Subscribe to the gold package and get unlimited access to Shamra Academy
Register a new userAccSS3D: Accelerator for Spatially Sparse 3D DNNs
84
0
0.0
(
0
)
Added by
Prashant Laddha
Publication date
2020
fields
Informatics Engineering
and research's language is
English
Ask ChatGPT about the research
No Arabic abstract
Semantic understanding and completion of real world scenes is a foundational primitive of 3D Visual perception widely used in high-level applications such as robotics, medical imaging, autonomous driving and navigation. Due to the curse of dimensionality, compute and memory requirements for 3D scene understanding grow in cubic complexity with voxel resolution, posing a huge impediment to realizing real-time energy efficient deployments. The inherent spatial sparsity present in the 3D world due to free space is fundamentally different from the channel-wise sparsity that has been extensively studied. We present ACCELERATOR FOR SPATIALLY SPARSE 3D DNNs (AccSS3D), the first end-to-end solution for accelerating 3D scene understanding by exploiting the ample spatial sparsity. As an algorithm-dataflow-architecture co-designed system specialized for spatially-sparse 3D scene understanding, AccSS3D includes novel spatial locality-aware metadata structures, a near-zero latency and spatial sparsity-aware dataflow optimizer, a surface orientation aware pointcloud reordering algorithm and a codesigned hardware accelerator for spatial sparsity that exploits data reuse through systolic and multicast interconnects. The SSpNNA accelerator core together with the 64 KB of L1 memory requires 0.92 mm2 of area in 16nm process at 1 GHz. Overall, AccSS3D achieves 16.8x speedup and a 2232x energy efficiency improvement for 3D sparse convolution compared to an Intel-i7-8700K 4-core CPU, which translates to a 11.8x end-to-end 3D semantic segmentation speedup and a 24.8x energy efficiency improvement (iso technology node)
rate research
Read More
Skeleton-based Graph Convolutional Networks (GCNs) models for action recognition have achieved excellent prediction accuracy in the field. However, limited by large model and computation complexity, GCNs for action recognition like 2s-AGCN have insufficient power-efficiency and throughput on GPU. Thus, the demand of model reduction and hardware acceleration for low-power GCNs action recognition application becomes continuously higher. To address challenges above, this paper proposes a runtime sparse feature compress accelerator with hybrid pruning method: RFC-HyPGCN. First, this method skips both graph and spatial convolution workloads by reorganizing the multiplication order. Following spatial convolution workloads channel-pruning dataflow, a coarse-grained pruning method on temporal filters is designed, together with sampling-like fine-grained pruning on time dimension. Later, we come up with an architecture where all convolutional layers are mapped on chip to pursue high throughput. To further reduce storage resource utilization, online sparse feature compress format is put forward. Features are divided and encoded into several banks according to presented format, then bank storage is split into depth-variable mini-banks. Furthermore, this work applies quantization, input-skipping and intra-PE dynamic data scheduling to accelerate the model. In experiments, proposed pruning method is conducted on 2s-AGCN, acquiring 3.0x-8.4x model compression ratio and 73.20% graph-skipping efficiency with balancing weight pruning. Implemented on Xilinx XCKU-115 FPGA, the proposed architecture has the peak performance of 1142 GOP/s and achieves up to 9.19x and 3.91x speedup over high-end GPU NVIDIA 2080Ti and NVIDIA V100, respectively. Compared with latest accelerator for action recognition GCNs models, our design reaches 22.9x speedup and 28.93% improvement on DSP efficiency.
This paper describes how 3D XPoint memory arrays can be used as in-memory computing accelerators. We first show that thresholded matrix-vector multiplication (TMVM), the fundamental computational kernel in many applications including machine learning, can be implemented within a 3D XPoint array without requiring data to leave the array for processing. Using the implementation of TMVM, we then discuss the implementation of a binary neural inference engine. We discuss the application of the core concept to address issues such as system scalability, where we connect multiple 3D XPoint arrays, and power integrity, where we analyze the parasitic effects of metal lines on noise margins. To assure power integrity within the 3D XPoint array during this implementation, we carefully analyze the parasitic effects of metal lines on the accuracy of the implementations. We quantify the impact of parasitics on limiting the size and configuration of a 3D XPoint array, and estimate the maximum acceptable size of a 3D XPoint subarray.
