Search

The expansion of large language models (LLMs) with hundreds of billions of parameters presents significant challenges to computational resources, particularly data movement and memory bandwidth. Long-context LLMs, which process sequences of tens of thousands of tokens, further increase the demand on the memory system as the complexity in attention layers and key-value cache sizes is proportional to the context length. Processing-in-Memory (PIM) maximizes memory bandwidth by moving compute to the data and can address the memory bandwidth challenges; however, PIM is not necessarily scalable to accelerate long-context LLM because of limited per-module memory capacity and the inflexibility of fixed-functional unit PIM architecture and static memory management. In this work, we propose LoL-PIM which is a multi-node PIM architecture that accelerates long context LLM through hardware-software co-design. In particular, we propose how pipeline parallelism can be exploited across a multi-PIM module while a direct PIM access (DPA) controller (or DMA for PIM) is proposed that enables dynamic PIM memory management and results in efficient PIM utilization across a diverse range of context length. We developed an MLIR-based compiler for LoL-PIM extending a commercial PIM-based compiler where the software modifications were implemented and evaluated, while the hardware changes were modeled in the simulator. Our evaluations demonstrate that LoL-PIM significantly improves throughput and reduces latency for long-context LLM inference, outperforming both multi-GPU and GPU-PIM systems (up to 8.54x and 16.0x speedup, respectively), thereby enabling more efficient deployment of LLMs in real-world applications., Comment: 15 pages, 12 figures

Accelerating end-to-end inference of transformer-based large language models (LLMs) is a critical component of AI services in datacenters. However, diverse compute characteristics of end-to-end LLM inference present challenges as previously proposed accelerators only address certain operations or stages (e.g., self-attention, generation stage, etc.). To address the unique challenges of accelerating end-to-end inference, we propose IANUS -- Integrated Accelerator based on NPU-PIM Unified Memory System. IANUS is a domain-specific system architecture that combines a Neural Processing Unit (NPU) with a Processing-in-Memory (PIM) to leverage both the NPU's high computation throughput and the PIM's high effective memory bandwidth. In particular, IANUS employs a unified main memory system where the PIM memory is used both for PIM operations and for NPU's main memory. The unified main memory system ensures that memory capacity is efficiently utilized and the movement of shared data between NPU and PIM is minimized. However, it introduces new challenges since normal memory accesses and PIM computations cannot be performed simultaneously. Thus, we propose novel PIM Access Scheduling that manages normal memory accesses and PIM computations through workload mapping and scheduling across the PIM and the NPU. Our detailed simulation evaluations show that IANUS improves the performance of GPT-2 by 6.2$\times$ and 3.2$\times$, on average, compared to the NVIDIA A100 GPU and the state-of-the-art accelerator. As a proof-of-concept, we develop a prototype of IANUS with a commercial PIM, NPU, and an FPGA-based PIM controller to demonstrate the feasibility of IANUS., Comment: Updated version of the paper accepted to ASPLOS 2024

Graph Neural Networks (GNNs) are emerging as a formidable tool for processing non-euclidean data across various domains, ranging from social network analysis to bioinformatics. Despite their effectiveness, their adoption has not been pervasive because of scalability challenges associated with large-scale graph datasets, particularly when leveraging message passing. To tackle these challenges, we introduce NeuraChip, a novel GNN spatial accelerator based on Gustavson's algorithm. NeuraChip decouples the multiplication and addition computations in sparse matrix multiplication. This separation allows for independent exploitation of their unique data dependencies, facilitating efficient resource allocation. We introduce a rolling eviction strategy to mitigate data idling in on-chip memory as well as address the prevalent issue of memory bloat in sparse graph computations. Furthermore, the compute resource load balancing is achieved through a dynamic reseeding hash-based mapping, ensuring uniform utilization of computing resources agnostic of sparsity patterns. Finally, we present NeuraSim, an open-source, cycle-accurate, multi-threaded, modular simulator for comprehensive performance analysis. Overall, NeuraChip presents a significant improvement, yielding an average speedup of 22.1x over Intel's MKL, 17.1x over NVIDIA's cuSPARSE, 16.7x over AMD's hipSPARSE, and 1.5x over prior state-of-the-art SpGEMM accelerator and 1.3x over GNN accelerator. The source code for our open-sourced simulator and performance visualizer is publicly accessible on GitHub https://neurachip.us, Comment: Visit https://neurachip.us for WebGUI based simulations

