b10472

ggml-org/llama.cppb10472Aug 17, 2026by github-actions[bot]same day as b10470

What changed since b10470

low risk

HIP builds now skip the UMA memory override so AMD APUs report memory accurately instead of over-promising.

Improvements

  • HIP builds now rely on hipMemGetInfo for memory reporting rather than the generic MemAvailable-based UMA override.
  • Memory accounting on AMD APUs with small carve-outs is more conservative and less likely to overstate available VRAM.

Bug fixes

  • Resolved issue #18159 where AMD APU systems reported inflated available memory in HIP builds.

Upgrade notes

No configuration or API changes are required; this is a drop-in patch-level bump from b10470. Only HIP/ROCm users on AMD APUs are affected, and they should see more realistic memory figures and fewer out-of-memory surprises during model loading. The CI release-tagging change from b10470 carries over unchanged.

Analyzed Oct 24, 2026

Upgrading?

Release notes summary

Fixes memory reporting on AMD APUs within HIP builds to prevent over-promising available memory on systems with small memory carve-outs.

Key Highlights

  • Skip UMA override for HIP builds
  • Fixes issue #18159 regarding memory reporting
  • Supports accurate memory reporting via hipMemGetInfo

New Features

  • Fix for AMD APU memory reporting on HIP builds

Release Notes (excerpt)

<details open> cuda : skip UMA override for HIP builds (#27083) AMD APUs report accurate memory via hipMemGetInfo. Using MemAvailable over-promises on small-carveout systems. fixes #18159 </details> **Website:** - <https://llama.app> **macOS/iOS:** - macOS Apple Silicon (arm64) - macOS Apple Silicon (arm64, KleidiAI enabled) DISABLED - macOS Intel (x64) - iOS XCFramework **Linux:** - Ubuntu x64 (CPU) - Ubuntu arm64 (CPU) - Ubuntu s390x (CPU) - Ubuntu x64 (Vulkan) - Ubuntu arm64 (Vulkan)…

Published by ggml-org on GitHub.

Read the full release notes

Related Repositories