Verified, then logged.
Sun 22 Nov 2026 · 00:24 KST / 15:24 UTC한국어 | ENSearch /

Breaking01 / AI

Naver cuts SNOW image AI resources through GPU sharing

Naver reported savings from switching to LitServe and sharing GPUs but did not publish before-and-after response times or throughput.

Smartphones and a laptop lined up on a daylight-lit testing bench show successive steps of an app journey, one screen halted on a red error panel, as an out-of-focus engineer leans in from behind. AI-generated image

Naver, a Korean internet company, reported cutting GPU resources equivalent to about 67 H100 graphics processors for SNOW's image AI services.

The company published the account on 16 Nov in a post covering a presentation at its internal engineering event. The work covered 16 image AI services, with detailed savings listed as the equivalent of 67.37 H100 GPUs.

Naver said it switched to LitServe for serving, the system that handles requests to run AI models. It removed repeated image encoding and decoding and reduced waiting time for central processing units (CPUs) and graphics processing units (GPUs).

The company said it adjusted worker processes, threads and CPU limits to match actual traffic. It also moved lighter tasks to CPUs and placed multiple services on shared GPUs.

The post lists response speed, throughput, error rates and cancellation rates among the improvements. It identifies automatic optimisation based on traffic as follow-up work.

Naver did not disclose before-and-after figures for response times or throughput in the post.