Achieving the UN Sustainable Development Goals (SDGs) demands a shift by industry, governments, society, and individuals to reach adequate levels of a…

Le SPN se nourrit des collaborations au sein de son réseau. Fin 2022, Game Partners évoquait le projet de construire un jeu sérieux sur le Numérique Responsable.

Smart coders can see the system is broken

Learn about optimizing your Azure carbon emissions.

Deep Learning (DL) frameworks such as PyTorch and TensorFlow include runtime infrastructures responsible for executing trained models on target hardware, managing memory, data transfers, and multi-accelerator execution, if applicable. Additionally, it is a common practice to deploy pre-trained models on environments distinct from their native development settings. This led to the introduction of interchange formats such as ONNX, which includes its runtime infrastructure, and ONNX Runtime, which work as standard formats that can be used across diverse DL frameworks and languages. Even though these runtime infrastructures have a great impact on inference performance, no previous paper has investigated their energy efficiency. In this study, we monitor the energy consumption and inference time in the runtime infrastructures of three well-known DL frameworks as well as ONNX, using three various DL models. To have nuance in our investigation, we also examine the impact of using different execution providers. We find out that the performance and energy efficiency of DL are difficult to predict. One framework, MXNet, outperforms both PyTorch and TensorFlow for the computer vision models using batch size 1, due to efficient GPU usage and thus low CPU usage. However, batch size 64 makes PyTorch and MXNet practically indistinguishable, while TensorFlow is outperformed consistently. For BERT, PyTorch exhibits the best performance. Converting the models to ONNX usually yields significant performance improvements but the ONNX converted ResNet model with batch size 64 consumes approximately 10% more energy and time than the original PyTorch model.

Today, we’re releasing a new guide explaining how to compel companies employing deceptive design to dodge data protection requests—like those for data deletion or access—to comply. It encapsulates five years...

The largest publicly available dataset of data brokers made available to the wider research community.

Alter Numeris, ce sont des chercheurs et des penseurs qui réfléchissent les enjeux de la société numérique.

If you want to save money in Enterprise IT, it turns out that sustainability as a KPI is more important than cost - this is one of the key findings coming from the recent GreenOps Survey, some research made possible by the recent ClimateAction MiniGrants fund.

Pistes d'interfaces pour un design écologique des paramètres sur mobile.

In this test face detection algorithms will determine how normal you are. 100% privacy friendly.

Le nouveau référentiel étend le RGAA pour satisfaire pleinement aux exigences européennes.

Chrome DevTools lets you slow down your website to observe and inspect how it loads.

In this post, I want to look at the two different carbon accounting models used for estimation, and in what scenario you might use one over the other.

Als ervaren digitale experts helpen we organisaties met toegankelijke technologie. Strategie, UX, websites, technologie, online marketing...

Récemment, on m’a posé la question de savoir si un site plein de visuels et de grandes images pouvait être considéré comme éco-conçu si on l’optimisait à mort. Alors : éco-conception web = frugalité des contenus ou optimisations techniques ?

Vers un numérique public responsable - Téléchargez le document au format PDF ou consultez-le gratuitement en ligne