Unscheduled treatment interruptions may lead to reduced quality of care in radiation therapy (RT). Identifying the RT prescription dose effects on the outcome of treatment interruptions, mediated through doses distributed into different organs-at-risk (OARs), can inform future treatment planning. The radiation exposure to OARs can be summarized by a matrix of dose-volume histograms (DVH) for each patient. Although various methods for high-dimensional mediation analysis have been proposed recently, few studies investigated how matrix-valued data can be treated as mediators. In this paper, we propose a novel Bayesian joint mediation model for high-dimensional matrix-valued mediators. In this joint model, latent features are extracted from the matrix-valued data through an adaptation of probabilistic multilinear principal components analysis (MPCA), retaining the inherent matrix structure. We derive and implement a Gibbs sampling algorithm to jointly estimate all model parameters, and introduce a Varimax rotation method to identify active indicators of mediation among the matrix-valued data. Our simulation study finds that the proposed joint model has higher efficiency in estimating causal decomposition effects compared to an alternative two-step method, and demonstrates that the mediation effects can be identified and visualized in the matrix form. We apply the method to study the effect of prescription dose on treatment interruptions in anal canal cancer patients.

Fully Homomorphic Encryption (FHE) enables the processing of encrypted data without decrypting it. FHE has garnered significant attention over the past decade as it supports secure outsourcing of data processing to remote cloud services. Despite its promise of strong data privacy and security guarantees, FHE introduces a slowdown of up to five orders of magnitude as compared to the same computation using plaintext data. This overhead is presently a major barrier to the commercial adoption of FHE. In this work, we leverage GPUs to accelerate FHE, capitalizing on a well-established GPU ecosystem available in the cloud. We propose GME, which combines three key microarchitectural extensions along with a compile-time optimization to the current AMD CDNA GPU architecture. First, GME integrates a lightweight on-chip compute unit (CU)-side hierarchical interconnect to retain ciphertext in cache across FHE kernels, thus eliminating redundant memory transactions. Second, to tackle compute bottlenecks, GME introduces special MOD-units that provide native custom hardware support for modular reduction operations, one of the most commonly executed sets of operations in FHE. Third, by integrating the MOD-unit with our novel pipelined $64$-bit integer arithmetic cores (WMAC-units), GME further accelerates FHE workloads by $19\%$. Finally, we propose a Locality-Aware Block Scheduler (LABS) that exploits the temporal locality available in FHE primitive blocks. Incorporating these microarchitectural features and compiler optimizations, we create a synergistic approach achieving average speedups of $796\times$, $14.2\times$, and $2.3\times$ over Intel Xeon CPU, NVIDIA V100 GPU, and Xilinx FPGA implementations, respectively.

Heterogeneous graph neural networks (HGNNs) have emerged as powerful algorithms for processing heterogeneous graphs (HetGs), widely used in many critical fields. To capture both structural and semantic information in HetGs, HGNNs first aggregate the neighboring feature vectors for each vertex in each semantic graph and then fuse the aggregated results across all semantic graphs for each vertex. Unfortunately, existing graph neural network accelerators are ill-suited to accelerate HGNNs. This is because they fail to efficiently tackle the specific execution patterns and exploit the high-degree parallelism as well as data reusability inside and across the processing of semantic graphs in HGNNs. In this work, we first quantitatively characterize a set of representative HGNN models on GPU to disclose the execution bound of each stage, inter-semantic-graph parallelism, and inter-semantic-graph data reusability in HGNNs. Guided by our findings, we propose a high-performance HGNN accelerator, HiHGNN, to alleviate the execution bound and exploit the newfound parallelism and data reusability in HGNNs. Specifically, we first propose a bound-aware stage-fusion methodology that tailors to HGNN acceleration, to fuse and pipeline the execution stages being aware of their execution bounds. Second, we design an independency-aware parallel execution design to exploit the inter-semantic-graph parallelism. Finally, we present a similarity-aware execution scheduling to exploit the inter-semantic-graph data reusability. Compared to the state-of-the-art software framework running on NVIDIA GPU T4 and GPU A100, HiHGNN respectively achieves an average 41.5$\times$ and 8.6$\times$ speedup as well as 106$\times$ and 73$\times$ energy efficiency with quarter the memory bandwidth of GPU A100., Comment: 16 pages, 17 figures; To appear in IEEE TPDS 2024

