AWS wellarchitected documentation change
Summary
Updated documentation links, grammatical corrections, and minor phrasing improvements in ML accelerator explanations. Changes include fixing a broken internal link, capitalizing 'Machine Learning', removing redundant adverbs, and expanding NPU acronym.
Security assessment
The changes are editorial improvements and documentation structure updates without any mention of security vulnerabilities, mitigations, or security-related features. The content focuses on performance optimization and hardware selection guidance for ML inference workloads.
Diff
diff --git a/wellarchitected/latest/iot-lens/use-accelerators-for-machine-learning-inference.md b/wellarchitected/latest/iot-lens/use-accelerators-for-machine-learning-inference.md index 1cc482a2b..f872bb4b5 100644 --- a//wellarchitected/latest/iot-lens/use-accelerators-for-machine-learning-inference.md +++ b//wellarchitected/latest/iot-lens/use-accelerators-for-machine-learning-inference.md @@ -3 +3 @@ -[Documentation](/index.html)[AWS Well-Architected](https://aws.amazon.com/architecture/well-architected/)[AWS Well-Architected Framework](abstract-and-introduction.html) +[Documentation](/index.html)[AWS Well-Architected](https://aws.amazon.com/architecture/well-architected/)[AWS Well-Architected Framework](iot-lens.html) @@ -7 +7 @@ -Performing machine learning (ML) inference on the IoT device can greatly reduce the amount of data transmitted to the cloud. ML inference is a computationally intensive process that might not be energy optimal when run on a CPU without the right instruction set. For ML applications, the use of a CPU with additional vector operations and specialized acceleration hardware can lower the energy consumption of IoT devices. +Performing Machine Learning (ML) inference on the IoT device can greatly reduce the amount of data transmitted to the cloud. ML inference is a computationally intensive process which might not be energy optimal when run on a CPU without the right instruction set. For ML applications, the use of a CPU with additional vector operations and specialized acceleration hardware can lower the energy consumption of IoT devices. @@ -9 +9 @@ Performing machine learning (ML) inference on the IoT device can greatly reduce -Inference-class devices can come in various forms such as GPUs (Graphics Processing Units), NPUs (Neural Processing Units), Digital Signal Processors (DSPs), and FPGA (Field-Programmable Gate Array) devices or CPUs with vector manipulation operators. +Inference-class devices can come in various forms such as GPUs (Graphics Processing Units), NPUs (Neural Processing Units), Digital Signal Processors (DSPs) and FPGA (Field-Programmable Gate Array) devices or CPUs with vector manipulation operators. @@ -15 +15 @@ When lower latency than what is achievable by the MCUs is desired, specialized h -DSPs are designed to perform mathematical functions very quickly without using the host MCU's clock cycles. They are more power efficient than an MCU. In a typical DSP-accelerated ML application, such as wake-word detection in a smart speaker, the DSP first processes an analog signal such as an audio or voice signal and then wakes up the host MCU from a deep-sleep mode via an interrupt. The processor can thus remain in a low-power mode while performing inference, and only wakes up when necessary for further processing or connection to the cloud. +DSPs are designed to perform mathematical functions quickly without using the host MCU's clock cycles. They are more power efficient than an MCU. In a typical DSP-accelerated ML application, such as wake-word detection in a smart speaker, the DSP first processes an analog signal such as an audio or voice signal and then wakes up the host MCU from a deep-sleep mode via an interrupt. The processor can thus remain in a low-power mode while performing inference, and only wakes up when necessary for further processing or connection to the cloud. @@ -17 +17 @@ DSPs are designed to perform mathematical functions very quickly without using t -MPUs are suitable for edge ML applications that require higher processing capabilities, such as running more complex ML models or handling larger input datasets. MPUs may also have built-in hardware accelerators for ML tasks, such as NPUs that improve ML inferencing performance. +MPUs are suitable for edge ML applications that require higher processing capabilities, such as running more complex ML models or handling larger input datasets. MPUs may also have built-in hardware accelerators for ML tasks, such as Neural Processing Units (NPUs) which improves ML inferencing performance. @@ -21 +21 @@ NPUs are optimized for artificial neural networks. If your application involves -GPUs are specialized processors designed for graphics-intensive tasks, but can offer high performance for ML inference. If you are already using a deep learning framework or software that is optimized for GPUs, it may be more convenient to continue using GPUs. GPUs are not very power efficient and should only be selected for the highest intensity workloads. +GPUs are specialized processors designed for graphics-intensive tasks, but can offer high performance for ML inference. If you are already using a deep learning framework or software that is optimized for GPUs, it may be more convenient to continue using GPUs. GPUs are not power efficient and should only be selected for the highest intensity workloads.