Generative AI depends on both best-in-class hardware and a lightning-fast network. Latency in cloud networks can lead to data traffic logjams that prevent companies from getting the results they need. That is why we have set up a non-blocking network for users with high demands on computing power, availability, sustainability and cost transparency. With its enhanced bandwidth, it guarantees that all data traffic will enjoy the speed required for optimal performance.
Taiga Cloud ensures that every connection from one switch to another always has sufficient capacity. That way, the full bandwidth of all connected servers can be passed on without running applications having to wait for data.
While this adds an extra layer of complexity for us, it benefits users. This is especially true for real-time applications, for which the absence of latency is a critical success factor. For example, rendering for simulations, presentations in virtual showrooms and streaming all depend on fast networks. The flow of data without interruption is also critical for Natural Language Processing and AI training models that are clustered because they are too large to run on a single machine.
NVIDIA DPUs for top performance with open design and no hidden costs
Intelligent NVIDIA BlueField-2 type network cards with specialized DPUs are a central component in Taiga’s non-blocking network. DPUs take the pressure off CPUs in managing network traffic, freeing up valuable CPU cycles for actual GenAI tasks, thereby further reducing latency.
Software engineers at Taiga Cloud are guaranteed the bandwidth they need for their applications. They can build their software stack to be scalable on a platform with an open design and no hidden costs. They receive maximum performance for GenAI applications requiring high processing power in connection with full cost transparency.
Crucially, this offer is flexible. Taiga Cloud offers on-demand GPU cloud services. That means if you want to change the resources of an instance, you can simply select a different offering.