From 8606cb5acf01796a741bf008a279aa45dc1985e9 Mon Sep 17 00:00:00 2001
From: marx161-cmd <222994158+marx161-cmd@users.noreply.github.com>
Date: Thu, 1 Jan 2026 21:31:08 +0100
Subject: [PATCH] Document AMD iGPU fork differences

---
 README.md | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/README.md b/README.md
index 2dd2ccc36..d65b58304 100644
--- a/README.md
+++ b/README.md
@@ -1,6 +1,13 @@
 > [!IMPORTANT]
 > This is a fork of Ollama with experimental ROCm support for AMD APUs / iGPUs (UMA/GTT memory) and stability tweaks. See `docs/rocm-apu.md` for setup details.
 
+## What’s Different in This Fork
+
+- Experimental ROCm support for AMD APUs / iGPUs using UMA/GTT memory.
+- GTT-aware VRAM reporting for HIP to avoid 512 MB false limits on APUs.
+- Safe defaults for mixed CPU/GPU offload via `OLLAMA_NUM_GPU`.
+- Setup guide and runtime notes in `docs/rocm-apu.md`.
+
 <div align="center">
   <a href="https://ollama.com">
     <img alt="ollama" width="240" src="https://github.com/ollama/ollama/assets/3325447/0d0b44e2-8f4a-4e99-9b52-a5c1c741c8f7">