The Intel 386 processor featured 32-bit architecture and a complex register set using six different circuit types for implementation.
Registers in the 386 are built with static RAM cells (6T, 8T, 10T) optimized for speed and features like dual-porting and triple-porting.
The register file includes interleaved bits for lower 16 bits to support 8-bit register access, requiring a shuffle network for bit alignment.
The processor's datapath uses a regular structure with horizontal functional blocks and vertical bus lines, maximizing performance while minimizing die area.
Backward compatibility with earlier x86 processors added complexity, such as partial register writes and interleaving, contrasting with RISC design simplicity.
The article maps physical register types to architectural registers, speculating on assignments like type (d) for EAX/EBX/ECX/EDX and type (f) for segment registers.
The Intel 8087 floating-point coprocessor, introduced in 1980, performed floating-point operations up to 100 times faster than early microprocessors, benefiting applications like AutoCAD and spreadsheets.
The 8087 was a complex chip with 40,000 to 75,000 transistors, featuring a datapath, constant ROM, microcode ROM, and stack-based architecture with eight 80-bit registers and tag bits.
Its stack design used a top-of-stack pointer and offset-addressing, but suffered from stack overflow issues and was difficult for compilers, requiring extensive workarounds like saving the full stack to memory.
Despite drawbacks, the 8087 was a commercial success, establishing Intel Israel and influencing the IEEE 754 floating-point standard; its stack approach was later replaced by SSE and AVX in modern processors.
Gauntlet, an open-source multi-agent pipeline, evaluated deep technical comprehension of computer architecture papers, surpassing human analysis in most cases.
In evaluations on 20 recent ISCA/HPCA papers, Gauntlet was preferred over human analysis in 15 out of 20 comparisons, with significant advantage in Critical Rigor.
The multi-agent structure, particularly the synthesis pass, was key to Gauntlet's performance, outperforming a single-agent approach on 96% of papers.
Human analyses were preferred only in aspects like trust and usefulness, often due to issues like confident but wrong claims or unprioritized breadth.
All analyses, scores, and the evaluation rubric are released as a community resource for further research.