Waste management & machine learning
Lixo works to develop the circular economy by creating intelligent vision solutions for the waste management and recovery industry.
By combining image sensors and visual recognition algorithms, Lixo enables all waste management stakeholders (collectors, sorting centers, recyclers, etc.) to analyze waste flows in order to better manage, sort and recycle them on an industrial scale.
Their ultimate goal: to make the recovered raw materials competitive in price and quality with virgin raw materials.
The use of cloud computing
At Lixo, we use deep learning models to detect waste in real time in collection trucks or waste treatment facilities.
Deep learning models require graphics cards, also called GPUs, to train them. This need for GPUs is irregular: sometimes we need one, then several the next day to train many models. The use of cloud computing therefore seemed relevant (scalable and payable by the minute) to meet our needs.
That said, GPUs are power-hungry and their rental is a major expense. So we were looking for a service that was both environmentally friendly and price competitive.
Qarnot, the choice of an efficient, responsible and competitive cloud provider
We decided to change our cloud provider to Qarnot because its services meet our expectations in terms of performance and cost, and because we share the same environmental values.
Qarnot values the heat released by the microprocessors whereas this energy is simply lost with other providers. Also, via the dashboard, we have easy access to information on the reduction of carbon footprint and energy consumption of our calculations.
We have access to more powerful CPUs and GPUs at a more competitive price. With Qarnot, we double the speed of our model drives.
Qarnot has a high level of customer service. The documentation and time required to set up the training environment is clearly communicated and the exchanges with the team are fluid and quick. This made the integration much easier.
Concretely, what are the benefits?
Since switching to Qarnot, our bill has been cut by a factor of four, our computer calculations are performed twice as fast and for half the cost. We estimate that our carbon footprint on model training is reduced by 80% with Qarnot.