Nuvolos Update: The Nvidia A100 GPU is Here

Nvidia’s flagship GPU, the A100, is so advanced – and high end – that reviewers coyly claimed it “isn’t for you“. Indeed, it’s a considerable investment for computational scientists or data analysts trying to maintain adequate computational hardware on their own. But with Nuvolos, it can be for you any time – as our clusters now have the option of running nodes with the Nvidia A100! This represents a huge upgrade for high performance computing (HPC) workloads. So what is it all about?

For those not in the know, a GPU is not a graphics card, but represents the computing power associated with a graphics card, needed to get the optimal performance out of the card. It handles the mathematical (especially geometrical) calculations required to make high end graphics rendering possible. GPUs tend to have more transistors and a larger density of computing cores than the CPUs most people are familiar with.

The Nvidia A100 GPU is the 8th generation GPU made by the company intended to be the state-of-the-art implementation for data centres. Nvidia markets it as specifically designed for machine learning and HPC applications, i.e. use cases such as data analytics and AI model training and deep learning inference.

In terms of architecture and design, the Nvidia A100 benefits from several improvements over its predecessors. Firstly, its Tensor cores are faster for AI and HPC workloads than the previous (CUDA) standard. Secondly, the A100 is based on Nvidia’s Ampere architecture, which combines these advanced cores with multi-instance GPUs, meaning that the GPU can be partitioned into separate instances separated at the hardware level, guaranteeing optimal usage of its capacity for concurrent workloads of different sizes and requirements. Moreover, the bandwidth has been considerably improved, with 600 GB/second GPU-to-GPU direct bandwidth and 2TB/sec for general use. This means the A100 is both smarter and faster in what it can do.

So if you have use cases relying on HPC workloads or machine learning model training, for example, you will not want to miss this opportunity. Using Nuvolos, the cloud platform for computational and data science, you can get experimenting in no time while availing yourself of market-leading GPU hardware. Everything runs directly in the browser and you can always use your favourite tools in your desired working environment. If you are not yet familiar with Nuvolos, try it out now with our free trial.