Matches in SemOpenAlex for { <https://semopenalex.org/work/W3113443077> ?p ?o ?g. }
Showing items 1 to 74 of
74
with 100 items per page.
- W3113443077 abstract "The ever-increasing size of modern deep neural network (DNN) architectures has put increasing strain on the hardware needed to implement them. Sparsified DNNs can greatly reduce memory costs and increase throughput over standard DNNs, if the loss of accuracy can be adequately controlled. However, sparse DNNs present unique computational challenges. Efficient model or data parallelism algorithms are extremely hard to design and implement. The recent effort MIT/IEEE/Amazon HPEC Graph Challenge has drawn attention to high-performance inference methods for large sparse DNNs. In this paper, we introduce SNIG, an efficient inference engine for large sparse DNN s. SNIG develops highly optimized inference kernels and leverages the power of CUDA Graphs to enable efficient decomposition of model and data parallelisms. Our decomposition strategy is flexible and scalable to different partitions of data volumes, model sizes, and GPU numbers. We have evaluated SNIG on the official benchmarks of HPEC Sparse DNN Challenge and demonstrated its promising performance scalable from a single GPU to multiple GPUs. Compared to the champion of the 2019 HPEC Sparse DNN Challenge, SNIG can finish all inference workloads using only a single GPU. At the largest DNN, which has more than 4 billion parameters across 1920 layers each of 65536 neurons, SNIG is up to 2.3x faster than a state-of-the-art baseline under a machine of 4 GPUs." @default.
- W3113443077 created "2021-01-05" @default.
- W3113443077 creator A5048830943 @default.
- W3113443077 creator A5088685794 @default.
- W3113443077 date "2020-09-22" @default.
- W3113443077 modified "2023-09-27" @default.
- W3113443077 title "A Novel Inference Algorithm for Large Sparse Neural Network using Task Graph Parallelism" @default.
- W3113443077 cites W2124653173 @default.
- W3113443077 cites W2625457103 @default.
- W3113443077 cites W2899244816 @default.
- W3113443077 cites W2969388332 @default.
- W3113443077 cites W2989899068 @default.
- W3113443077 cites W3100809954 @default.
- W3113443077 cites W3100839241 @default.
- W3113443077 cites W2859678021 @default.
- W3113443077 cites W2990514424 @default.
- W3113443077 doi "https://doi.org/10.1109/hpec43674.2020.9286218" @default.
- W3113443077 hasPublicationYear "2020" @default.
- W3113443077 type Work @default.
- W3113443077 sameAs 3113443077 @default.
- W3113443077 citedByCount "7" @default.
- W3113443077 countsByYear W31134430772021 @default.
- W3113443077 countsByYear W31134430772022 @default.
- W3113443077 countsByYear W31134430772023 @default.
- W3113443077 crossrefType "proceedings-article" @default.
- W3113443077 hasAuthorship W3113443077A5048830943 @default.
- W3113443077 hasAuthorship W3113443077A5088685794 @default.
- W3113443077 hasConcept C113775141 @default.
- W3113443077 hasConcept C11413529 @default.
- W3113443077 hasConcept C121332964 @default.
- W3113443077 hasConcept C154945302 @default.
- W3113443077 hasConcept C163716315 @default.
- W3113443077 hasConcept C173608175 @default.
- W3113443077 hasConcept C2776214188 @default.
- W3113443077 hasConcept C2778119891 @default.
- W3113443077 hasConcept C2781172179 @default.
- W3113443077 hasConcept C41008148 @default.
- W3113443077 hasConcept C48044578 @default.
- W3113443077 hasConcept C56372850 @default.
- W3113443077 hasConcept C61483411 @default.
- W3113443077 hasConcept C62520636 @default.
- W3113443077 hasConcept C77088390 @default.
- W3113443077 hasConceptScore W3113443077C113775141 @default.
- W3113443077 hasConceptScore W3113443077C11413529 @default.
- W3113443077 hasConceptScore W3113443077C121332964 @default.
- W3113443077 hasConceptScore W3113443077C154945302 @default.
- W3113443077 hasConceptScore W3113443077C163716315 @default.
- W3113443077 hasConceptScore W3113443077C173608175 @default.
- W3113443077 hasConceptScore W3113443077C2776214188 @default.
- W3113443077 hasConceptScore W3113443077C2778119891 @default.
- W3113443077 hasConceptScore W3113443077C2781172179 @default.
- W3113443077 hasConceptScore W3113443077C41008148 @default.
- W3113443077 hasConceptScore W3113443077C48044578 @default.
- W3113443077 hasConceptScore W3113443077C56372850 @default.
- W3113443077 hasConceptScore W3113443077C61483411 @default.
- W3113443077 hasConceptScore W3113443077C62520636 @default.
- W3113443077 hasConceptScore W3113443077C77088390 @default.
- W3113443077 hasLocation W31134430771 @default.
- W3113443077 hasOpenAccess W3113443077 @default.
- W3113443077 hasPrimaryLocation W31134430771 @default.
- W3113443077 hasRelatedWork W1595151633 @default.
- W3113443077 hasRelatedWork W1784521533 @default.
- W3113443077 hasRelatedWork W2009196736 @default.
- W3113443077 hasRelatedWork W2950263839 @default.
- W3113443077 hasRelatedWork W2982613029 @default.
- W3113443077 hasRelatedWork W3005521981 @default.
- W3113443077 hasRelatedWork W3013976982 @default.
- W3113443077 hasRelatedWork W3153190191 @default.
- W3113443077 hasRelatedWork W3192514716 @default.
- W3113443077 hasRelatedWork W4294538768 @default.
- W3113443077 isParatext "false" @default.
- W3113443077 isRetracted "false" @default.
- W3113443077 magId "3113443077" @default.
- W3113443077 workType "article" @default.