Benchmark for KV-cache tiering in LLM inference, measuring when evicted KV blocks should be kept in CPU RAM, moved to disk, or discarded.
Central monorepo for benchmarking, telemetry, monitoring, and systems-performance projects spanning GPU and CPU benchmarks, agent runtime analysis, Windows and AMD APU monitoring, and web-based diagnostics.
A systematic benchmark evaluating threading and multiprocessing strategies in the pydre analytics pipeline, with a focus on workload-dependent performance and practical default execution strategies.
Serving cost model for 2026 LLM inference stacks, integrating empirical data from 30+ portfolio benchmarks into a single auditable model.
Reproducible CPU inference benchmarking framework for deep learning workloads. Implements GDEV-AI performance evaluation across legacy and modern Intel Xeon architectures.
gem5-based simulation study of Matrix–Matrix Multiplication (MM) across CPU models, clock frequencies, and memory hierarchies (DDR3/DDR4/DDR5). Includes config scripts, automation, stats collection, plots, and final report.
Simulation benchmark measuring how LoRA adapter swap churn interacts with KV cache fragmentation in multi-adapter LLM serving systems.
Discrete-event simulation of KV-cache fragmentation dynamics in continuous batching LLM servers, comparing contiguous vs paged allocation and compaction policies over sustained load.
Simulation of speculative KV-cache prefetching for multi-turn LLM conversations, measuring stall reduction, hit rate, and wasted GPU prefetch bandwidth.
🚀 Benchmark parallel execution strategies in the pydre analytics pipeline to identify the best approach for diverse workload scenarios.
To associate your repository with the systems-performance topic, visit your repo's landing page and select "manage topics."