More Like this

1. Multi-dimensional balanced graph partitioning via projected gradient descent

   https://doi.org/10.14778/3324301.3324307  

   Avdiukhin, Dmitrii; Pupyrev, Sergey; Yaroslavtsev, Grigory (April 2019, Proceedings of the VLDB Endowment)

   Motivated by performance optimization of large-scale graph processing systems that distribute the graph across multiple machines, we consider the balanced graph partitioning problem. Compared to most of the previous work, we study the multi-dimensional variant in which balance according to multiple weight functions is required. As we demonstrate by experimental evaluation, such multi-dimensional balance is essential for achieving performance improvements for typical distributed graph processing workloads. We propose a new scalable technique for the multidimensional balanced graph partitioning problem. It is based on applying randomized projected gradient descent to a non-convex continuous relaxation of the objective. We show how to implement the new algorithm efficiently in both theory and practice utilizing various approaches for the projection step. Experiments with large-scale graphs containing up to hundreds of billions of edges indicate that our algorithm has superior performance compared to the state of the art. 

   more » « less
2. Decentralized Offload-based Execution on Memory-centric Compute Cores

   https://doi.org/10.1145/3422575.3422778  

   Baskaran, Saambhavi; Sampson, Jack (September 2020, International Symposium on Memory Systems (MEMSYS))

   With the end of Dennard scaling, power constraints have led to increasing compute specialization in the form of differently specialized accelerators integrated at various levels of the general-purpose system hierarchy. The result is that the most common general-purpose computing platform is now a heterogeneous mix of architectures even within a single die. Consequently, mapping application code regions into available execution engines has become a challenge due to different interfaces and increased software complexity. At the same time, the energy costs of data movement have become increasingly dominant relative to computation energy. This has inspired a move towards data-centric systems, where computation is brought to data, in contrast to traditional processing-centric models. However, enabling compute nearer memory entails its own challenges, including the interactions between distance-specialization and compute-specialization. The granularity of any offload to near(er) memory logic would impact the potential data transmission reduction, as smaller offloads will not be able to amortize the transmission costs of invocation and data return, while very large offloads can only be mapped onto logic that can support all of the necessary operations within kernel-scale codes, which exacerbates both area and power constraints. For better energy efficiency, each set of related operations should be mapped onto the execution engine that, among those capable of running the set of operations, best balances the data movement and the degree of compute specialization of that engine for this code. Further, this offload should proceed in a decentralized way that keeps both the data and control movement low for all transitions among engines and transmissions of operands and results. To enable such a decentralized offload model, we propose an architecture interface that enables a common offload model for accelerators across the memory hierarchy and a tool chain to automatically identify (in a distance-aware fashion) and map profitable code regions on specialized execution engines. We evaluate the proposed architecture for a wide range of workloads and show energy reduction compared to an energy-efficient in-order core. We also demonstrate better area efficiency compared to kernel-scale offloads. 

   more » « less
3. QUIDAM: A Framework for Qu ant i zation-Aware D NN A ccelerator and M odel Co-Exploration

   https://doi.org/10.1145/3555807  

   Inci, Ahmet; Virupaksha, Siri Garudanagiri; Jain, Aman; Chin, Ting-Wu; Thallam, Venkata Vivek; Ding, Ruizhou; Marculescu, Diana (September 2022, ACM Transactions on Embedded Computing Systems)

   As the machine learning and systems communities strive to achieve higher energy-efficiency through custom deep neural network (DNN) accelerators, varied precision or quantization levels, and model compression techniques, there is a need for design space exploration frameworks that incorporate quantization-aware processing elements into the accelerator design space while having accurate and fast power, performance, and area models. In this work, we present QUIDAM , a highly parameterized quantization-aware DNN accelerator and model co-exploration framework. Our framework can facilitate future research on design space exploration of DNN accelerators for various design choices such as bit precision, processing element type, scratchpad sizes of processing elements, global buffer size, number of total processing elements, and DNN configurations. Our results show that different bit precisions and processing element types lead to significant differences in terms of performance per area and energy. Specifically, our framework identifies a wide range of design points where performance per area and energy varies more than 5 × and 35 ×, respectively. With the proposed framework, we show that lightweight processing elements achieve on par accuracy results and up to 5.7 × more performance per area and energy improvement when compared to the best INT16 based implementation. Finally, due to the efficiency of the pre-characterized power, performance, and area models, QUIDAM can speed up the design exploration process by 3-4 orders of magnitude as it removes the need for expensive synthesis and characterization of each design. 

