The system uses the same GB10 Grace Blackwell Superchip as the 128GB DGX Spark. The main difference is unified memory capacity.
DGX Spark 64GB vs. 128GB
| Specification | DGX Spark 64GB | DGX Spark 128GB |
| Superchip | NVIDIA GB10 | NVIDIA GB10 |
| CPU | 20-core Arm | 20-core Arm |
| GPU architecture | Blackwell | Blackwell |
| CUDA cores | 6,144 | 6,144 |
| Unified memory | 64GB | 128GB |
| Memory bandwidth | 273 GB/s | 273 GB/s |
| AI compute | Up to 1 PFLOP FP4 | Up to 1 PFLOP FP4 |
| Networking | ConnectX-7 | ConnectX-7 |
| Maximum model size | Up to ~100B parameters | Up to ~200B parameters |
| Price | From $4,999 | $6,950 Founders Edition |
| Original launch price | — | $3,999 |
The 64GB model keeps the same processor, memory bandwidth, networking and NVIDIA software stack as the 128GB version.
Its main limitation is model size. NVIDIA positions the 64GB system for local AI models of up to roughly 100 billion parameters, while the 128GB version can handle models up to around 200 billion parameters.
Two 64GB DGX Spark systems can also be connected through ConnectX-7. Together they provide:
- 128GB combined memory;
- roughly twice the memory bandwidth;
- support for models up to around 200B parameters;
- up to 1.7× higher performance in supported workloads.
Pricing has changed sharply
The original 128GB DGX Spark launched at $3,999. The current 128GB Founders Edition is priced at $6,950, while the new 64GB model starts at $4,999. That means the new 64GB system costs $1,000 more than the original 128GB model did at launch.
Half the memory now costs more than the original 128GB DGX Spark launch price.
NVIDIA has previously cited constraints in memory supply as a reason for DGX Spark price increases.
The 64GB model therefore lowers the current entry price for GB10 hardware, but it is significantly more expensive per gigabyte of unified memory than the original DGX Spark.
