11:53Machine-translated from Azerbaijani by AIRead in Azerbaijani
Nvidia Unveils DGX Spark Computer With 64 GB of Memory, Starting at 8500 Manats
Nvidia has unveiled a more affordable version of DGX Spark with 64 GB of memory for those who want to run artificial intelligence models on their own devices. The system is based on the GB10 chip and will go on sale on October 23. Models from Acer, Asus, Dell, Gigabyte, HP and MSI will start at 4999 dollars (approximately 8500 manats). The company is seeking to reduce the initial cost for users who do not need 128 GB of memory.
The development of the new version was influenced by the more efficient operation of artificial intelligence models and the sharp rise in RAM prices. For example, a powerful model such as Qwen 3.8 27B fits into 32 GB of memory with a limited context. Context refers to the amount of information that the model considers at the same time. Therefore, not everyone needs 128 GB of memory to get responses from a ready-made model. DGX Spark with 128 GB of memory and similar GB10 systems currently sell for 7000 - 9000 dollars (approximately 11900 - 15300 manats). These prices are also higher than Nvidia's updated recommended retail price. Versions with more memory will remain on sale for running large models, expanding clusters and further training models with specific data. Since the prices of memory and storage devices change frequently, the initial price of the 64 GB models may also change later.
Although the amount of memory has been reduced, the system's core hardware remains unchanged. The 64 GB models have a 20-core Arm processor. The CPU and GPU use the same memory, with a data transfer speed of 273 GB/s. The ConnectX 7 network adapter has also been retained. It can be used to connect several DGX Spark systems in a cluster. Such a setup can increase total memory and speed up the model's response generation.
Nvidia is also releasing a new software tool called Nvidia Sync Cluster Assistant to make cluster setup easier. The software automates the process of connecting the systems to one another. It complements the Nvidia Sync app. That app allows users to connect remotely to DGX Spark from a Mac or PC and use its computing power. It will also make it easier to manage several devices when additional memory and power are needed. A Model Launcher feature is also being added to the Nvidia Sync app. This feature automatically downloads and launches models such as Qwen 3.8 27B on one or more systems. The model is then connected to the browser-based OpenCode coding agent, allowing the user to begin local artificial intelligence tasks. Although the 64 GB version is cheaper, the system can be expanded later through cluster support.
The development of the new version was influenced by the more efficient operation of artificial intelligence models and the sharp rise in RAM prices. For example, a powerful model such as Qwen 3.8 27B fits into 32 GB of memory with a limited context. Context refers to the amount of information that the model considers at the same time. Therefore, not everyone needs 128 GB of memory to get responses from a ready-made model. DGX Spark with 128 GB of memory and similar GB10 systems currently sell for 7000 - 9000 dollars (approximately 11900 - 15300 manats). These prices are also higher than Nvidia's updated recommended retail price. Versions with more memory will remain on sale for running large models, expanding clusters and further training models with specific data. Since the prices of memory and storage devices change frequently, the initial price of the 64 GB models may also change later. Although the amount of memory has been reduced, the system's core hardware remains unchanged. The 64 GB models have a 20-core Arm processor. The CPU and GPU use the same memory, with a data transfer speed of 273 GB/s. The ConnectX 7 network adapter has also been retained. It can be used to connect several DGX Spark systems in a cluster. Such a setup can increase total memory and speed up the model's response generation. Nvidia is also releasing a new software tool called Nvidia Sync Cluster Assistant to make cluster setup easier. The software automates the process of connecting the systems to one another. It complements the Nvidia Sync app. That app allows users to connect remotely to DGX Spark from a Mac or PC and use its computing power. It will also make it easier to manage several devices when additional memory and power are needed. A Model Launcher feature is also being added to the Nvidia Sync app. This feature automatically downloads and launches models such as Qwen 3.8 27B on one or more systems. The model is then connected to the browser-based OpenCode coding agent, allowing the user to begin local artificial intelligence tasks. Although the 64 GB version is cheaper, the system can be expanded later through cluster support.