25 March 2024 in IEEE Micro
Large Language Model (LLM) changes our lives, while it requires unprecedented computing resources, especially it requires large memory capacity and high bandwidth to process weights. However, while the logic process was developing, the speed of development of the memory process could not keep up, causing problems that resulted in the performance of LLM being hindered by memory. Samsung have introduced breakthrough Processing-in-Memory/Processing-near-Memory (PIM/PNM) solutions that enhance the main memory bandwidth. With the HBM-PIM-based GPU-cluster system and LPDDR5-PIM-based system, the performance of transformer-based LLMs improved by up to 1.9× and 2.7×, respectively. The CXL-based PNM solution serves memory-centric computing systems by implementing logic inside the CXL memory controller. This results in a performance gain of over 4.4× with an energy reduction of about 53% with PNM. Furthermore, we provide PIM/PNM software stacks, including an AI compiler targeting the acceleration of AI models.
2 April 2020 in IEEE Transactions on Computers
Recurrent Neural Networks (RNNs) spend most of their execution time performing matrix-vector multiplication (MV-mul). Because the matrices in RNNs have poor reusability and the ever-increasing size of the matrices becomes too large to fit in the on-chip storage of mobile/IoT devices, the performance and energy efficiency of MV-mul is determined by those of main-memory DRAM. Therefore, computing MV-mul within DRAM draws much attention. However, previous studies lacked consideration for the matrix sparsity, the power constraints of DRAM devices, and concurrency in accessing DRAM from processors while performing MV-mul. We propose a main-memory architecture called MViD, which performs MV-mul by placing MAC units inside DRAM banks. For higher computational efficiency, we use a sparse matrix format and exploit quantization. Because of the limited power budget for DRAM devices, we implement the MAC units only on a portion of the DRAM banks. We architect MViD to slow down or pause MV-mul for concurrently processing memory requests from processors while satisfying the limited power budget. Our results show that MViD provides 7.2× higher throughput compared to the baseline system with four DRAM ranks (performing MV-mul in a chip-multiprocessor) while running inference of Deep Speech 2 with a memory-intensive workload.
7 December 2020 in IEEE Computer Architecture Letters
Personalized recommendation systems are gaining significant traction due to their industrial importance. An important building block of recommendation systems consists of what is known as the embedding layers, which exhibit a highly memory-intensive characteristics. Fundamental primitives of embedding layers are the embedding vector gathers followed by vector reductions, which exhibit low arithmetic intensity and becomes bottlenecked by the memory throughput. To address this issue, recent proposals in this research space employ a near-data processing (NDP) solution at the DRAM rank-level, achieving a significant performance speedup. We observe that prior NDP solutions based on rank-level parallelism leave significant performance left on the table, as they do not fully reap the abundant data transfer throughput inherent in DRAM datapaths. Based on the observation that the datapath of the DRAM has a hierarchical tree structure, we propose a novel, fine-grained NDP architecture for recommendation systems, which augments the DRAM datapath with an "in-DRAM" reduction unit at the DDR4/5 rank/bank-group/bank level, achieving significant performance improvements over state-of-the-art approaches. We also propose hot embedding-vector replication to alleviate the load imbalance across the reduction units.
7 June 2023 in IEEE Transactions on Computers
Graph Neural Network (GNN) models have attracted attention, given their high accuracy in interpreting graph data. One of the primary building blocks of a GNN model is aggregation, which gathers and averages the feature vectors corresponding to the nodes adjacent to each node. Aggregation works by multiplying the adjacency and feature matrices. The size of both matrices exceeds the on-chip cache capacity for many realistic datasets, and the adjacency matrix is highly sparse. These characteristics lead to little data reuse, causing intensive main-memory accesses during the aggregation process. Thus, aggregation exhibits memory-intensive characteristics and dominates most of the total execution time. In this paper, we propose GraNDe, an NDP architecture that accelerates memory-intensive aggregation operations by locating NDP modules near DRAM datapath to exploit rank-level parallelism. GraNDe maximizes bandwidth utilization by separating the memory channel path with the buffer chip in between so that pre-/post-processing in the host processor and reduction in NDP modules operate simultaneously. By exploring the preferred data mappings of the operand matrices to DRAM ranks, we architect GraNDe to support adaptive matrix mapping that applies the optimal mapping for each layer depending on the dimension of the layer and the configuration of a memory system. We also propose adj-bundle broadcasting and re-tiling optimizations to reduce the transfer time for adjacency matrix data and to improve feature vector data reusability by exploiting tiling with consideration of adjacency between nodes. GraNDe achieves 3.01× and 1.69× on average, and up to 4.00× and 1.98× speedups of GCN aggregation over the baseline system and the state-of-the-art NDP architecture for GCN, respectively.
13 June 2022 in IEEE Computer Architecture Letters
Graph Convolutional Network (GCN) models have attracted attention given their high accuracy in interpreting graph data. One of the primary building blocks of a GCN model is aggregation, which gathers and averages the feature vectors corresponding to the vertices adjacent to each individual vertex. Aggregation works by multiplying the adjacency and feature matrices. The size of both matrices exceeds the on-chip cache capacity, and the adjacency matrix is highly sparse. These lead to little data reuse and cause numerous main-memory accesses during the aggregation process. Thus, aggregation exhibits memory-intensive characteristics. We propose GraNDe, an NDP architecture that accelerates memory-intensive aggregation operations by locating processing elements near the DRAM datapath to exploit rank-level parallelism. By exploring the data mapping of the operand matrices to DRAM ranks, we discover that the optimal mapping differs depending on the configuration of a specific GCN layer. With our optimal layer-by-layer mapping scheme, GraNDe shows a speedup up to 4.3× compared to the baseline system on open-graph benchmark datasets.