skip to main content

Explore Research Products in the NSF-PAR

Title: AutoTM: Automatic Tensor Movement in Heterogeneous Memory Systems using Integer Linear Programming
Memory capacity is a key bottleneck for training large scale neural networks. Intel® Optane DC PMM (persistent memory modules) which are available as NVDIMMs are a disruptive technology that promises significantly higher read bandwidth than traditional SSDs at a lower cost per bit than traditional DRAM. In this work we show how to take advantage of this new memory technology to minimize the amount of DRAM required without compromising performance significantly. Specifically, we take advantage of the static nature of the underlying computational graphs in deep neural network applications to develop a profile guided optimization based on Integer Linear Programming (ILP) called AutoTM to optimally assign and move live tensors to either DRAM or NVDIMMs. Our approach can replace 50% to 80% of a system's DRAM with PMM while only losing a geometric mean 27.7% performance. This is a significant improvement over first-touch NUMA, which loses 71.9% of performance. The proposed ILP based synchronous scheduling technique also provides 2x performance over using DRAM as a hardware-controlled cache for very large networks.  more » « less
Award ID(s):
1850566
NSF-PAR ID:
10166524
Author(s) / Creator(s):
; ; ; ;
Date Published:
Journal Name:
ASPLOS '20: Proceedings of the Twenty-Fifth International Conference on Architectural Support for Programming Languages and Operating Systems
Page Range / eLocation ID:
875 to 890
Format(s):
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. ; ; ; ( , 2021 IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS))
    null (Ed.)
    Non-volatile memory (NVRAM) based on phase-change memory (such as Optane DC Persistent Memory Module) is making its way into Intel servers to address the needs of emerging applications that have a huge memory footprint. These systems have both DRAM and NVRAM on the same memory channel with the smaller capacity DRAM serving as a cache to the larger capacity NVRAM in the so called 2LM mode. In this work we analyze the performance of such DRAM caches on real hardware using a broad range of synthetic and real-world benchmarks. We identify three key limitations of DRAM caches in these emerging systems which prevent large-scale, bandwidth bound applications from taking full advantage of NVRAM read and write bandwidth. We show that software based techniques are necessary for orchestrating the data movement between DRAM and PMM for such workloads to take full advantage of these new heterogeneous memory systems. 
    more » « less
  2. ; ; ; ; ; ; ; ( , Proceedings of the VLDB Endowment)
    Deep neural networks (DNNs) are becoming increasingly deeper, wider, and non-linear due to the growing demands on prediction accuracy and analysis quality. Training wide and deep neural networks require large amounts of storage resources such as memory because the intermediate activation data must be saved in the memory during forward propagation and then restored for backward propagation. However, state-of-the-art accelerators such as GPUs are only equipped with very limited memory capacities due to hardware design constraints, which significantly limits the maximum batch size and hence performance speedup when training large-scale DNNs. Traditional memory saving techniques either suffer from performance overhead or are constrained by limited interconnect bandwidth or specific interconnect technology. In this paper, we propose a novel memory-efficient CNN training framework (called COMET) that leverages error-bounded lossy compression to significantly reduce the memory requirement for training in order to allow training larger models or to accelerate training. Our framework purposely adopts error-bounded lossy compression with a strict error-controlling mechanism. Specifically, we perform a theoretical analysis on the compression error propagation from the altered activation data to the gradients, and empirically investigate the impact of altered gradients over the training process. Based on these analyses, we optimize the error-bounded lossy compression and propose an adaptive error-bound control scheme for activation data compression. Experiments demonstrate that our proposed framework can significantly reduce the training memory consumption by up to 13.5X over the baseline training and 1.8X over another state-of-the-art compression-based framework, respectively, with little or no accuracy loss. 
