Parallel Computing seminars

Statistical workloads such as Markov chain Monte Carlo, bootstrapping, cross-validation and simulation can benefit from parallel execution, but practical obstacles include blocked R sessions, duplicated data, worker setup and rewritten code. This seminar introduces open-source tools built on asynchronous input/output and messaging, combining asynchronous parallel execution with shared R objects and zero-copy data access. A practical tour covers mirai for scaling work from laptops to high-performance computing, servers and cloud systems, and mori for shared memory. Applications include responsive Shiny interfaces and event-driven programming. The presentation also discusses migrating existing parallel and future code and the prospects for a zero-copy parallel computing model.

Katarzyna Swirydowicz uses an asynchronous solver on an AMD system to examine the practical implementation of computational linear algebra across CPUs and GPUs. The talk introduces the relevant computational ideas, programming models, and software tools, then considers how algorithmic structure interacts with hardware capabilities. This case study illustrates both the opportunities and implementation challenges of asynchronous methods for large-scale scientific computing on GPU-accelerated systems.

Matthias Bolten discusses the scalability of multigrid solvers for linear systems arising from discretized partial differential equations. On modern supercomputers, heterogeneous CPUs and GPUs and the widening gap between computation, network, and memory speeds complicate parallelization. Classical multigrid analysis relies on tightly coupled multiplicative components, whereas additive and asynchronous variants relax this coupling. The talk compares approaches to improving high-performance multigrid scalability, including asynchronous execution.

Vasileios Kalantzis develops iterative linear algebra algorithms that tolerate incomplete matrix-vector products in controller-worker cloud systems. Richardson and Chebyshev schemes solve linear systems using randomly available product entries, replacing missing entries with zero. For dominant eigenvectors, modified power iterations substitute zeros, previous entries, or averages of partial iterates for delayed components. The talk presents convergence results in expectation and numerical experiments on sparse matrices for both problem classes.

Evan Coleman studies how asynchronous solvers can recover convergence quality while tolerating stale data, stragglers, and variable delays. At the coordinator, Anderson acceleration connects asynchronous stationary iterations to Krylov methods with changing preconditioners and flexible GMRES. Controlled-delay experiments on high-performance computing infrastructure show that its effectiveness depends on the iteration's coupling density. At the worker, residual-weighted randomized coordinate descent includes Boltzmann weights that interpolate between uniform and greedy selection while preserving convergence guarantees. Both approaches seek better use of computation when information is inconsistent.

Daniel Szyld extends randomized point and block Gauss-Seidel and Gauss-Southwell convergence results from Hermitian positive-definite matrices to certain non-Hermitian classes, including overlapping variables in domain decomposition. The analysis treats a range of sampling probabilities and greedy selection strategies and identifies choices that optimize the bounds. The best expected convergence bounds for randomized methods match those of more expensive deterministic Gauss-Southwell algorithms. These results establish a convergence rate for asynchronous iterations. Joint work with Andreas Frommer.

Erik Boman discusses preconditioning for asynchronous linear solvers. Inner products create synchronization requirements in Krylov methods, while preconditioners can also improve iterations such as Richardson's method. The talk focuses on asynchronous incomplete factorizations and introduces ATS-ILU, an iterative incomplete LU method with synchronous and asynchronous versions that performs competitively with ParILU.

Edmond Chow examines how asynchronous updates improve parallel iterative computation. The first part covers asynchronous versions of classical first- and second-order linear iterations, Chebyshev methods, and multigrid, with attention to efficiency and fault tolerance. The second introduces reinforcement learning and asynchronous state-value estimation for finding optimal policies. When the state space is too large to enumerate, these updates focus computational effort on frequently visited regions.

Robert Tarjan asks how much data-structure operations can be accelerated using multiple unsynchronised processes. The presentation examines the challenges of exploiting concurrency through joint work with Siddhartha Jayanti on efficient concurrent disjoint-set union. It includes recent approaches that remove the need for randomisation.

