While reading this great article https://evanhahn.com/worlds-smallest-png/ about the smallest png image possible I started thinking about how old the png standard is and how many more newer formats are out there. PNG is 27 years old and JPG is 31. Still png with jpg still seems to be the standard when looking at software out there. For example the standard wordpress themes have pngs for their images. Also the wordpress photos page returns a lot pf JPG images. We have written about the savings of using avif in comparison to png in one of our case studies: AVIF VS. PNG

A friend recently sent me a news headline in which a company is promoting that they will use their servers 15 years. While this initially sounds like a great idea, once you think about it, for a little while, you realize modern servers are far more energy efficient than old ones. This got me thinking, when is the point where the energy efficiency of the modern server outweighs the embodied carbon of the new machine. The more I thought about this the more I went down the rabbit hole. What if, the servers are powered with solar. How do I account for this? What are the servers doing? If the server is doing a lot of vector calculations a new one with a H100 card might make sense. How are they cooled? A modern server will surely produce less heat for the same operation. So it really depends on so many factors in the specific case that a general statement is hard to make. But I still wanted to get some sort of feeling. So I broke it down to the following inputs for the new and old machine:

Artificial Intelligence is on the rise and it is using huge amounts of resources. [1] [2] [3]

People are still debating on how much, but it is clear that it is quite a lot. While the training of LLMs is a prevalent discussion the inference stage is hardly discussed. This is the process of getting a model to answer a prompt. Every query that is put to a model also takes up quite a considerable amount of resources.

We started researching this topic quite some time back and wanted to see how much resources this really is and how difference models compare to each other and if different types of queries also take different amounts of energy. Coming out of this research we want to present:

Wir sind von der dena (Deutsche Energie Agentur) beauftragt worden, eine hoffentlich bahnbrechende Studie zum Thema Green Coding als Standard in der Softwareentwicklung für eine stromsparende Gestaltung der Digitalisierung zu erstellen.

Something we often encounter when talking to clients is that they often have no idea of how much energy the various components in their pipeline use. We have developed multiple projects like Eco-CI that measures the energy usage of CI/CD pipelines or Power Hog which logs energy consumption on desktops. But to date is was not possible to see how much your servers use.

In the age of digital revolution, our reliance on technology is ever-increasing, but so is our carbon footprint. Amidst this tech boom, a pressing question arises: how can we leverage technology to combat, rather than contribute to, climate change?

CPU utilization is an interesting value to use if you ware interested in either energy usage or optimizing your code.

We have written a long case study on if it is useful in the first place. Long story short: not really. It used to be a good metrics before cpus would scale dynamically, you had on type of cpu and not multiple performance/ energy cores that each do their own thing and things were a lot simpler. But in modern computers the utilization is somehow interesting but not really much worth in terms of an exact metric.

While reading this great article https://evanhahn.com/worlds-smallest-png/ about the smallest png image possible I started thinking about how old the png standard is and how many more newer formats are out there. PNG is 27 years old and JPG is 31. Still png with jpg still seems to be the standard when looking at software out there. For example the standard wordpress themes have pngs for their images. Also the wordpress photos page returns a lot pf JPG images. We have written about the savings of using avif in comparison to png in one of our case studies: AVIF VS. PNG

Standard benchmarks play a pivotal role in the field of software engineering, serving as a foundational element for ensuring quality and efficiency. The Transaction Processing Performance Council (TPC) benchmarks, in particular, have established a high standard across the database sector. These benchmarks are designed to simulate real-world database usage, offering a platform for comparing various database products in a manner that is fair, fundamental, and reproducible.

While writing the “Software Life Cycle Assessment done in the wild” article I needed to develop a little server that could do three simple tasks. Because the article became too long I decided to separate the methodology how to develop with environmental impact in mind into a dedicated part. Feel free to head over to the SLCA article and read the first part to see why we are developing this server.