10 repository-uri
Utilities for tracking GPU memory and throughput during model inference.
Explore 10 awesome GitHub repositories matching testing & quality assurance · GPU Performance Profilers. Refine with filters or upvote what's useful.
Faceswap is a comprehensive framework for automated media manipulation and neural face synthesis. It provides a modular pipeline that manages the entire lifecycle of facial feature extraction, deep learning model training, and image conversion. By coordinating complex computer vision workflows, the system enables users to map facial identities between source and destination datasets while maintaining structural alignment and lighting consistency across video frames. The project distinguishes itself through a highly extensible plugin-based architecture that handles hardware-accelerated process
Benchmarks graphics hardware by tracking memory usage and throughput across varying batch sizes to refine pipeline performance.
nvtop este un dashboard bazat pe terminal utilizat pentru monitorizarea performanței, utilizării memoriei și temperaturii mai multor procesoare grafice și acceleratoare hardware. Acesta funcționează ca un instrument de administrare centralizat pentru a urmări starea și sarcina de calcul a mai multor dispozitive pe un singur sistem. Instrumentul se distinge prin corelarea ID-urilor proceselor de sistem cu consumul de resurse hardware, permițând utilizatorilor să identifice aplicații specifice care consumă resurse GPU. Acesta utilizează un strat de abstractizare agnostic față de furnizor pentru a suporta hardware de la mai mulți producători diferiți în cadrul unei singure interfețe. Software-ul oferă metrici de performanță în timp real și atribuirea resurselor per proces folosind o interfață bazată pe text. Utilizatorii pot gestiona layout-urile interfeței și pot salva preferințele de afișare printr-un fișier de configurare local pentru a menține setările între sesiuni.
Analyzes individual process resource consumption on GPUs to identify bottlenecks and memory leaks.
jetson-inference is a set of libraries and tools for executing optimized deep learning models on embedded GPU hardware. Its primary purpose is to enable real-time computer vision and AI inference at the edge with low latency and high throughput. The project distinguishes itself through high-performance streaming analytics and the ability to execute concurrent AI pipelines on auto-grade silicon. It provides specialized support for multi-sensor stream processing, utilizing zero-copy data transport to load camera frames directly into GPU memory. The codebase covers a broad surface of capabiliti
Analyzes and debugs GPU-accelerated workloads to optimize AI, graphics, and compute performance.
This project is a comprehensive collection of educational examples and reference implementations for building vision and language models using PyTorch. It serves as a deep learning tutorial covering the end-to-end process of developing neural networks, from initial architecture definition to final production deployment. The repository provides detailed guides on implementing a wide range of domain-specific models, including convolutional neural networks for object detection and segmentation, as well as transformer and recurrent architectures for natural language processing. It emphasizes gene
Tracks CPU and GPU utilization and data throughput to identify system-level application bottlenecks.
Lists running GPU processes with PID, user, and memory, and allows termination from the interface.
Measures GPU throughput, utilization, cache hit rates, and memory throughput to identify optimization opportunities.
capa is a binary capability scanner that identifies high-level behaviors and actions an executable can perform, such as network communication or file manipulation. It functions as a malware behavior analysis tool and a MITRE ATT&CK mapping framework, scanning PE, ELF, .NET, and shellcode files through both static analysis and dynamic sandbox report processing. The tool distinguishes itself through a YAML-based detection rule engine that defines detection logic in human-readable files, with conditions expressed as feature combinations and logical operators. It integrates with IDA Pro, Ghidra,
Limits analysis to specific processes by PID when processing dynamic sandbox reports.
btrace este un instrument de tracing dinamic și profiler de performanță pentru JVM, utilizat pentru injectarea scripturilor de instrumentare sigure într-o mașină virtuală Java care rulează, fără a necesita repornirea procesului. Acesta funcționează ca un framework de agent Java și un server Model Context Protocol, expunând operațiunile de diagnosticare JVM și instrumentele de tracing către modelele de limbaj mari și asistenții AI. Proiectul se distinge prin activarea injectării de cod în timp real și a instrumentării la nivel de bytecode printr-un protocol binar securizat. Asigură stabilitatea în producție printr-un motor de analiză statică de siguranță care blochează modelele de cod instabile, cum ar fi buclele și alocările, în timpul fazei de compilare a scriptului. Sistemul acoperă o gamă largă de capabilități de observabilitate, inclusiv tracing-ul execuției metodelor, urmărirea alocării obiectelor și monitorizarea accesului la câmpuri. Oferă profilarea performanței prin metrici de latență și eșantionarea datelor de execuție și poate emite evenimente personalizate Java Flight Recorder pentru analiză nativă. Agentul suportă strategii de implementare flexibile, inclusiv bootstrapping la pornire sau atașarea la procese live în medii containerizate folosind pachete fat JAR.
Includes utilities for tracking GPU memory and throughput during deep learning model inference.
Perfetto is a platform for system-level performance tracing and analysis on Linux and Android. It combines a high-throughput trace recorder, a SQL-based query engine, and a browser-based visualizer into a single toolchain. The platform covers CPU scheduling and call-stack profiling, native and Java heap memory allocation tracking, GPU and graphics events, and system-wide counters such as CPU frequency and power consumption. The architecture decouples trace recording from offline analysis, using a compact protobuf format for event encoding and columnar storage for efficient SQL queries. The we
Organizes GPU traces into timelines by device and process, displaying per-kernel metric tables and details.
regl is a declarative WebGL library that manages graphics state and GPU resources through functional commands instead of manual binding and state tracking. It provides a command-based drawing abstraction where shaders, attributes, and render state are encapsulated into reusable, compiled functions that can be executed efficiently. What sets regl apart is its scoped state inheritance system, which allows nested drawing commands to inherit and override render state from parent scopes for organized rendering. The library automatically recovers from GPU context loss by restoring buffer and textur
Regl collects CPU and GPU timing data and draw call counts per command for rendering performance diagnostics.