New leaks have emerged regarding Nvidia's next-generation GPU plans. According to the shared information, the name Blackwell-Next, which the company has not yet officially announced, was found in a new patch submitted to the Linux kernel. This indicates that Nvidia engineers have already begun software preparations for hardware beyond Blackwell.
Blackwell-Next Name Appears for the First Time
Blackwell-Next is mentioned in a Linux kernel patch for the NVGrace-GPU VFIO driver. The patch states that while the BAR0 polling method used for existing Grace Blackwell hardware will be maintained, a new CXL DVSEC-based readiness check has been added for Blackwell-Next GPUs. In short, Nvidia is updating the infrastructure for how GPU memory will be controlled by the system in the next-generation platform.
These changes are part of the work being done on the VFIO subsystem within the Linux 7.2 cycle. The primary purpose of the patch is to update the method for checking if GPU memory is ready after initialization or reset. However, the most striking aspect is that the name "Blackwell-Next" has now become directly visible on the software side.
The shared image also includes Nvidia's data center roadmap. According to this, the company is expected to move to the Rubin architecture after Blackwell, and then to the Feynman architecture further down the line. It is also rumored that these names used for data center GPUs might be carried over to consumer graphics cards. Nvidia's earlier mention of new N-series solutions based on Blackwell also supports this possibility.
For now, the technical details of Blackwell-Next are unknown. However, the preparation of Linux support indicates that the company is not only preparing for distant architectures like Rubin and Feynman but also for an intermediate generation that could be positioned between Blackwell and Rubin. Therefore, the name Blackwell-Next may become more frequently discussed in the data center GPU sector in the coming period.
0 Comments: