NVIDIA’s mysterious “Blackwell-Next” GPU architecture designed to support worst-case timeout scenarios.
NVIDIA has begun laying the foundational software work for its upcoming computing hardware well before its official release. Tech site Phoronix spotted a recent Linux kernel patch submission for the Linux 7.2 cycle containing a reference to “Blackwell-Next”.
The update introduces “CXL DVSEC-based” readiness polling for Blackwell-Next hardware inside the existing nvgrace-gpu vfio-pci variant driver. Authored by Ankit Agrawal, the patch utilizes interruptible, lockless waits designed to natively support spec-defined worst-case timeout scenarios.
Because NVIDIA’s official roadmaps already position Blackwell Ultra and the Rubin architecture as successive iterations to the original Blackwell chip, the “Blackwell-Next” nomenclature remains slightly ambiguous. However, analyzing the code gives a clear hint: the change modifies the nvgrace-gpu driver, which handles systems pairing a Grace CPU with Blackwell hardware. While Rubin is slated to pair with a new “Vera” CPU, engineers speculate that this patch represents early, foundational integration for the upcoming Rubin generation.
Crucially, this patch is strictly an enterprise affair. It targets data center platforms, AI infrastructure, and virtualization tasks, which means it does not confirm specifications for consumer-oriented GeForce graphics cards. While long-term roadmaps show that architectures like Rubin and the Feynman generation will eventually span both AI and consumer lines, NVIDIA has yet to announce details on next-gen consumer chips like the rumored GeForce RTX 50 “SUPER”. For now, NVIDIA’s attention remains on readying its multi-generational data center portfolio.
