Nvidia T4 GPUs now generally available on Google Cloud

Google Cloud has made Nvidia T4 GPUs generally available, making it the first platform to offer them globally as cloud instances

Nvidia’s T4 GPUs, unveiled earlier this year, are aimed at machine learning training and inference, data analytics and graphics workloads. Google first made them available in Brazil, India, Tokyo and Singapore in January; AWS followed up with limited G4 cloud instances in March.

Machine learning training and inference is high on the agenda for modern enterprises, but training models can be cumbersome at scale using CPUs. GPUs, once only piquing the interest of gamers, are increasingly the weapon of choice for data scientists and engineers, as they offer improved handling of multiple (if simpler) calculations in parallel. Nvidia says that up to 90 percent of the cost of machine learning is inference training.

Google already offers the pricey Nvidia V100 GPUs on its platform. In a blog post, Google said the T4 instances offered more affordable machine learning training compared to the V100s.

“The T4 offers the acceleration benefits of Turing Tensor Cores, but at a lower price,” read the blog post. “This is great for large training workloads, especially as you scale up more resources to train faster, or to train larger models.”

The Nvidia T4 GPUs have 16 GB of memory each, offering a range of precision support including FP32, FP16, INT8, and INT4 and 260 TOPs of computing performance.

The instances are now available in three US and Asia regions, one Europe and one South America region, and are all interconnected by a high-speed network.

Price wise, they come in as low as $0.29 (£0.22) per hour per GPU, with “on-demand” instances coming in at $0.95 (£0.72) per hour. Google added that users will be offered discounts for  “sustained use.”