   more » « less
4. ViTCoD: Vision Transformer Acceleration via Dedicated Algorithm and Accelerator Co-Design

   https://doi.org/10.1109/HPCA56546.2023.10071027  

   You, Haoran; Sun, Zhanyi; Shi, Huihong; Yu, Zhongzhi; Zhao, Yang; Zhang, Yongan; Li, Chaojian; Li, Baopu; Lin, Yingyan (February 2023, 2023 IEEE International Symposium on High-Performance Computer Architecture (HPCA))

   Vision Transformers (ViTs) have achieved state-of-the-art performance on various vision tasks. However, ViTs’ self-attention module is still arguably a major bottleneck, limiting their achievable hardware efficiency and more extensive applications to resource constrained platforms. Meanwhile, existing accelerators dedicated to NLP Transformers are not optimal for ViTs. This is because there is a large difference between ViTs and Transformers for natural language processing (NLP) tasks: ViTs have a relatively fixed number of input tokens, whose attention maps can be pruned by up to 90% even with fixed sparse patterns, without severely hurting the model accuracy (e.g., <=1.5% under 90% pruning ratio); while NLP Transformers need to handle input sequences of varying numbers of tokens and rely on on-the-fly predictions of dynamic sparse attention patterns for each input to achieve a decent sparsity (e.g., >=50%). To this end, we propose a dedicated algorithm and accelerator co-design framework dubbed ViTCoD for accelerating ViTs. Specifically, on the algorithm level, ViTCoD prunes and polarizes the attention maps to have either denser or sparser fixed patterns for regularizing two levels of workloads without hurting the accuracy, largely reducing the attention computations while leaving room for alleviating the remaining dominant data movements; on top of that, we further integrate a lightweight and learnable auto-encoder module to enable trading the dominant high-cost data movements for lower-cost computations. On the hardware level, we develop a dedicated accelerator to simultaneously coordinate the aforementioned enforced denser and sparser workloads for boosted hardware utilization, while integrating on-chip encoder and decoder engines to leverage ViTCoD’s algorithm pipeline for much reduced data movements. Extensive experiments and ablation studies validate that ViTCoD largely reduces the dominant data movement costs, achieving speedups of up to 235.3×, 142.9×, 86.0×, 10.1×, and 6.8× over general computing platforms CPUs, EdgeGPUs, GPUs, and prior-art Transformer accelerators SpAtten and Sanger under an attention sparsity of 90%, respectively. Our code implementation is available at https://github.com/GATECH-EIC/ViTCoD. 

   more » « less
5. AccPar: Tensor Partitioning for Heterogeneous Deep Learning Accelerators

   https://doi.org/10.1109/HPCA47549.2020.00036  

   Song, Linghao; Chen, Fan; Zhuo, Youwei; Qian, Xuehai; Li, Hai; Chen, Yiran (February 2020, 2020 IEEE International Symposium on High Performance Computer Architecture (HPCA))

   Deep neural network (DNN) accelerators as an example of domain-specific architecture have demonstrated great success in DNN inference. However, the architecture acceleration for equally important DNN training has not yet been fully studied. With data forward, error backward and gradient calculation, DNN training is a more complicated process with higher computation and communication intensity. Because the recent research demonstrates a diminishing specialization return, namely, “accelerator wall”, we believe that a promising approach is to explore coarse-grained parallelism among multiple performance-bounded accelerators to support DNN training. Distributing computations on multiple heterogeneous accelerators to achieve high throughput and balanced execution, however, remaining challenging. We present ACCPAR, a principled and systematic method of determining the tensor partition among heterogeneous accelerator arrays. Compared to prior empirical or unsystematic methods, ACCPAR considers the complete tensor partition space and can reveal previously unknown new parallelism configurations. ACCPAR optimizes the performance based on a cost model that takes into account both computation and communication costs of a heterogeneous execution environment. Hence, our method can avoid the drawbacks of existing approaches that use communication as a proxy of the performance. The enhanced flexibility of tensor partitioning in ACCPAR allows the flexible ratio of computations to be distributed among accelerators with different performances. The proposed search algorithm is also applicable to the emerging multi-path patterns in modern DNNs such as ResNet. We simulate ACCPAR on a heterogeneous accelerator array composed of both TPU-v2 and TPU-v3 accelerators for the training of large-scale DNN models such as Alexnet, Vgg series and Resnet series. The average performance improvements of the state-of-the-art “one weird trick” (OWT) and HYPAR, and ACCPAR, normalized to the baseline data parallelism scheme where each accelerator replicates the model and processes different input data in parallel, are 2.98×, 3.78×, and 6.30×, respectively. 

   more » « less