Homomorphic encryption (HE) enables computations on encrypted data by concealing information under noise for security. However, the process of bootstrapping, which resets the noise level in the ciphertext, is computationally expensive and requires a large bootstrapping key. The TFHE scheme offers a faster and programmable bootstrapping algorithm called PBS, crucial for security-focused applications like machine learning. Nevertheless, the current TFHE scheme lacks support for ciphertext packing, resulting in low throughput. This work thoroughly analyzes TFHE bootstrapping, identifies the bottleneck in GPUs caused by the blind rotation fragmentation problem, and proposes a hardware TFHE accelerator called Strix. Strix introduces a two-level batching approach to enhance the batch size in PBS, utilizes a specialized microarchitecture for efficient streaming data processing, and incorporates a fully-pipelined FFT microarchitecture to improve performance. It achieves significantly higher throughput than state-of-the-art implementations on both CPUs and GPUs, outperforming existing TFHE accelerators by a factor of 7.4.

The goal of radiation therapy for cancer is to deliver prescribed radiation dose to the tumor while minimizing dose to the surrounding healthy tissues. To evaluate treatment plans, the dose distribution to healthy organs is commonly summarized as dose-volume histograms (DVHs). Normal tissue complication probability (NTCP) modelling has centered around making patient-level risk predictions with features extracted from the DVHs, but few have considered adapting a causal framework to evaluate the safety of alternative treatment plans. We propose causal estimands for NTCP based on deterministic and stochastic interventions, as well as propose estimators based on marginal structural models that impose bivariable monotonicity between dose, volume, and toxicity risk. The properties of these estimators are studied through simulations, and their use is illustrated in the context of radiotherapy treatment of anal canal cancer patients.

While providing low latency is a fundamental requirement in deploying recommendation services, achieving high resource utility is also crucial in cost-effectively maintaining the datacenter. Co-locating multiple workers of a model is an effective way to maximize query-level parallelism and server throughput, but the interference caused by concurrent workers at shared resources can prevent server queries from meeting its SLA. Hera utilizes the heterogeneous memory requirement of multi-tenant recommendation models to intelligently determine a productive set of co-located models and its resource allocation, providing fast response time while achieving high throughput. We show that Hera achieves an average 37.3% improvement in effective machine utilization, enabling 26% reduction in required servers, significantly improving upon the baseline recommedation inference server.

Homomorphic Encryption (HE) enables users to securely outsource both the storage and computation of sensitive data to untrusted servers. Not only does HE offer an attractive solution for security in cloud systems, but lattice-based HE systems are also believed to be resistant to attacks by quantum computers. However, current HE implementations suffer from prohibitively high latency. For lattice-based HE to become viable for real-world systems, it is necessary for the key bottlenecks - particularly polynomial multiplication - to be highly efficient. In this paper, we present a characterization of GPU-based implementations of polynomial multiplication. We begin with a survey of modular reduction techniques and analyze several variants of the widely-used Barrett modular reduction algorithm. We then propose a modular reduction variant optimized for 64-bit integer words on the GPU, obtaining a 1.8x speedup over the existing comparable implementations. Next, we explore the following GPU-specific improvements for polynomial multiplication targeted at optimizing latency and throughput: 1) We present a 2D mixed-radix, multi-block implementation of NTT that results in a 1.85x average speedup over the previous state-of-the-art. 2) We explore shared memory optimizations aimed at reducing redundant memory accesses, further improving speedups by 1.2x. 3) Finally, we fuse the Hadamard product with neighboring stages of the NTT, reducing the twiddle factor memory footprint by 50%. By combining our NTT optimizations, we achieve an overall speedup of 123.13x and 2.37x over the previous state-of-the-art CPU and GPU implementations of NTT kernels, respectively., Comment: Accepted, to be pusblished at SEED 2022 conference (IEEE International Symposium on Secure and Private Execution Environment Design)

