LPDDR5X unified memory, 273 GB/s bandwidth, up to 4 TB NVMe, and a 240 W power supply. The compact system connects to a monitor and network and ships with DGX OS.

NVIDIA's 'up to 1 PFLOP FP4' is a theoretical performance figure using sparsity. It is not a measured token-generation rate. Compare actual model benchmarks instead.

02

Official price changed

NVIDIA announced a Founders Edition MSRP change from $3,999 to $4,699 in February 2026 because of memory supply constraints. OEM GB10 computers can have different pricing. Check the exact configuration and authorized retailer availability in your country; import duty and shipping can materially change the final price.

03

What community results suggest

Community benchmark roundups contain multiple GB10/DGX Spark builds and model results, but configurations vary in model format, runtime, context, and software version. Treat each row separately and look for the raw benchmark. Capacity does not equal speed: a 128 GB system can load more weights than a 24 GB GPU, while the GPU's memory bandwidth is much lower than a high-end discrete card's.

The platform makes sense when you need NVIDIA's integrated software environment, a compact developer box, and the ability to load larger models. If you only run models that fit in 12–24 GB, compare a GPU desktop's speed and upgrade path.

04

Check your software and connectivity needs

  • Check that model formats and serving frameworks support the Arm CPU and GB10 software stack.
  • Confirm whether your workload is inference, fine-tuning, development, or multi-node experimentation.
  • Include external display, storage, network, and support needs in the budget.
  • Ask whether the 64 GB OEM configuration is available in your region and how its price compares.
  • Do not buy based only on the 'supercomputer' name or theoretical FP4 TOPS.
PLAIN ENGLISH

Words you may see in this guide

RAM
The working memory used by your computer and apps.
VRAM
The memory on a dedicated graphics card. Shared graphics may use system RAM.
Context
The text and other input the model can use for its current answer.
Quantization
A way to make model weights smaller. It can change memory use and answer quality.
Engine
The software that loads the model and runs it.
RAG
A workflow that finds useful document passages before asking a model to answer.
Benchmark
A repeatable test. A result is useful only when you know the setup and task.
RESEARCH SOURCES

Official facts and real user evidence

Official documentation supports product and model facts. Community discussions show real setups, failures, and questions. A community result is supporting evidence, not a promise that another computer will perform the same way.

YOUR NEXT STEP

Watch a real test. Check what fits your computer.

See practical local AI tests from DeviceTerra, then use MamiLens to build a hardware-aware shortlist for your own setup.

This guide is educational. Model software, licenses, and hardware support can change. Check official sources before an important deployment.