Research map / Architectures and efficiency / RNNs and normalization
Nimble: Lightweight and Parallel GPU Task Scheduling for Deep Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use Nimble instead of PyTorch | ImageNet neural-network inference · NASNet-A (mobile), batch size 1 | inference speed: 22.34× | Helped |
| Use Nimble instead of TensorRT | ImageNet neural-network inference · NASNet-A (mobile), batch size 1 | inference speed: 2.81× | Helped |
| Use Nimble instead of TVM | ImageNet neural-network inference · EfficientNet-B5, batch size 1 | inference speed: 1.70× | Helped |
| Use Nimble instead of TVM | ImageNet neural-network inference · MobileNetV2, batch size 1 | inference speed | Hurt |
| Use Nimble instead of PyTorch | neural-network training · CIFAR-10 | training throughput: 3.61× | Helped |
| Enable multi-stream execution (ablation) | Nimble · Inception-v3 inference | speedup vs single-stream execution: 1.09× | Helped |
| Enable multi-stream execution (ablation) | Nimble · DARTS inference | speedup vs single-stream execution: 1.37× | Helped |
| Enable multi-stream execution (ablation) | Nimble · AmoebaNet inference | speedup vs single-stream execution: 1.45× | Helped |
| Enable multi-stream execution (ablation) | Nimble · NASNet-A (mobile) inference | speedup vs single-stream execution: 1.88× | Helped |
| Enable multi-stream execution (ablation) | Nimble · NASNet-A (large) inference | speedup vs single-stream execution: 1.31× | Helped |
| Minimize the run-time scheduling procedure | PyTorch inference execution · ResNet-50 | inference speed: 2.37× | Helped |
Closest papers by meaning
- CHoRD: Coordinating Scheduling and Data Placement for Efficient Deep Neural Network Inference on Chiplet-Based GPUs (NeurIPS 2026 · similarity 0.88)
- Poplar: Efficient Scaling of Distributed DNN Training on Heterogeneous GPU Clusters (AAAI 2025 · similarity 0.86)
- DEPrune: Depth-wise Separable Convolution Pruning for Maximizing GPU Parallelism (NeurIPS 2024 · similarity 0.85)
- HiRemate: Hierarchical Approach for Efficient Re-materialization of Neural Networks (ICML 2025 · similarity 0.84)
- Pfeife: Automatic Pipeline Parallelism for PyTorch (ICML 2025 · similarity 0.84)
- FlashMoE: Fast Distributed MoE in a Single Kernel (NeurIPS 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.