Quick Answer
Nvidia’s 64GB DGX Spark is a lower-memory desktop AI computer that starts at $4,999 and goes on sale October 23, 2026, through six named manufacturers. Nvidia says it runs supported models up to 100 billion parameters locally, while two linked units can pool 128GB for models up to 200 billion parameters. Confirm the final seller configuration before buying.
Key Takeaways
- The Nvidia DGX Spark 64GB starts at $4,999 in the United States.
- Acer, Asus, Dell, Gigabyte, HP, and MSI will offer the new configuration.
- Nvidia says one system can run supported models up to 100 billion parameters locally.
- Two systems can pool 128GB of memory through NVIDIA Sync Cluster Assistant.
- The 64GB model uses the same GB10 Grace Blackwell Superchip as the 128GB DGX Spark.
What is the Nvidia DGX Spark 64GB?
The Nvidia DGX Spark 64GB is a new desktop AI computer configuration with 64GB of unified memory and a starting price of $4,999. Nvidia announced the system on October 2, 2026, and said the lower-memory model uses the same GB10 Grace Blackwell Superchip, DGX OS, and Nvidia AI software stack as the existing 128GB DGX Spark. Nvidia’s announcement sets out the October 23 availability date and the company’s local-model claims.
The most important difference is the amount of unified memory available for AI workloads. Nvidia positions the 64GB configuration as a lower-cost way to run supported AI models locally, rather than sending inference work to a cloud service. That local approach can matter for users whose work requires an on-device setup, but Nvidia’s model-size figures apply to supported workloads and do not establish that every model or software workflow will perform the same way.
The 64GB DGX Spark is not a conventional consumer PC replacement based on the information Nvidia has provided. The product is a desktop system aimed at local AI work, and prospective buyers should evaluate its memory capacity, local software requirements, and seller configuration before treating the $4,999 starting price as a complete setup cost.
When does the Nvidia DGX Spark 64GB go on sale?
The Nvidia DGX Spark 64GB goes on sale Friday, October 23, 2026. Nvidia says Acer, Asus, Dell, Gigabyte, HP, and MSI will offer the configuration, which means availability will come through partner systems rather than a single Nvidia retail listing.
The October 23 date matters because the $4,999 figure is a starting price, not a confirmed final price for every manufacturer’s version. A seller may offer different storage choices or other configuration differences, so shoppers should compare the specific system being sold instead of assuming every listing matches the entry-level price. Independent launch coverage also identifies October 23 and the $4,999 starting point. Tom’s Hardware’s launch coverage confirms the announced schedule.
The practical response is to wait for the manufacturer listing that matches the intended workload. Buyers who need more than the stated local-model capacity should compare the single-system configuration against Nvidia’s option to connect two 64GB systems, rather than buying one unit solely because it has the lowest entry price.
How much local AI can one 64GB DGX Spark run?
The Nvidia DGX Spark 64GB can run supported models up to 100 billion parameters entirely on-device, according to Nvidia. The company says this arrangement avoids cloud inference for supported workloads, which means the model runs on the local DGX Spark rather than requiring a cloud inference service for that stated use case.
The parameter figure is useful as a capacity guide, but it is not a complete performance measurement. Nvidia’s announcement does not say that every model at or below 100 billion parameters will have identical speed, software support, or behavior. The practical limit depends on whether the model and local software workflow are supported by the DGX Spark environment.
Local AI also does not eliminate the need to consider data handling. Users working with sensitive prompts should establish what the local application stores, which accounts connect to the workflow, and whether any optional service sends data elsewhere. Readers assessing AI privacy practices can apply the same caution used for sensitive AI prompts, even when the model itself runs on local hardware.
What hardware and connectivity does the DGX Spark platform include?
The DGX Spark platform includes a 20-core Arm CPU, Wi-Fi 7, Bluetooth 5.4, four USB-C ports, up to 4TB of NVMe storage, and a 200Gbps ConnectX-7 network interface. Nvidia lists those platform specifications on its product page, although the company’s October 2 announcement does not specify which storage configuration every 64GB partner system will include. Nvidia’s DGX Spark specifications page provides the platform-level hardware details.
| DGX Spark capability | What Nvidia lists | Why it matters |
|---|---|---|
| Unified memory | 64GB on the new configuration | Nvidia says one system supports models up to 100 billion parameters for supported local workloads. |
| Processor platform | GB10 Grace Blackwell Superchip and 20-core Arm CPU | The 64GB model uses the same GB10 platform named for the 128GB DGX Spark. |
| Storage | Up to 4TB NVMe | Buyers should verify the storage amount in the exact partner configuration. |
| Networking | 200Gbps ConnectX-7 interface | The platform supports the connectivity Nvidia describes for linked-system use. |
| Wireless | Wi-Fi 7 and Bluetooth 5.4 | The system includes current wireless connectivity alongside its wired interface. |
The hardware list clarifies that the DGX Spark is a complete desktop platform, not merely a memory module or accelerator card. At the same time, “up to 4TB” is a platform maximum rather than a guarantee for every $4,999 system. Confirm the storage capacity, included accessories, and software support on the selected manufacturer’s product page before placing an order.
