Intel Contributes AI Acceleration to PyTorch 2.0
March 17, 2023 | IntelEstimated reading time: 1 minute
?In the release of Python 2.0, contributions from Intel using Intel® Extension for PyTorch , oneAPI Deep Neural Network Library (oneDNN) and additional support for Intel® CPUs enable developers to optimize inference and training performance for artificial intelligence (AI).
As part of the PyTorch 2.0 compilation stack, the TorchInductor CPU backend optimization by Intel Extension for PyTorch and PyTorch ATen CPU achieved up to 1.7 times faster FP32 inference performance when benchmarked with TorchBench, HuggingFace and timm.1 This update brings notable performance improvements to graph compilation over the PyTorch eager mode.
Other optimizations include:
- Improved message-passing between adjacent neural network nodes to support graph neural network in PyTorch Geometric (PyG) for enhanced inference and performance training on Intel CPUs.
- New x86 quantization backend – a combination of FBGEMM (Facebook General Matrix-Matrix Multiplication) and oneDNN backends – replaces FBGEMM as the default quantization backend for x86 CPU platforms to enable better end-to-end int8 inference performance.
- Extended use of oneDNN with oneDNN Graph API to maximize efficient code generation on AI hardware by automatically identifying the graph partitions to be accelerated through fusion. BFloat16 and Float32 data types are supported and only inference workloads can be optimized; BF16 is only optimized on machines with AVX512_BF16 ISA support.
Suggested Items
Argonne, Toyota Collaborate on Cutting-Edge Battery Recycling Process
05/01/2024 | BUSINESS WIREThe U.S. Department of Energy’s (DOE) Argonne National Laboratory has recently launched a collaboration with Toyota Motor North America that could reduce the nation’s reliance on foreign sources of battery materials.
Micron First to Ship Critical Memory for AI Data Centers
05/01/2024 | MicronMicron Technology, Inc. announced it is leading the industry by validating and shipping its high-capacity monolithic 32Gb DRAM die-based 128GB DDR5 RDIMM memory in speeds up to 5,600 MT/s on all leading server platforms.
Danfoss Awarded Scanfil
05/01/2024 | ScanfilScanfil received an award from Danfoss in regards of delivery performance, quality, and customer service. We are honored to receive this award.
TSMC Celebrates 30th North America Technology Symposium
04/29/2024 | TSMCTSMC unveiled its newest semiconductor process, advanced packaging, and 3D IC technologies for powering the next generation of AI innovations with silicon leadership at the Company’s 2024 North America Technology Symposium.
CACI Awarded $1.3 Billion Task Order to Provide Communications and Information Technology Expertise
04/26/2024 | CACI International Inc.CACI International Inc announced that it has been awarded a five-year task order worth a total estimated value of $1.3 billion to provide communications and information technology expertise to U.S. European Command (USEUCOM) and U.S. Africa Command (USAFRICOM).
Copyright © 2024 I-Connect007 | IPC Publishing Group Inc. All rights reserved.
Log in