    more » « less
  3. ; ; ; ( , ACM Journal on Emerging Technologies in Computing Systems)
    Recent advances in GPU-based manycore accelerators provide the opportunity to efficiently process large-scale graphs on chip. However, real world graphs have a diverse range of topology and connectivity patterns (e.g., degree distributions) that make the design of input-agnostic hardware architectures a challenge. Network-on-Chip (NoC)- based architectures provide a way to overcome this challenge as the architectural topology can be used to approximately model the expected traffic patterns that emerge from graph application workloads. In this paper, we first study the mix of long- and short-range traffic patterns generated on-chip using graph workloads, and subsequently use the findings to adapt the design of an optimal NoC-based architecture. In particular, by leveraging emerging three-dimensional (3D) integration technology, we propose design of a small-world NoC (SWNoC)- enabled manycore GPU architecture, where the placement of the links connecting the streaming multiprocessors (SM) and the memory controllers (MC) follow a power-law distribution. The proposed 3D manycore GPU architecture outperforms the traditional planar (2D) counterparts in both performance and energy consumption. Moreover, by adopting a joint performance-thermal optimization strategy, we address the thermal concerns in a 3D design without noticeably compromising the achievable performance. The 3D integration technology is also leveraged to incorporate Near Data Processing (NDP) to complement the performance benefits introduced by the SWNoC architecture. As graph applications are inherently memory intensive, off-chip data movement gives rise to latency and energy overheads in the presence of external DRAM. In conventional GPU architectures, as the main memory layer is not integrated with the logic, off-chip data movement negatively impacts overall performance and energy consumption. We demonstrate that NDP significantly reduces the overheads associated with such frequent and irregular memory accesses in graph-based applications. The proposed SWNoC-enabled NDP framework that integrates 3D memory (like Micron's HMC) with a massive number of GPU cores achieves 29.5% performance improvement and 30.03% less energy consumption on average compared to a conventional planar Mesh-based design with external DRAM. 
    more » « less
  4. ; ; ; ; ; ; ; ; ( , IEEE Transactions on Neural Networks and Learning Systems)
    The record-breaking performance of deep neural networks (DNNs) comes with heavy parameter budgets, which leads to external dynamic random access memory (DRAM) for storage. The prohibitive energy of DRAM accesses makes it nontrivial for DNN deployment on resource-constrained devices, calling for minimizing the movements of weights and data in order to improve the energy efficiency. Driven by this critical bottleneck, we present SmartDeal, a hardware-friendly algorithm framework to trade higher-cost memory storage/access for lower-cost computation, in order to aggressively boost the storage and energy efficiency, for both DNN inference and training. The core technique of SmartDeal is a novel DNN weight matrix decomposition framework with respective structural constraints on each matrix factor, carefully crafted to unleash the hardware-aware efficiency potential. Specifically, we decompose each weight tensor as the product of a small basis matrix and a large structurally sparse coefficient matrix whose nonzero elements are readily quantized to the power-of-2. The resulting sparse and readily quantized DNNs enjoy greatly reduced energy consumption in data movement as well as weight storage, while incurring minimal overhead to recover the original weights thanks to the required sparse bit-operations and cost-favorable computations. Beyond inference, we take another leap to embrace energy-efficient training, by introducing several customized techniques to address the unique roadblocks arising in training while preserving the SmartDeal structures. We also design a dedicated hardware accelerator to fully utilize the new weight structure to improve the real energy efficiency and latency performance. We conduct experiments on both vision and language tasks, with nine models, four datasets, and three settings (inference-only, adaptation, and fine-tuning). Our extensive results show that 1) being applied to inference, SmartDeal achieves up to 2.44x improvement in energy efficiency as evaluated using real hardware implementations and 2) being applied to training, SmartDeal can lead to 10.56x and 4.48x reduction in the storage and the training energy cost, respectively, with usually negligible accuracy loss, compared to state-of-the-art training baselines. Our source codes are available at: https://github.com/VITA-Group/SmartDeal. 
    more » « less
  5. ; ( , International Symposium on Memory Systems (MEMSYS))
    Recent advancements in 3D-stacked DRAM such as hybrid memory cube (HMC) and high-bandwidth memory (HBM) promise higher bandwidth and lower power consumption compared to traditional DDR-based DRAM. However, taking advantage of this additional bandwidth for improving the performance of real-world applications requires carefully laying out the data in memory which incurs significant programmer effort. To alleviate this programmer burden, we investigate application-specific address mapping to improve performance while minimizing manual effort. Our approach is guided by the following insights: (i) toggling activity of address bits can help determine strategies to improve parallelism within memory but this metric underestimates conflicts and (ii) modern memory controllers reorder address requests and therefore any toggling activity measured from an address trace is non-deterministic. Furthermore, our position is that analyzing individual address bits results in poor estimates for actual conflicts and exploited parallelism and that entropy needs to be calculated for groups of address bits. Therefore, we calculate window-based probabilistic entropy for groups of address bits to determine a near-optimal address mapping. We present simulation results for ten applications that show a performance improvement up to 25% over fixed address-mapping and up to 8% over previous application-specific address mapping for our proposed approach. 
    more » « less
  • Citation Formats
  • MLA
  • APA
  • Chicago
  • BibTeX
  • Save / Share this Record
  • Send to Email