Can two 64GB DGX Spark systems work together?
Two Nvidia DGX Spark 64GB systems can link through NVIDIA Sync Cluster Assistant to pool 128GB of memory and run models up to 200 billion parameters, according to Nvidia. The paired configuration expands the stated model-size ceiling beyond a single 64GB system, which gives buyers a defined path when one system’s memory is insufficient.
Nvidia says two linked 64GB systems delivered up to 1.7 times the performance of one system in the company’s Qwen 3.8 27B test. That result is self-reported, applies to Nvidia’s named test, and should not be treated as a universal performance result for every model, prompt pattern, or application. Buyers should use the figure as an indication of Nvidia’s scaling claim rather than as a guaranteed speed increase.
Two-system setups also change the purchasing decision. Two systems at the announced starting price would begin at $9,998 before any configuration differences, although final partner prices may vary. The most sensible approach is to compare a single 64GB system, two linked 64GB systems, and the 128GB DGX Spark only after identifying the largest model and local workflow required.
Which local AI tools are coming to the 64GB DGX Spark?
Nvidia Sync Model Launcher is due at the end of October 2026 and is designed to let users download and launch Qwen3.8 27B locally. Nvidia also says the setup will include OpenCode in a browser, which identifies a planned software path for users who want to begin with the company’s supported local workflow.
The planned launcher matters because local AI hardware requires a usable way to obtain and start compatible models. A model launcher can reduce setup work for the specific software Nvidia names, but the announcement does not establish support for every local model tool or every developer environment. Buyers should verify the final launcher availability and supported-model list when the software arrives.
Local AI tools also need operational limits, particularly when a system can access files, code, or connected services. Readers considering agent-style workflows should apply safeguards such as permission limits and review steps, which are also central to AI agent safety controls. A local deployment can change where inference runs, but it does not remove the need to review what an AI tool is authorized to do.
Who should consider the Nvidia DGX Spark 64GB?
The Nvidia DGX Spark 64GB is most relevant to buyers who need Nvidia’s stated local capacity for supported models up to 100 billion parameters and want a desktop system that starts below the 128GB configuration. The $4,999 starting price makes the 64GB version a distinct entry point in the DGX Spark lineup, while the shared GB10 Grace Blackwell Superchip and software stack preserve the platform Nvidia describes for the larger-memory model.
The Nvidia DGX Spark 64GB is less straightforward for buyers who only need occasional cloud AI access or who have not identified a supported local model workflow. The product’s value depends on using its local memory capacity, software stack, and hardware platform for a real workload. A lower starting price does not make 64GB interchangeable with 128GB for users whose model requirements exceed Nvidia’s stated single-system capacity.
The practical buying checklist has 4 parts: verify the exact partner configuration, identify the largest supported model needed, determine whether one or two systems are necessary, and confirm the local software path. Users who also need broader AI platform context can consider how private AI memory systems approach data handling differently from a dedicated local desktop computer.
FAQ
How much does the Nvidia DGX Spark 64GB cost?
The Nvidia DGX Spark 64GB starts at $4,999 in the United States. Nvidia announced that price on October 2, 2026, but final pricing can vary by manufacturer configuration.
When will the Nvidia DGX Spark 64GB be available?
The Nvidia DGX Spark 64GB will be available on October 23, 2026. Nvidia says Acer, Asus, Dell, Gigabyte, HP, and MSI will offer systems.
Can the Nvidia DGX Spark 64GB run AI models locally?
The Nvidia DGX Spark 64GB can run supported models up to 100 billion parameters entirely on-device, according to Nvidia. The stated limit applies to supported workloads, so buyers should confirm compatibility for their intended model and software tools.
Can two Nvidia DGX Spark 64GB systems be connected?
Two Nvidia DGX Spark 64GB systems can connect through NVIDIA Sync Cluster Assistant and pool 128GB of memory. Nvidia says the linked setup can run models up to 200 billion parameters.
Does the Nvidia DGX Spark 64GB have the same chip as the 128GB model?
The Nvidia DGX Spark 64GB uses the same GB10 Grace Blackwell Superchip as the 128GB DGX Spark. The main stated configuration difference is the 64GB unified-memory capacity.