Flopper.io Flopper.io

NVIDIA DGX Spark vs NVIDIA GB10 Grace Blackwell

Side-by-side specifications, performance, and rental pricing for datacenter AI workloads.

DGX Spark: Grace Blackwell, 2025 GB10 Grace Blackwell: Blackwell, 2025
Memory
128 vs 128 GB
Same capacity
Bandwidth
273 GB/s vs 273 GB/s
Even
TDP
140 W vs 180 W
DGX Spark draws 40 W less
Download comparison sheet

Share this page

NVIDIA DGX Spark vs NVIDIA GB10 Grace Blackwell

A post with this link shows the card on the Image tab as its preview.

Post on X Share on LinkedIn
 Printable version →

NVIDIA DGX Spark

Full specs →

What is the NVIDIA DGX Spark?

The DGX Spark is an APU from NVIDIA on the Grace Blackwell architecture, an processor that pairs CPU and GPU on one chip. DGX Spark full specifications →

Can I rent the DGX Spark in the cloud?

Flopper is not tracking any live rental listings for the DGX Spark yet. The provider directory lists the clouds that rent comparable accelerators on demand. GPU cloud providers →

How does the DGX Spark compare with other accelerators?

This page sets the DGX Spark against the GB10 Grace Blackwell. Both parts launched in the same year. Both are NVIDIA parts. Flopper also compares it head to head with the Ryzen AI Max+ 395 Radeon 8060S (Strix Halo), Jetson AGX Thor T5000 128GB and M5 Ultra, across every published precision, memory and power. DGX Spark vs Ryzen AI Max+ 395 Radeon 8060S (Strix Halo) →

Where do the DGX Spark figures come from?

Every DGX Spark figure on this page traces to NVIDIA's own datasheet or specification table. How Flopper verifies specifications →

NVIDIA GB10 Grace Blackwell

Full specs →

What is the NVIDIA GB10 Grace Blackwell?

The GB10 Grace Blackwell is an APU from NVIDIA on the Blackwell architecture, an processor that pairs CPU and GPU on one chip. It is integrated into a larger processor rather than sold as a separate card. GB10 Grace Blackwell full specifications →

Can I rent the GB10 Grace Blackwell in the cloud?

Flopper is not tracking any live rental listings for the GB10 Grace Blackwell yet. The provider directory lists the clouds that rent comparable accelerators on demand. GPU cloud providers →

How does the GB10 Grace Blackwell compare with other accelerators?

This page sets the GB10 Grace Blackwell against the DGX Spark. Both parts launched in the same year. Both are NVIDIA parts. The compare tool builds the same comparison for any two parts in the catalogue. All GPU comparisons →

Where do the GB10 Grace Blackwell figures come from?

NVIDIA published the GB10 Grace Blackwell figures in marketing material rather than a datasheet, and Flopper has no independent confirmation of them. How Flopper verifies specifications →

Specifications

DGX Spark
GB10 Grace Blackwell
Architecture
Grace BlackwellBlackwell
Launch Year
20252025
Form Factor
Desktop WorkstationIntegrated
Memory
128 GB128 GB
Memory Bandwidth
273 GB/s273 GB/s
TDP
140 W180 W
Process Node
4nm—

Performance (TFLOPS)

DGX Spark
GB10 Grace Blackwell
No verified data available No verified data available
No verified data available31 TFLOPS
No verified data available No verified data available
No verified data available No verified data available
No verified data available No verified data available
No verified data available No verified data available
FP6
No verified data available No verified data available
500 TFLOPS
1,000 TFLOPS sparse
500 TFLOPS
1,000 TFLOPS sparse
No verified data available No verified data available

Figures are dense. Where a vendor publishes a figure it sits under the dense one and never decides a winner. Hover any precision for its definition, or read the GPU glossary.

FLOPS by Precision

What actually differs

Pick a GB10 box on its enclosure, because the chip is the same in all of them. The GB10 page carries across every system built on it: 128 GB of unified memory at 273 GB/s, the FP4 rating and the model sizes that fit. Those are settled the moment you choose GB10, whoever makes the case.

The DGX Spark page describes NVIDIA’s reference build, and power shows where the enclosure matters. NVIDIA rates the complete system at 140 W with a 240 W maximum. Those figures belong to NVIDIA’s box, not the chip: other makers set their own cooling and power budgets for their GB10 systems.

So compare GB10 systems on what each maker publishes for its enclosure, and treat the chip figures as a shared baseline. For a different 128 GB chip, see DGX Spark vs Ryzen AI Max+ 395.

New to the terms? The glossary explains FP32, FP4, Sparsity and TFLOPS. For how the generations relate, see NVIDIA architectures explained.

Frequently asked questions

What is the difference between the DGX Spark and the GB10?

The GB10 is the Grace Blackwell superchip; the DGX Spark is NVIDIA's own desk-side system built around it. Memory, bandwidth and FP4 throughput belong to the chip and match on both pages. The DGX Spark page adds what NVIDIA's enclosure defines, such as its 140 W system rating and 240 W maximum.

Is the DGX Spark faster than other GB10 systems?

Not on paper. Every GB10 system shares 128 GB of unified memory at 273 GB/s and 500 TFLOPS FP4 dense, or 1,000 TFLOPS FP4 with sparsity. Differences between makers come from the enclosure, mainly cooling and power limits, so compare those on each maker's datasheet rather than the chip figures.

How much power does a GB10 system use?

NVIDIA rates the DGX Spark at 140 W for the whole system, with a 240 W maximum. Other makers set their own cooling and power budgets for their GB10 boxes, so check each maker's datasheet rather than assuming the DGX Spark rating carries over.

Can you rent a DGX Spark or a GB10?

Flopper lists no rental listings for either. The GB10 ships only inside desk-side systems, the DGX Spark among them, and you buy the whole machine outright. To test a large model before buying, rent a datacenter GPU with similar memory by the hour.

Get Comparison Updates

New GPUs added weekly. Be the first to see how they compare.

Open in Full Compare Tool → DGX Spark full specs GB10 Grace Blackwell full specs Estimate cluster cost in the calculator →

© 2026 Flopper.io - Compare the GPUs Powering AI