NVIDIA Unveils RTX Spark Roadmap Through 2030: Rubin and Feynman Generations Expected

At the margins of the presentation of RTX Spark, NVIDIA showcased a roadmap that confirms the arrival of new generations of AI platforms for PCs and workstations based on the future architectures Rubin and Feynman.

For the period between 2027 and 2028, the debut of Vera Rubin Spark platforms is expected, which will adopt LPDDR6 memory. Subsequently, between 2029 and 2030, the Rosa Feynman Spark solutions will arrive, mentioned publicly for the first time by NVIDIA and set to utilize a future generation of memory that has not yet been announced.

Alongside the evolution of the Spark platforms, NVIDIA also plans to upgrade the DGX Station for Windows. After the debut of the version based on the GB300 Grace Blackwell Ultra Desktop Superchip, the roadmap indicates the arrival of systems built on the future Rubin and Feynman architectures. The subsequent generations will integrate components such as Vera and Rosa CPUs, HBM4 and HBM Next memory, and new generations of ultra-high-speed networking, namely ConnectX-9 at 1600 Gbps and ConnectX-10.

DGX Station for Windows targets a different audience compared to the RTX Spark systems. These are not PCs intended for gaming, but workstations designed for activities such as AI model development and training, data science, high-intensity inference, and deployment of local AI agents.

The initial configuration based on the GB300 combines a Blackwell Ultra GPU with a Grace CPU with 72 cores via NVLink-C2C interconnect, offering up to 748 GB of coherent memory and performance that can reach 20 petaFLOPS in FP4. According to NVIDIA, these systems will be able to locally run models with sizes up to one trillion parameters. The first DGX Stations for Windows are expected in Q4 2026 and will be produced by partners such as ASUS, Dell Technologies, GIGABYTE, HP, MSI, and Supermicro.

However, perhaps the most significant aspect of the announcement concerns the message sent to the ecosystem of hardware manufacturers and software developers. NVIDIA is looking to build a comprehensive platform for the era of agent-based AI on Windows, a goal that requires investments not only in hardware but also across the entire software stack and collaborations with Microsoft and application developers.

The publication of an extended roadmap through 2030 thus represents a guarantee of continuity for partners and customers and demonstrates that RTX Spark is not just a simple experimental project.