[Microsoft] GreenAI

[will-iamalpine]

Overview

Machine Learning training consumes vast amounts of energy. In this test case, we will calculate the SCI delta between two convolutional neural networks (InceptionV3 and DenseNet) for an image classification scenario.

Sites for Software Sustainability Actions

Energy Efficiency

1. Training to be run on Azure Machine Learning GPU
2. Prior analysis has shown that InceptionV3 Outperforms DenseNet:

• 10.3% higher accuracy than DenseNet
• 13.0% less $USD than DenseNet 
• 20.0% less energy than DenseNet
• 9.83% less time to train than DenseNet

Hardware Efficiency (N/A)

This will not be an action taken in this test case. One could propose that a reduced training time would consequently reduce embodied carbon, but this is out of scope for the calculations.

Carbon Awareness

1. Time-shifting workloads
2. Using WattTime's API and the GSF Carbon Aware SDK project, we will shift the workloads to the optimal time within a 24-hour period.

Procedure

(What) Software boundary

• Cloud instance for containerized workload (containerized workloads)

(Scale) Functional unit

r = Machine Learning training job

(How) Quantification method

• energy measurement will be provided through new GPU energy telemetry that is available in Azure Machine Learning
• carbon-aware data will leverage WattTime's marginal carbon intensity API
• machine: Low priority GPU: NVIDIA T4 NC16as v3 @1.204/hr pay-as-you-go price
• dataset: Tiny ImageNet (1% subset of ImageNet): 100k images, 200 classes
• repository
• Duration: 10 epochs

(Quantify) SCI Value Calculation

Energy efficiency:

carbon-aware findings:

(Report - WIP)

Disclose the software boundary and your calculation methodology, including items that you might not have included in the previous sections

[atg-abhishek]

@buchananwp to ask the UW students who will be working on this to make a PR referencing this issue. The PR will be against an appendix.

[srini1978]

@buchananwp Will the final SCI value include the calculated value of M?

[will-iamalpine]

Yes, I will ask that the teams attempt to calculate M. I expect it to be quite difficult, but it would be a good challenge for them!

On Mon, Jan 24, 2022 at 11:11 AM Srinivasan ***@***.***> wrote: @buchananwp <https://github.com/buchananwp> Will the final SCI value include the calculated value of M? — Reply to this email directly, view it on GitHub <#33>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACEFHEUAKL3PTIGVPMTJG63UXUQU7ANCNFSM5KQ6OZKA> . Triage notifications on the go with GitHub Mobile for iOS <https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675> or Android <https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub>. You are receiving this because you were mentioned.Message ID: <Green-Software-Foundation/software_carbon_intensity/issues/216/1019929455 @github.com>

-- Best, Will

[atg-abhishek]

@buchananwp we didn't end up creating a case study for this did we? Perhaps we can pick this up again given that you've published a paper on this? We have the folder for it here: https://github.com/Green-Software-Foundation/software_carbon_intensity/tree/dev/case-studies

cc @Henry-WattTime

[will-iamalpine]

Correct: we didn't create a case study. Happy to put together a summary in the format that's required, but I'd prefer to recycle existing content if possible (e.g. reference the paper directly. What's the timeline on this? Note: we didn't incorporate embodied emissions (M). Unfortunately, I don't have bandwidth to apply these new numbers into our work. Best, Will

…

On Thu, Jul 28, 2022 at 3:06 PM Abhishek Gupta ***@***.***> wrote: @buchananwp <https://github.com/buchananwp> we didn't end up creating a case study for this did we? Perhaps we can pick this up again given that you've published a paper on this? We have the folder for it here: https://github.com/Green-Software-Foundation/software_carbon_intensity/tree/dev/case-studies cc @Henry-WattTime <https://github.com/Henry-WattTime> — Reply to this email directly, view it on GitHub <#33>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ACEFHEQNN3R4GFHB2UKO453VWKAWPANCNFSM5KQ6OZKA> . You are receiving this because you were mentioned.Message ID: <Green-Software-Foundation/software_carbon_intensity/issues/216/1198116038 @github.com>

[Henry-WattTime]

Move to guidance document, link to academic paper that relays same information in more depth.