You already know we chose Exoscale to host our infrastructure, but they also offer some outstanding features when it comes to GPU power.
Our AI stack relies heavily on the availability of GPUs, which are traditionally very expensive resources. When we started working with Exoscale, we ran a beta test with them on their more budget-friendly offerings, such as the Nvidia A5000 and a commercial-grade 3080TI. This significantly reduces the cost of these resources, allowing us to offer AI systems at a relatively low price.
Liquid-cooled goodness
So what makes this so special? Exoscale has invested heavily in water cooling in collaboration with DIGGERS GmbH. This innovation allows them to reduce power consumption by 50% and repurpose the excess heat for public heating (yes, some of Vienna’s spas will literally be heated by AI!). Of course, this also has a big impact on performance, stable temperatures mean you can extract more raw power from each chip.
Why does this matter to Kubiqo?
For us, it’s a game changer in terms of the performance you get for your money. It also makes LLM use cases accessible to more customers. From an engineering perspective, integrating water cooling into a data center is a powerful way to do more with the same amount of energy. This is especially important for the future of infrastructure technologies, where unchecked power usage could become a serious issue.
For further information on this topic, please visit the official website of Exoscale