Mmastodon TechnologySoftware first seen 5 d ago, last 5 d ago, peak #6

Intel Arc GPU doubles AI token speed with a second engine

Original: Da 5 a 13 token al secondo senza comprare nulla: la mia GPU Intel Arc aveva un secondo motore che non usava. Una serata

An Italian developer reports boosting local AI inference on his Intel Arc GPU from 5 to 13 tokens per second without buying any hardware, after discovering the card had an unused second compute engine. He spent an evening working with Vulkan and SYCL, and says an initial benchmark had misled him about real performance. He published a full write-up of the process.

Why now: Practical tips for speeding up local AI inference on consumer hardware are attracting attention from enthusiasts running models at home

Intel ArcVulkanSYCLIntel

Open on mastodon →

Evidence

• Da 5 a 13 token al secondo senza comprare nulla: la mia GPU Intel Arc aveva un secondo motore che non usava. Una serata con Vulkan, SYCL e un benchmark che mi ha mentito in faccia. Il resoconto completo. https:// noblogo.org/itsmelarry/gpu-int… · itsmelarry@mastodon.uno

API: https://socialmediatrends-api.osmike.com/v1/trends/4765385