Search arXiv⌕ Search
Home/Archive/arXiv

arXiv · 1803.04014

NVIDIA Tensor Core Programmability, Performance & Precision

Stefano Markidis·Steven Wei Der Chien·Erwin Laure·Ivy Bo Peng·Jeffrey S. Vetter
View original source ↗Read source document ↗Export metadata ↓

Abstract

The NVIDIA Volta GPU microarchitecture introduces a specialized unit, called "Tensor Core" that performs one matrix-multiply-and-accumulate on 4x4 matrices per clock cycle. The NVIDIA Tesla V100 accelerator, featuring the Volta microarchitecture, provides 640 Tensor Cores with a theoretical peak performance of 125 Tflops/s in mixed precision. In this paper, we investigate current approaches to program NVIDIA Tensor Cores, their performances and the precision loss due to computation in mixed precision. Currently, NVIDIA provides three different ways of programming matrix-multiply-and-accumulate on Tensor Cores: the CUDA Warp Matrix Multiply Accumulate (WMMA) API, CUTLASS, a templated library based on WMMA, and cuBLAS GEMM. After experimenting with different approaches, we found that NVIDIA Tensor Cores can deliver up to 83 Tflops/s in mixed precision on a Tesla V100 GPU, seven and three times the performance in single and half precision respectively. A WMMA implementation of batched GEMM reaches a performance of 4 Tflops/s. While precision loss due to matrix multiplication with half precision input might be critical in many HPC applications, it can be considerably reduced at the cost of increased computation. Our results indicate that HPC applications using matrix multiplications can strongly benefit from using of NVIDIA Tensor Cores.

Explore related subjects

cs.DCcs.PF

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Stefano Markidis, Steven Wei Der Chien, Erwin Laure, Ivy Bo Peng, Jeffrey S. Vetter. 2018-03-11. NVIDIA Tensor Core Programmability, Performance & Precision. https://doi.org/10.1109/ipdpsw.2018.00091

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

Learning to Remember: Attentive Reinforcement Learning for Edge Serverless Autoscaling

In edge computing, the stochastic and bursty nature of serverless workloads challenges autonomous resource orchestration. Traditional reactive controllers, such as the Kubernetes Horizontal Pod Autoscaler (HPA), suffer from reaction latency, leading to Service Level Objective (SLO) violations during traffic spikes and resource flapping during ramp-downs. While Deep Reinforcement Learning (DRL) offers a pathway toward proactive management, standard agents suffer from \textit{temporal blindness}, an inability to exploit the recent temporal context in non-Markovian edge environments. To bridge this gap, we propose a stability-aware autoscaling framework unifying short-horizon temporal context and control via an Attention-Enhanced Double-Stacked LSTM architecture integrated within a Proximal Policy Optimization (PPO) agent. Unlike shallow recurrent models, our approach employs a learned attention mechanism that weights recent historical states non-uniformly, suppressing high-frequency jitter while preserving the trend that precedes demand shifts. We validate the framework on two independent Kubernetes clusters using real-world Azure Functions traces. Against the single-layer LSTM ablation and the static HPA baseline, our approach reduces P90 latency by $\approx$67\%, and holds average latency within the 50ms hard SLO for 98.8\% of the run against 49.6\% and 43.5\% respectively. Against Kubernetes Event-Driven Autoscaling (KEDA), it matches latency performance at 75\% fewer replica-steps and 59\% less churn, with P90 hard-SLO violation bursts of at most 5 consecutive intervals against up to 24 for KEDA. These results indicate that mitigating temporal blindness through deep attentive memory improves the reliability and stability of Kubernetes autoscaling under bursty edge workloads.

cs.DC↗
arXiv2026

Scheduling Coflows in Multi-Core OCS Networks with Performance Guarantee

The coflow abstraction captures application-level communication patterns and enables coordinated scheduling of parallel flows to reduce job completion times in distributed systems. Modern data center networks (DCNs) are employing multiple independent optical circuit switching (OCS) cores operating concurrently to meet the massive bandwidth demands of application jobs. However, existing coflow scheduling research primarily focuses on the single-core setting, while studies of multi-core fabrics have largely considered electrical packet switching (EPS) networks. To address this gap, this paper studies the coflow scheduling problem in multi-core OCS networks under the not-all-stop reconfiguration model, in which the reconfiguration of one circuit does not interrupt other circuits. The challenges stem from two aspects: (i) cross-core coupling induced by traffic assignment across heterogeneous cores; and (ii) per-core OCS scheduling constraints, namely \textit{port exclusivity} and \textit{reconfiguration delay}. We propose an approximation algorithm that jointly integrates cross-core flow assignment and per-core circuit scheduling to minimize the total weighted coflow completion time (CCT) and establish a provable worst-case performance guarantee. Furthermore, our algorithm framework can be applied to the multi-core EPS scenario with a corresponding approximation guarantee for packet-switched fabrics. Trace-driven simulations using real Facebook workloads demonstrate that our algorithm can reduce the total weighted CCT and tail CCT.

cs.DC↗
arXiv2026

PipeLive: Efficient Live In-place Pipeline Parallelism Reconfiguration for Dynamic LLM Serving

Pipeline parallelism (PP) is widely used to partition layers of large language models (LLMs) across GPUs, enabling scalable inference for large models. However, existing systems rely on static PP configurations that fail to adapt to dynamic settings, such as serverless platforms and heterogeneous GPU environments. Reconfiguring PP by stopping and redeploying service incurs prohibitive downtime, so reconfiguration must instead proceed live and in place, without interrupting inference. However, live in-place PP reconfiguration is fundamentally challenging. GPUs are already saturated with model weights and KV cache, leaving little room for new layer placements and necessitating KV cache resizing, at odds with systems like vLLM that preallocate for throughput. Moreover, maintaining KV consistency during execution is difficult: stop-and-copy introduces large pauses, while background synchronization risks inconsistency as states evolve. We present PipeLive, which enables live in-place PP reconfiguration with minimal disruption. PipeLive introduces a redesigned KV cache layout together with a co-designed extension to PageAttention, forming a unified mechanism for live KV resizing. It further adopts an incremental KV patching mechanism, inspired by live virtual machine migration, to synchronize KV states between source and target configurations and identify a safe switch point. PipeLive achieves a 2.5X reduction in time-to-first-token (TTFT) without KV cache overflow compared to disabling KV resizing. Furthermore, compared to a variant without KV patching, it reduces reconfiguration overhead from seconds to under 10ms, and improves TTFT and time-per-output-token (TPOT) by up to 54.7% and 14.7%, respectively.

cs.DC↗