Articles Posted in GPU-as-a-Service (GPUaaS)

Key Takeaways

• Creating an effective contract structure is a critical part of the infrastructure GPUaaS providers need to scale, helping them monetize different forms of capacity without assuming risks beyond their control.
• GPUaaS offerings now range from interruptible, on-demand capacity to reserved servers, dedicated clusters, and managed endpoints. Service levels, billing rules and remedies should reflect the unique nature of the particular offering and measure performance at the level at which a failure actually affects the customer.
• Customer-facing commitments should be tested against the host, data center, power, fiber and platform arrangements supporting the service, so providers understand where they may be taking on risk controlled by others.

GPU-as-a-Service is moving beyond arrangements focused on large, dedicated clusters. Customers can increasingly rent individual GPUs or servers on demand, reserve capacity for a defined period, purchase interruptible compute at a discount, or consume inference through a managed endpoint.

Continue reading