Skip to content

DGX Spark 64GB Price: $4,999 PC Arrives October 23

NVIDIA's DGX Spark 64GB arrives October 23 from $4,999 with the same GB10 Grace Blackwell platform as the 128GB model, but less unified memory.

DGX Spark 64GB Price: $4,999 PC Arrives October 23

On this page

NVIDIA is adding a 64GB version of its DGX Spark desktop AI computer on October 23, with systems from Acer, ASUS, Dell, Gigabyte, HP and MSI starting at $4,999. The unusual part is that this is not simply a cheaper version of the existing machine: it keeps the same GB10 Grace Blackwell Superchip and software stack while cutting unified memory from 128GB to 64GB.

The DGX Spark 64GB keeps the expensive parts intact

The new configuration uses the same Grace Blackwell platform as the larger DGX Spark, combining a 20-core Arm processor with a Blackwell GPU, unified system memory and NVIDIA's local artificial intelligence software stack. NVIDIA says the 64GB machine can run models with up to 100 billion parameters entirely on the device, which makes the memory reduction more significant than a simple specification downgrade. The machine is designed for developers and researchers who want to run inference, fine-tuning and AI agents locally instead of sending every workload to a cloud service.

Unified memory is particularly useful here because the processor and graphics hardware draw from the same memory pool rather than keeping separate CPU and GPU memory. That means the 64GB figure is not equivalent to a conventional desktop with 64GB of system RAM plus a separate graphics card with its own video memory. The available memory still has to accommodate the operating system, model weights, cache and other software, so the maximum model size NVIDIA advertises should not be read as a guarantee that every 100-billion-parameter model will run comfortably.

The $4,999 price needs more context

At $4,999, the new DGX Spark 64GB is not cheap by normal desktop standards. It is also $300 above NVIDIA's previously announced $4,699 Founders Edition price for the 128GB DGX Spark, although the official NVIDIA marketplace has shown that 128GB model at $4,699 and with stock availability varying. NVIDIA raised that Founders Edition price from its original $3,999 level in February because of memory supply constraints.

That comparison explains why the new 64GB system should not be described simply as a bargain DGX Spark. Its $4,999 starting price buys half the unified memory of the original configuration, while the existing 128GB hardware has become considerably harder to price consistently across NVIDIA's own store and partner systems. Independent coverage also reports a much higher current price for some 128GB configurations, showing why the real comparison will depend on which manufacturer and configuration a buyer can actually obtain.

One 64GB Spark is aimed at models that fit the smaller memory pool

NVIDIA's argument for the new configuration is that many current open models are small enough to make 64GB useful for local development. The company says the system can handle models up to 100 billion parameters, while its existing 128GB platform is designed for models up to 200 billion parameters in a single system. In practical terms, that makes the 64GB version more suitable when the model, operating system and supporting workload fit comfortably inside the available memory, rather than when a developer routinely needs the largest models possible.

The software is also largely unchanged. NVIDIA lists its Agent Toolkit, CUDA-X AI libraries, Nemotron models and support for runtimes including Ollama, vLLM and PyTorch with CUDA. That matters because the value of the DGX Spark is not just the processor; it is the combination of hardware and a preconfigured environment for building and running local AI applications.

Two 64GB systems can become a 128GB cluster

NVIDIA is using the same networking capability to make the smaller system scalable. Each DGX Spark includes a ConnectX-7 network interface, and two 64GB units can be connected directly over a 200-gigabit-per-second link. NVIDIA's Sync Cluster Assistant handles the network configuration and allows the two systems to operate as a larger pool with 128GB of combined memory.

There is a performance catch to that approach. NVIDIA reports up to 1.7 times the performance of a single system when two 64GB Sparks are used together in its Qwen 3.8 27B test. That is a vendor result from one workload, not an independent benchmark or a promise that every application will scale by the same amount. The advantage is therefore flexibility: a developer can start with one machine and add another when the workload no longer fits, rather than buying a larger system immediately.

The memory market is part of the story

The timing of the 64GB model makes more sense against NVIDIA's earlier explanation for its DGX Spark price increase. In February, NVIDIA said worldwide memory supply constraints had forced the Founders Edition price from $3,999 to $4,699 without changing the hardware. The new product gives NVIDIA and its manufacturing partners another configuration that uses less memory while preserving the same GB10 platform.

That does not mean the 64GB model will automatically be cheaper to manufacture by exactly half, because memory is only one part of the system and the final products come from different manufacturers. It does mean NVIDIA now has a lower-memory entry point for workloads that do not require 128GB, while the more expensive configurations remain available for larger models. The result is a product range shaped as much by memory capacity and supply as by raw processor performance.

Who the DGX Spark 64GB actually makes sense for

The new machine is aimed at a very specific PC buyer: someone who already has a reason to run sizeable AI models locally and values NVIDIA's integrated software environment. Developers working with local agents, inference, model experimentation or data that should remain on a private machine can use the 64GB configuration without paying for memory they may not need. For an ordinary desktop user, however, the $4,999 starting price and specialized software stack put it far outside the role of a conventional productivity PC.

The October 23 launch will also make the partner configurations more important than the headline price. Acer, ASUS, Dell, Gigabyte, HP and MSI are responsible for the systems, so storage capacity, warranty terms, regional pricing and actual availability may differ between machines. Until those systems reach retailers, $4,999 is best treated as NVIDIA's announced starting point rather than a universal street price.

October will show whether 64GB is enough

The interesting test begins after the machines ship. A 64GB local AI computer does not need to beat the 128GB model at everything; its usefulness depends on whether the models developers actually want to run fit within its smaller memory pool and whether the partner systems remain close to the announced price. NVIDIA is also preparing a Sync Model Launcher for later in October to simplify deployment across one or more Sparks, which could make the two-machine approach easier to use.

For now, the important change is straightforward: NVIDIA has turned DGX Spark from a single 128GB configuration into a platform with a 64GB entry point. The $4,999 price will attract attention, but the real question for buyers is not whether 64GB sounds large. It is whether the workloads they plan to run can stay inside it.

Copy this post

0 Comments

No comments yet. Be the first to share your thoughts.

Join the conversation

Log in or create a free account to leave a comment. You can edit or delete your own comments any time.