Efficient GPU Usage in Cloud Services: Is Pay-per-Use Feasible?
Do others feel that a significant amount of time on a rented GPU-based instance is spent not utilizing the GPU? Often, the time is consumed debugging code, which mainly involves the CPU rather than the GPU. This leads to paying for GPU time that isn’t actually being used. Are there any cloud providers that offer a pay-per-use model specifically for GPU usage, rather than paying for the entire instance? Is this an issue for anyone else?
(
2 comments
| on HN
)