A new ×86 core architecture for the next generation of computing

A Metric-Guided Method for Discovering Impactful Features and Architectural Insights for Skylake-Based Processors

“…For example, profiling results shared by this article, indicate that integer benchmarks highly benefit from better instruction fetch, while Floating-Point (FP) benchmarks benefit from an optimized execution engine. Therefore, hardware vendors introduce several changes at once in new generations [5][6][7]. While this can benefit certain workloads (as Figure 2 demonstrates), it induces inter-feature interactions.…”

“…Figure 2 shows significant speedup for a particular workload on recent processors from Intel and AMD. While both Skylake [6] and Zen [5] microarchitectures show significant speedup over prior generations, the primary question this article explores is what are the architectural changes that enable this speedup? Processor architects lump multiple architectural modifications into a new processor generation.…”

Cache Exclusivity and Sharing

“…for all h, l ≥ 0. We call Equation (7) the VCR equation. For any specific cache, the VCR equation is a single equation, but for the split LRU stack, it is parameterized-the equality must hold for all nonnegative integers h, l. We prove the VFP theorem, which states that the VFP is not only correct (i.e., it satisfies the VCR equation) but also unique (i.e., no other solution exists that satisfies the VCR equation).…”

“…Theorem 3.1 (VFP Theorem). The VFP defined by Equation (4) is the only solution that satisfies the VCR stated in Equation (7).…”

“…AMD Bulldozer processors use cluster-based multithreading (CMT), where each core (or module) has two clusters. The new Zen processors use simultaneous multithreading (SMT) [7]. The two have the same effect of multilevel cache sharing.…”