NVIDIA Expands RTX Spark Roadmap Through 2030

NVIDIA presented a roadmap for its RTX Spark PC platform at Computex 2026, showing multi-generation releases stretching to 2030, multiple outlets report. PC Gamer and Tom's Hardware report the slide places the initial Grace/Blackwell RTX Spark in 2026, a Vera Rubin Spark generation in 2027/2028, and a Rosa Feynman Spark generation in 2029/2030. Tom's Hardware reports NVIDIA committed that every future generation of its platform will include a Spark chip. PC Gamer reports the Blackwell RTX Spark package includes 6144 GPU cores, 20 Grace CPU cores, and up to 128 GB of unified LPDDR5X memory. Wccftech and Heise note the Vera Rubin roadmap entry explicitly references support for LPDDR6 memory.

Moving from LPDDR5X to LPDDR6 typically implies higher memory bandwidth and efficiency for mobile and compact systems. Industry hardware vendors adopting LPDDR6 aim for improved sustained throughput and lower energy per bit, which matters for unified memory architectures where CPU and GPU share the same pool.

The published Blackwell Spark hardware figures (as reported by PC Gamer) and Wccftech's mention of a 600 GB/s unified memory bandwidth for Spark-class devices point to system designs prioritizing on-package memory bandwidth rather than discrete DRAM channels. For machine learning workloads that rely on large shared tensors and frequent CPU-GPU transfers, those bandwidth characteristics materially affect real-world throughput and latency.

Public reporting frames NVIDIA's roadmap as an explicit push into Windows on Arm and a multi-year commitment to Spark platforms, with Tom's Hardware and Wccftech describing the move as part of a broader effort to bring agentic AI capabilities to laptops and desktops. For OEMs and software vendors, a multi-generation hardware roadmap reduces uncertainty around platform continuity and can influence decisions about driver investments and system designs.

The choice to name CPU architectures (Vera Rubin, Rosa Feynman) and to indicate a memory-generation change signals a hardware cadence that could affect timelines for OS tuning, compiler toolchains, and ISV optimization. Historically, shifts in memory standards and CPU microarchitectures require updates to power delivery, thermal design, and memory controllers, which ripple into system validation cycles.

monitor OEM design wins and announced notebooks/mini-PCs that adopt the Blackwell Spark parts; early silicon samples and thermal/benchmark data will reveal how the unified memory bandwidth performs in mixed CPU/GPU ML workloads. For the memory ecosystem: track LPDDR6 module availability, JEDEC finalization status, and vendor pricing, since those factors shape time to market. For software: watch for Microsoft and ISV support signals for Windows on Arm optimizations targeting the Spark unified memory model.