EXPLORINEER
en

Archive — US — turtlesai

Nvidia Opens NVLink Fusion to Custom CPUs and ASICs: More Flexibility in the AIEcosystem

At Computex 2025, Nvidia unveiled NVLink Fusion, a high-speed interconnect allowing for the integration of custom CPUs and ASICs with Nvidia GPUs. This development aims to enhance the AI ecosystem by promoting flexible design in accelerated systems, supported by partnerships with major tech companies. Additionally, Nvidia introduced DGX Cloud Lepton, a platform optimizing access to GPU computing resources.

Key points

  • Nvidia launched NVLink Fusion, enabling integration with custom CPUs and ASICs.
  • The new technology offers up to 14 times the bandwidth of PCIe 5.0.
  • NVLink Fusion supports two configurations: connecting custom CPUs to Nvidia GPUs and integrating Grace CPUs with non-Nvidia accelerators.
  • Strategic partnerships with Fujitsu, Qualcomm, MediaTek, and others support NVLink Fusion's ecosystem.
  • The DGX Cloud Lepton platform facilitates access to Nvidia GPUs through a decentralized 'ride-sharing' model, connecting users with GPU providers.
  • Nvidia retains control over the interconnect, requiring the use of Nvidia CPUs or GPUs for leveraging NVLink Fusion.

Why it matters

  • The rise of AI technologies and the demand for flexible, high-performance computing solutions reflect ongoing trends in the IT industry as of 2025.
  • The shift towards modular and interoperable systems aligns with increased collaboration in technology, reminiscent of practices seen during the early 2020s AI boom.
  • Decentralized computing models like DGX Cloud Lepton reflect a broader trend in cloud services, emphasizing resource sharing and accessibility in tech ecosystems.

Nvidia's introduction of NVLink Fusion and DGX Cloud Lepton at Computex 2025 marks a significant advancement in the AI ecosystem, emphasizing flexibility and modularity in computing, crucial for meeting increasing industry demands.