Transfer learning in natural language processing (NLP), as realized using models like BERT (Bi-directional Encoder Representation from Transformer), has significantly improved language representation with models that can tackle challenging language problems. Consequently, these applications are driving the requirements of future systems. Thus, we focus on BERT, one of the most popular NLP transfer learning algorithms, to identify how its algorithmic behavior can guide future accelerator design. To this end, we carefully profile BERT training and identify key algorithmic behaviors which are worthy of attention in accelerator design. We observe that while computations which manifest as matrix multiplication dominate BERTs overall runtime, as in many convolutional neural networks, memory-intensive computations also feature prominently. We characterize these computations, which have received little attention so far. Further, we also identify heterogeneity in compute-intensive BERT computations and discuss software and possible hardware mechanisms to further optimize these computations. Finally, we discuss implications of these behaviors as networks get larger and use distributed training environments, and how techniques such as micro-batching and mixed-precision training scale. Overall, our analysis identifies holistic solutions to optimize systems for BERT-like models.
Genomics is the foundation of precision medicine, global food security and virus surveillance. Exact-match is one of the most essential operations widely used in almost every step of genomics such as alignment, assembly, annotation, and compression. Modern genomics adopts Ferragina-Manzini Index (FM-Index) augmenting space-efficient Burrows-Wheeler transform (BWT) with additional data structures to permit ultra-fast exact-match operations. However, FM-Index is notorious for its poor spatial locality and random memory access pattern. Prior works create GPU-, FPGA-, ASIC- and even process-in-memory (PIM)-based accelerators to boost FM-Index search throughput. Though they achieve the state-of-the-art FM-Index search throughput, the same as all prior conventional accelerators, FM-Index PIMs process only one DNA symbol after each DRAM row activation, thereby suffering from poor memory bandwidth utilization. In this paper, we propose a hardware accelerator, EXMA, to enhance FM-Index search throughput. We first create a novel EXMA table with a multi-task-learning (MTL)-based index to process multiple DNA symbols with each DRAM row activation. We then build an accelerator to search over an EXMA table. We propose 2-stage scheduling to increase the cache hit rate of our accelerator. We introduce dynamic page policy to improve the row buffer hit rate of DRAM main memory. We also present CHAIN compression to reduce the data structure size of EXMA tables. Compared to state-of-the-art FM-Index PIMs, EXMA improves search throughput by $4.9times$, and enhances search throughput per Watt by $4.8times$.
Deformable convolution networks (DCNs) proposed to address the image recognition with geometric or photometric variations typically involve deformable convolution that convolves on arbitrary locations of input features. The locations change with different inputs and induce considerable dynamic and irregular memory accesses which cannot be handled by classic neural network accelerators (NNAs). Moreover, bilinear interpolation (BLI) operation that is required to obtain deformed features in DCNs also cannot be deployed on existing NNAs directly. Although a general purposed processor (GPP) seated along with classic NNAs can process the deformable convolution, the processing on GPP can be extremely slow due to the lack of parallel computing capability. To address the problem, we develop a DCN accelerator on existing NNAs to support both the standard convolution and deformable convolution. Specifically, for the dynamic and irregular accesses in DCNs, we have both the input and output features divided into tiles and build a tile dependency table (TDT) to track the irregular tile dependency at runtime. With the TDT, we further develop an on-chip tile scheduler to handle the dynamic and irregular accesses efficiently. In addition, we propose a novel mapping strategy to enable parallel BLI processing on NNAs and apply layer fusion techniques for more energy-efficient DCN processing. According to our experiments, the proposed accelerator achieves orders of magnitude higher performance and energy efficiency compared to the typical computing architectures including ARM, ARM+TPU, and GPU with 6.6% chip area penalty to a classic NNA.
Log in to be able to interact and post comments
comments
Fetching comments
Sign in to be able to follow your search criteria
