gcp Google Cloud Blog ·

Mirendil to use Google Cloud's AI Hypercomputer with TPUs and GPUs

blogaigcpengineer
announcement feature

Mirendil, an AI research lab, will utilize Google Cloud's AI Hypercomputer, leveraging a mix of TPUs and NVIDIA GPUs for model pre-training and post-training applications. This partnership aims to accelerate AI development and democratize research by providing scalable and flexible compute infrastructure. Mirendil is already deploying TPU v5p chips, with NVIDIA systems to follow, and is utilizing managed training clusters within Gemini Enterprise Agent Platform.

  • Mirendil partners with Google Cloud for AI Hypercomputer access
  • Flexible compute options for AI workloads
  • Collaborative infrastructure design and deployment
  • Mirendil's current and upcoming infrastructure deployment
Features (1)
  • Mirendil partners with Google Cloud for AI Hypercomputer access

    Mirendil will use Google Cloud's AI Hypercomputer, combining Google's TPU AI accelerators with NVIDIA AI infrastructure. This setup is designed to support Mirendil's model pre-training and post-training applications, aiming to accelerate and democratize AI research.

Enhancements (1)
  • Flexible compute options for AI workloads

    The ability to choose a mix of TPUs and NVIDIA's accelerated computing platform allows Mirendil to quickly access critical compute and match workloads to the most suitable architecture. This flexibility is key to managing complex, end-to-end training workflows and reinforcement learning at scale.

Maintenance (1)
  • Collaborative infrastructure design and deployment

    Google Cloud partnered with Mirendil on the end-to-end design and deployment of combined TPU and NVIDIA AI infrastructure, including compute, storage, networking, and control planes. Collaboration also extended to a system using managed training clusters in Gemini Enterprise Agent Platform to streamline provisioning and management.

Notes (1)
  • Mirendil's current and upcoming infrastructure deployment

    Mirendil has already deployed a cluster of TPU v5p chips and will soon bring NVIDIA AI accelerated computing systems online. This expansion on Google Cloud is intended to provide the scale and flexibility needed to push AI systems further.

Read the original announcement →

https://cloud.google.com/blog/topics/startups/mirendil-selects-ai-hypercomputer/

Related releases