New setup speeds up AI research
Mirendil is expanding its partnership with Google Cloud to harness the AI Hypercomputer. This move supports the pre-training and post-training of advanced artificial intelligence models. This collaboration aims to break down barriers in AI development by making the research process faster and more effective.
How the system works
Mirendil’s approach involves using Google’s TPU v5P chips alongside NVIDIA hardware. This dual-architecture setup allows them to match specific workloads to the most suitable computing resources. Google Cloud has worked closely with Mirendil to deploy a comprehensive infrastructure that spans computing, storage, networking, and control planes. A key component is the managed training clusters through the Gemini Enterprise Agent Platform.
What it means for AI research
Mirendil’s ultimate goal is to expand access to frontier AI research tools. Their CEO, Behnam Neyshabur, highlighted that progress in AI has often been limited by the speed at which humans can conduct research. We’re building AI systems that can accelerate and improve that loop itself.”
The new infrastructure with Google Cloud offers Mirendil the necessary scale and flexibility to advance its AI systems further. By combining these resources, the company aims to help scientists and engineers work faster and more efficiently. This deployment reflects a growing trend in AI research, where labs increasingly need infrastructure that supports multiple types of accelerators rather than being limited to a single type of chip.
As NVIDIA accelerated computing systems come online, Mirendil will have even more computing power at its disposal. This expansion is part of a broader strategy to make advanced AI capabilities more widely available, enabling more researchers to contribute to cutting-edge projects.