Didem Unat

Didem Unat, Ilyas Turimbetov, Mohammed Kefah Taha Issa et al.

In recent years, GPUs have become the preferred accelerators for HPC and ML applications due to their parallelism and fast memory bandwidth. While GPUs boost computation, inter-GPU communication can create scalability bottlenecks, especially as the number of GPUs per node and cluster grows. Traditionally, the CPU managed multi-GPU communication, but advancements in GPU-centric communication now challenge this CPU dominance by reducing its involvement, granting GPUs more autonomy in communication tasks, and addressing mismatches in multi-GPU communication and computation. This paper provides a landscape of GPU-centric communication, focusing on vendor mechanisms and user-level library supports. It aims to clarify the complexities and diverse options in this field, define the terminology, and categorize existing approaches within and across nodes. The paper discusses vendor-provided mechanisms for communication and memory management in multi-GPU execution and reviews major communication libraries, their benefits, challenges, and performance insights. Then, it explores key research paradigms, future outlooks, and open research questions. By extensively describing GPU-centric communication techniques across the software and hardware stacks, we provide researchers, programmers, engineers, and library designers insights on how to exploit multi-GPU systems at their best.

Tugba Torun, Ameer Taweel, Didem Unat

Sparse tensor operations are increasingly important in diverse applications such as social networks, deep learning, diagnosis, crime, and review analysis. However, a major obstacle in sparse tensor research is the lack of large-scale sparse tensor datasets. Another challenge lies in analyzing sparse tensor features, which are essential not only for understanding the nonzero pattern but also for selecting the most suitable storage format, decomposition algorithm, and reordering methods. However, due to the large size of real-world tensors, even extracting these features can be computationally expensive without careful optimization. To address these limitations, we have developed a smart sparse tensor generator that replicates key characteristics of real sparse tensors. Additionally, we propose efficient methods for extracting a comprehensive set of sparse tensor features. The effectiveness of our generator is validated through the quality of extracted features and the performance of decomposition on the generated tensors. Both the sparse tensor feature extractor and the tensor generator are open source with all the artifacts available at https://github.com/sparcityeu/FeaTensor and https://github.com/sparcityeu/GenTensor, respectively.

Mohamed Wahib, Muhammed Abdullah Soyturk, Didem Unat

To reduce the computational and memory overhead of Large Language Models, various approaches have been proposed. These include a) Mixture of Experts (MoEs), where token routing affects compute balance; b) gradual pruning of model parameters; c) dynamically freezing layers; d) dynamic sparse attention mechanisms; e) early exit of tokens as they pass through model layers; and f) Mixture of Depths (MoDs), where tokens bypass certain blocks. While these approaches are effective in reducing overall computation, they often introduce significant workload imbalance across workers. In many cases, this imbalance is severe enough to render the techniques impractical for large-scale distributed training, limiting their applicability to toy models due to poor efficiency. We propose an autonomous dynamic load balancing solution, DynMo, which provably achieves maximum reduction in workload imbalance and adaptively equalizes compute loads across workers in pipeline-parallel training. In addition, DynMo dynamically consolidates computation onto fewer workers without sacrificing training throughput, allowing idle workers to be released back to the job manager. DynMo supports both single-node multi-GPU systems and multi-node GPU clusters, and can be used in practical deployment. Compared to static distributed training solutions such as Megatron-LM and DeepSpeed, DynMo accelerates the end-to-end training of dynamic GPT models by up to 1.23x for MoEs, 3.18x for parameter pruning, 2.23x for layer freezing, 4.02x for sparse attention, 4.52x for early exit, and 1.17x for MoDs.