Transformers have become a predominant machine learning workload, they are not only the de-facto standard for natural language processing tasks, but they are also being deployed in other domains such as vision and speech recognition. Many of the transformer-based applications are real-time systems such as machine translation and web search. These real time systems often come with strict end-to-end inference latency requirements. Unfortunately, while the majority of the transformer computation comes from matrix multiplications, transformers also include several non-linear components that tend to become the bottleneck during an inference. In this work, we accelerate the inference of BERT models on the tensor streaming processor. By carefully fusing all the nonlinear components with the matrix multiplication components, we are able to efficiently utilize the on-chip matrix multiplication units resulting in a deterministic tail latency of 130 $\mu$s for a batch-1 inference through BERT-base, which is 6X faster than the current state-of-the-art.

Homomorphic Encryption (HE) is one of the most promising post-quantum cryptographic schemes that enable privacy-preserving computation on servers. However, noise accumulates as we perform operations on HE-encrypted data, restricting the number of possible operations. Fully HE (FHE) removes this restriction by introducing the bootstrapping operation, which refreshes the data; however, FHE schemes are highly memory-bound. Bootstrapping, in particular, requires loading GBs of evaluation keys and plaintexts from off-chip memory, which makes FHE acceleration fundamentally bottlenecked by the off-chip memory bandwidth. In this paper, we propose ARK, an Accelerator for FHE with Runtime data generation and inter-operation Key reuse. ARK enables practical FHE workloads with a novel algorithm-architecture co-design to accelerate bootstrapping. We first eliminate the off-chip memory bandwidth bottleneck through runtime data generation and inter-operation key reuse. This approach enables ARK to fully exploit on-chip memory by substantially reducing the size of the working set. On top of such algorithmic enhancements, we build ARK microarchitecture that minimizes on-chip data movement through an efficient, alternating data distribution policy based on the data access patterns and a streamlined dataflow organization of the tailored functional units -- including base conversion, number-theoretic transform, and automorphism units. Overall, our co-design effectively handles the heavy computation and data movement overheads of FHE, drastically reducing the cost of HE operations, including bootstrapping., Comment: 18 pages, 9 figures

Homomorphic encryption (HE) enables the secure offloading of computations to the cloud by providing computation on encrypted data (ciphertexts). HE is based on noisy encryption schemes in which noise accumulates as more computations are applied to the data. The limited number of operations applicable to the data prevents practical applications from exploiting HE. Bootstrapping enables an unlimited number of operations or fully HE (FHE) by refreshing the ciphertext. Unfortunately, bootstrapping requires a significant amount of additional computation and memory bandwidth as well. Prior works have proposed hardware accelerators for computation primitives of FHE. However, to the best of our knowledge, this is the first to propose a hardware FHE accelerator that supports bootstrapping as a first-class citizen. In particular, we propose BTS - Bootstrappable, Technologydriven, Secure accelerator architecture for FHE. We identify the challenges of supporting bootstrapping in the accelerator and analyze the off-chip memory bandwidth and computation required. In particular, given the limitations of modern memory technology, we identify the HE parameter sets that are efficient for FHE acceleration. Based on the insights gained from our analysis, we propose BTS, which effectively exploits the parallelism innate in HE operations by arranging a massive number of processing elements in a grid. We present the design and microarchitecture of BTS, including a network-on-chip design that exploits a deterministic communication pattern. BTS shows 5,556x and 1,306x improved execution time on ResNet-20 and logistic regression over a CPU, with a chip area of 373.6mm^2 and up to 163.2W of power., Comment: 15 pages, 10 figures