Scalable Parallel Computing Lab, SPCL @ ETH Zurich Courses

Performance-Detective: Automatic Deduction of Cheap and Accurate Performance Models

Automatische Ableitung günstiger und präziser Leistungsmodelle für komplexe Anwendungen durch intelligente Codeanalyse und optimiertes Experimentdesign.

• YouTube
• 16 minutes
• On-Demand
• Free Video

ML for High-Performance Climate: Data Post Processing, Compression, and Earth Virtualization Engines

Explore advanced machine learning techniques for climate data processing, compression, and Earth virtualization, enhancing high-performance computing in climate science and research.

• YouTube
• 1 hour 8 minutes
• On-Demand
• Free Video

Noise in the Clouds - Influence of Network Performance Variability on Application Scalability

Explora el impacto del ruido de red en la escalabilidad de aplicaciones HPC en la nube, analizando el rendimiento, la escalabilidad y los costos en comparación con sistemas HPC locales.

• YouTube
• 22 minutes
• On-Demand
• Free Video

Streaming Task Graph Scheduling for Dataflow Architectures

Explora la programación eficiente de grafos de tareas en arquitecturas de flujo de datos, presentando un nuevo modelo y técnicas de análisis para mejorar el rendimiento y la utilización de dispositivos.

• YouTube
• 29 minutes
• On-Demand
• Free Video

Earth Virtualization Engines (EVE) for Climate Change Modeling

Explore Earth Virtualization Engines (EVE) for climate research, discussing its potential impact and significance in addressing global environmental challenges.

• YouTube
• 23 minutes
• On-Demand
• Free Video

From Two Strong Oxen to Billions of Fleas - The Evolution of Massively Parallel Distributed Supercomputers

Explore massively parallel distributed supercomputers, focusing on distributed-memory parallelism and Remote Direct Memory Access. Gain insights into current designs and future challenges in high-performance computing.

• YouTube
• 33 minutes
• On-Demand
• Free Video

HPVM: Performance, Programmability and Retargetability for Heterogeneous Parallel Systems

Explore HPVM project's approach to programming heterogeneous parallel systems, focusing on hardware-agnostic programming and automatic generation of retargetable, high-performance code generators for diverse architectures.

• YouTube
• 1 hour 5 minutes
• On-Demand
• Free Video

The Role of MPI in Modern AI - Rusty Lusk's Legacy

Explore MPI's impact on modern AI through Torsten Hoefler's tribute to Rusty Lusk, examining the evolution and future of parallel computing in artificial intelligence.

• YouTube
• 40 minutes
• On-Demand
• Free Video

Realizing Petabit/s IO and sub-pJ/bit System-wide Communication with Silicon Photonics

Explore integrated silicon photonics for high-bandwidth, low-energy chip communication, enabling Pb/s speeds and flexible architectures for distributed ML applications.

• YouTube
• 1 hour 1 minute
• On-Demand
• Free Video

A Chiplet-Based Generative Inference Architecture with Block Floating Point Datatypes

Explore a chiplet-based architecture for generative inference, featuring block floating point datatypes, quantization techniques, and optimizations for large language models.

• YouTube
• 1 hour 6 minutes
• On-Demand
• Free Video

Deinsum: Practically I/O Optimal Multi-Linear Algebra for Distributed Systems

Optimiza el rendimiento de álgebra multilineal distribuida mediante derivación automática de mosaicos y programaciones óptimas, mejorando la intensidad aritmética y minimizando el movimiento de datos.

• YouTube
• 33 minutes
• On-Demand
• Free Video

HPC and AI/ML: A Synergistic Relationship

Explore synergies between HPC and AI/ML: scalable algorithms for extreme-scale deep learning, optimizations using neural network properties, and AI/ML applications in HPC performance modeling and code generation.

• YouTube
• 59 minutes
• On-Demand
• Free Video

Follow the Data: Memory-Centric Designs for Modern Datacenters

Explore data-driven approaches for memory architecture in exascale supercomputing and cloud memory disaggregation, addressing critical system performance and cost challenges.

• YouTube
• 1 hour 2 minutes
• On-Demand
• Free Video

Accelerating Data Serialization and Deserialization Protocols with In-Network Compute

Explore in-network compute for accelerating data serialization/deserialization protocols, improving efficiency in datacenter and HPC systems through SmartNIC offloading and on-the-fly deserialization techniques.

• YouTube
• 24 minutes
• On-Demand
• Free Video

Democratizing Deep Learning with DeepHyper

Explore DeepHyper, a software package automating deep neural network design for scientific data using scalable architecture and hyperparameter search, with insights on uncertainty estimation.

• YouTube
• 1 hour 1 minute
• On-Demand
• Free Video