Intel Buried This AI CPU... We Dug It Up! - Centaur Hauls

The video opens with a mystery around a Centaur CPU that can fit in an Intel socket but is not an Intel or AMD product, setting the stage for a deep dive into a long-lost x86 design. The hosts establish Centaur Hauls as a historically significant prototype line born from Centaur Technology, a unit that is connected to IDT and Via through a winding patent and licensing history. Early on they painstakingly examine the board’s layout, showing a motherboard without a traditional chipset, a setup full of jumpers and diagnostic pads, and a USB-C port that serves multiple purposes. The presentation foregrounds Centaur’s past: a shared lineage with the Xian and Via Nano lines, and the famous licensing battles that shaped the x86 landscape in the 1990s and 2000s. The segment concludes that the CHA SoC inside this prototype was designed as a high-bandwidth, AI-accelerated platform, with an onboard neural processor called encor intended to leverage large L3 caches and memory for speed, but the execution faced feasibility and timing challenges that prevented commercial release. Transitioning to hands-on discovery, the hosts boot the system to reveal a BIOS built in 2021, well after Centaur’s heyday, and realize this board is a development platform rather than a finished product. They describe a flexible, test-focused environment with fuses and jumpers to alter CPU identity, clock speeds, and feature sets, including AVX 512 support and a bidirectional AI interface that promised unprecedented acceleration. They highlight the absence of a traditional chipset, noting that the entire system is effectively the CPU itself plus a handful of I/O like USB, PCIe, and M.2, emphasizing the experimental nature of the build. The video then maps out Centaur’s historical arc: from the original Centaur Hauls concept to Via’s crossover with Intel, and the eventual fade as the broader market shifted toward GPUs and dedicated AI accelerators. The mounting curiosity pivots toward answering why this project never finished, and the hosts begin a historical tour of Centaur’s lifecycle, patent disputes, and strategic shifts that culminated in Centaur’s decline and absorption of engineers by Intel. With the historical groundwork in place, the hosts perform live hardware testing to demonstrate what the CHA can do. They show BIOS-level experimentation, power-on behavior, and basic readouts from CPU-Z, which reveal the ENCOr AI accelerator, the eight-core configuration, and a speed ceiling around 2.5 GHz. They test basic Windows 10 operation, note the absence of a firmware TPM, and confirm the platform is still able to boot despite the age and development status. Benchmark attempts reveal a sobering reality: C bench and Cinebench interactions produce results that are well behind contemporary CPUs, with 1660 and even modern Ryzen-class performances appearing as fair but distant benchmarks. They also experiment with Linux, finding more instability and driver gaps, which further illustrate the practical challenges of reviving an abandoned, under-supported platform. The narrative then circles back to the strategic question: why did Centaur Hauls not reach completion? The hosts chart the competitive landscape of the late 2000s and early 2010s, where Intel, AMD, and VIA were locked in licensing and patent fights, and where embedded x86 strategies faced a rapidly changing market fueled by AI-focused accelerators and GPUs. They discuss Via’s pivot to embedded markets and China, Centaur’s acquisition by Via and later involvement with Intel, and the broader lesson that timing, licensing, and market readiness can doom even technically promising designs. The video ends on a reflective note about what Centaur’s design represented at the cusp of a new AI era, and why some of the most interesting hardware concepts fail to become mainstream despite strong technical merit. The final takeaway is that Centaur Hauls stands as a historical milestone illustrating the friction between silicon innovation and market structure, a reminder that not every promising prototype makes the leap to production.”

Topics · technology · history · hardware · embedded systems · ai