Graphics processing units (GPUs) are widely available and have been used with great success to accelerate scientific computing in the last decade. These advances, however, are often not available to researchers interested in simulating spiking neural networks, but lacking the technical knowledge to write the necessary low-level code. Writing low-level code is not necessary when using the popular Brian simulator, which provides a framework to generate efficient CPU code from high-level model definitions in Python. Here, we present Brian2CUDA, an open-source software that extends the Brian simulator with a GPU backend. Our implementation generates efficient code for the numerical integration of neuronal states and for the propagation of synaptic events on GPUs, making use of their massively parallel arithmetic capabilities. We benchmark the performance improvements of our software for several model types and find that it can accelerate simulations by up to three orders of magnitude compared to Brian’s CPU backend. Currently, Brian2CUDA is the only package that supports Brian’s full feature set on GPUs, including arbitrary neuron and synapse models, plasticity rules, and heterogeneous delays. When comparing its performance with Brian2GeNN, another GPU-based backend for the Brian simulator with fewer features, we find that Brian2CUDA gives comparable speedups, while being typically slower for small and faster for large networks. By combining the flexibility of the Brian simulator with the simulation speed of GPUs, Brian2CUDA enables researchers to efficiently simulate spiking neural networks with minimal effort and thereby makes the advancements of GPU computing available to a larger audience of neuroscientists.

Large-scale numerical simulations of brain circuit models are important for identifying hypotheses on brain functions and testing their consistency and plausibility. Similarly, spiking neural networks are also gaining traction in machine learning with the promise that neuromorphic hardware will eventually make them much more energy efficient than classical ANNs. In this session, we will present the GeNN (GPU-enhanced Neuronal Networks) framework, which aims to facilitate the use of graphics accelerators for computational models of large-scale spiking neuronal networks to address the challenge of efficient simulations. GeNN is an open source library that generates code to accelerate the execution of network simulations on NVIDIA GPUs through a flexible and extensible interface, which does not require in-depth technical knowledge from the users. GeNN was originally developed as a pure C++ and CUDA library but, subsequently, we have added a Python interface and OpenCL backend. We will briefly cover the history and basic philosophy of GeNN and show some simple examples of how it is used and how it interacts with other Open Source frameworks such as Brian2GeNN and PyNN.

Last year’s SNUFA workshop report concluded “Moving toward neuron numbers comparable with biology and applying these networks to real-world data-sets will require the development of novel algorithms, software libraries, and dedicated hardware accelerators that perform well with the specifics of spiking neural networks” [1]. Taking inspiration from machine learning libraries — where techniques such as parallel batch training minimise latency and maximise GPU occupancy — as well as our previous research on efficiently simulating SNNs on GPUs for computational neuroscience [2,3], we are extending our GeNN SNN simulator to pursue this vision. To explore GeNN’s potential, we use the eProp learning rule [4] — which approximates RTRL — to train SNN classifiers on the Spiking Heidelberg Digits and the Spiking Sequential MNIST datasets. We find that the performance of these classifiers is comparable to those trained using BPTT [5] and verify that the theoretical advantages of neuron models with adaptation dynamics [5] translate to improved classification performance. We then measured execution times and found that training an SNN classifier using GeNN and eProp becomes faster than SpyTorch and BPTT after less than 685 timesteps and much larger models can be trained on the same GPU when using GeNN. Furthermore, we demonstrate that our implementation of parallel batch training improves training performance by over 4⨉ and enables near-perfect scaling across multiple GPUs. Finally, we show that performing inference using a recurrent SNN using GeNN uses less energy and has lower latency than a comparable LSTM simulated with TensorFlow [6].

Kilosort is a spike sorting pipeline for large-scale electrophysiology. Advances in silicon probe technology mean that in vivo electrophysiological recordings from hundreds of channels will soon become commonplace. To interpret these recordings we need fast, scalable and accurate methods for spike sorting, whose output requires minimal time for manual curation. Kilosort is a spike sorting framework that meets these criteria, and show that it allows rapid and accurate sorting of large-scale in vivo data. Kilosort models the recorded voltage as a sum of template waveforms triggered on the spike times, allowing overlapping spikes to be identified and resolved. Rapid processing is achieved thanks to a novel low-dimensional approximation for the spatiotemporal distribution of each template, and to batch-based optimization on GPUs. Kilosort is an important step towards fully automated spike sorting of multichannel electrode recordings, and is freely available.