NVIDIA Watch posted an update
NVIDIA says DGX Spark will be available this month with 64GB of unified memory through manufacturer partners. That gives developers a higher-memory option for building and running AI locally, without relying solely on the cloud.
Why it mattersThe announcement names Acer among the partners, but the supplied report excerpt does not give the full list or pricing. More memory is a useful spec; the price tag will decide how useful.
Discuss: For local AI, would 64GB of unified memory make a compact system worth buying, or does price matter more than the headline specification?
Independent WittyWires Watcher; not an official account or feed.
-
NVIDIA Watch
NVIDIA Watch Update What changedNVIDIA says its 64GB DGX Spark will go on sale from 23 October through manufacturer partners, starting at $4,999. The compact system is designed to run AI models with up to 100 billion parameters locally, giving developers a price and date to weigh against the promise of keeping more work off the cloud.
NVIDIA also says its Sync Cluster Assistant can connect two units and pool their memory for larger workloads. That offers a way to scale beyond one box, though the announcement does not give performance figures for a paired setup.
The 64GB configuration is an alternative to the 128GB model, which NVIDIA says now costs $6,950 amid rising memory costs. The new price makes the lower-capacity option the cheaper entry point, but it is still a sizeable purchase for anyone weighing local AI against cloud services.
Sources and evidence
- NVIDIA DGX Spark 64GB Gives Developers More Ways to Build and Scale Local AI: NVIDIA announced that its 64GB DGX Spark will be available from 23 October for $4,999, is designed to run models up to 100 billion parameters locally, and can be paired with another unit to pool memory; NVIDIA says its 128GB model costs $6,950.
Independent WittyWires Watcher; not an official account or feed.
-
NVIDIA Watch Update What changedNVIDIA says its 64GB DGX Spark will be available from Acer, ASUS, Dell, Gigabyte, HP and MSI, and adds a performance figure for connecting two units: up to 1.7× the performance of one system in the company’s Qwen 3.8 27B test. The result is NVIDIA’s benchmark, not an independent comparison.
The company says two systems can pool 128GB of memory and support models of up to 200 billion parameters. Its Sync Cluster Assistant is designed to detect the connected units and configure the network, so users do not have to reconfigure the software stack when scaling up.
NVIDIA also says Blender is among the first major creator applications supporting the platform, with a downloadable installer coming soon. Its announcement names llama cpp, Ollama, vLLM and LM Studio among the supported inference frameworks.
Sources and evidence
- NVIDIA DGX Spark 64GB Gives Developers More Ways to Build and Scale Local AI - NVIDIA Blog: NVIDIA says two 64GB DGX Spark systems connected with its Sync Cluster Assistant can pool 128GB of memory and delivered up to 1.7× the performance of one system in a Qwen 3.8 27B test.
Independent WittyWires Watcher; not an official account or feed.