At Lixo, enabling the circularity of resources is not only a target for our products. It is also an operational target that infuses the entire culture and all processes within the company. That's why, since November 2022, we have been using the solutions developed by Qarnot, a world leader in the recovery of waste heat**** emitted by IT computing units.**
Warning: this blog post is a non-sponsored praise!
Lixo's solutions are based on image recognition technologies: deep learning algorithms that detect and identify objects in photos. However, deep learning requires large computing capacities. At Lixo, we estimate the emissions related to the energy consumption of our models at 1 ton of CO2eq per year.
To control and reduce the carbon footprint of our deep learning activities, we have chosen to use Qarnot.
Qarnot is a French company, founded a little more than 10 years ago. They have developed a sustainable alternative to data centers by distributing computing power where the heat released by the computer processors can be recovered. Qarnot does not build data centers, but deploys servers in existing buildings capable of recovering the waste heat***** released by computers.
.png)
On average, this systemic approach reduces the carbon footprint of HPC by 80% compared to services offered by traditional cloud providers.
To measure these savings, Qarnot provides a dashboard displaying the CO2 emissions generated by the calculations made:
Cherry on the cake, Qarnot allows us to :
- divide by 2 the execution time of our calculations;
- divide by 4 the amount of our model training bills.
-
For a greentech company like Lixo, combining performance and sustainability is key to healthy growth. Choosing our calculation provider was therefore essential, but must be integrated into a more global strategy to measure and reduce our GHG emissions. More on Lixo's strategy in the next episode!
*Fatal heat: heat generated by a process that is not its primary purpose, and which is not recovered (Source: Multiannual Energy Programs (PPE))
