Five years ago, Jim Keller hinted at the concept of Rentable Units!
In 2019, during the Hot Chips conference, I had the privilege of attending an exclusive hot wings event where I met Jim Keller. I recorded his engaging conversation with Raj Kadori, which I highly recommend watching, especially for Keller referring to quantum computers as "pixie dust." I still cherish the badge I wore at the event as a reminder of that memorable interaction.
During our discussion, I gathered the courage to ask Keller about Intel's future CPU architectures, specifically about the possibility of combining threads to enhance IPC (instructions per cycle). Contrary to my expectations, he replied that the approach should be the opposite. I didn't grasp his meaning at the time, as I’m not a CPU architect, but the significance of his comments became clearer this month.
After releasing my videos on the Royal Core project, I spoke with a senior Intel official who wished to clarify the situation. This person confirmed several vital points: First, Beast Lake has indeed been canceled. Second, the decision to halt it was made earlier this year. Third, Keller's departure from Intel was due to internal political conflicts, which ultimately hurt the project. However, the individual emphasized that there were no intrinsic problems with the Royal Core; it was progressing smoothly and achieving its milestones until Pat Gelsinger decided to abruptly terminate it.
Gelsinger reportedly believed that high-performance cores were unnecessary, as he intended for CPUs to serve primarily as connections to GPUs. While this might reflect Nvidia's successful operational model, which currently doesn’t rely on high-performance CPU cores for competitiveness, I believe this perspective may be misguided for Intel. Gelsinger’s decision to terminate the Royal Core project, which was costly and complex, could hinder Intel's positioning as the landscape shifts toward AI.
Despite holding potential in AI and new technologies, Intel has missed crucial opportunities over the years, such as the recent GPU crash, while still investing significantly in underperforming projects like Arc. This misallocation of resources has led to a situation where Gelsinger’s decision to dissolve the Royal Core project may have dire long-term consequences for Intel's competitiveness in CPUs.
Recent reports indicate that many Intel CPU architects have left the company, particularly from the Royal Core design team in Oregon, who were frustrated by the lack of ambition toward high-performance cores and feared they would be relegated to minor projects. There were plans to eventually consolidate E-cores and P-cores into high-performance configurations, which Keller had been advocating for.
I have acquired extensive details about what Jim Keller had been working on that Gelsinger ultimately canceled. Information from various Intel sources corroborates the narrative that Beast Lake was poised to be a groundbreaking architecture. Initially, AOL Lake was to bridge to Royal Core but evolved as Intel faced challenges with hyperthreading.
The launch of Nova Lake in 2026 was anticipated to showcase Royal Core 1.0, featuring Rentable Units, emphasizing massive single-thread performance gains while forgoing hyperthreading. However, with the cancellation of Beast Lake, the fate of Nova Lake and associated designs remains uncertain, particularly regarding features like hyperthreading and Rentable Units.
Beast Lake was expected to reveal a design that utilized 12 performance cores (P-cores) and 16 efficiency cores (E-cores), potentially allowing cores to dynamically manage workloads. This design aimed for unprecedented performance, particularly in gaming, compelling several Intel employees to advocate for a focus on high-performing cores during Gelsinger’s tenure.
Looking forward, the outlook for Intel remains unclear. Rumors suggest Nova Lake might adopt different configurations but might not prioritize high efficiency or IPC anymore. The return of hyperthreading might indicate a shift away from developing advanced architectures like Royal Core, a concern for many looking at Intel's future viability in competitive CPU markets.
Despite cancellation setbacks, I remain cautiously optimistic. The groundwork for the core design is reportedly solid, with potential for future development of Rentable Units still existent. However, reinvestment in the design team and a commitment to advancing core performance will be crucial if Intel hopes to regain its footing under current leadership.