# ROCm with gfx803 on blender3 Finally, I can use cycles renderer of blender3 on the gfx803. But I cannot find the reason of problem, I just changed all of `__noinline__` to `__inline__`, and compile cycles for gfx803. At the same time, ROCm-5.2.3 always report a mem fault error, then ROCm-5.3 released, everything is ok. I am afraid the next version of ROCm may cause cycles break again, but right now, it can run successfully. |OS |linux|Python|ROCm |GPU |Blender| |--------------|-----|------|-----|-----|-------| |Ubuntu-20.04.5|5.15 |3.8.10|5.3.0|RX580|3.3.1 | Install ROCm First <https://docs.amd.com/bundle/ROCm-Installation-Guide-v5.3/page/Introduction_to_ROCm_Installation_Guide_for_Linux.html> |component |version |size |link| |--------------|----------|-------|----| |rocr-runtime |1.7.0 |902K |https://github.com/xuhuisheng/rocm-gfx803/releases/download/rocm530/hsa-rocr_1.7.0.50300-63.20.04_amd64.deb| |blender-gfx803|3.3.1 |300M |https://github.com/xuhuisheng/rocm-gfx803/releases/download/rocm530/blender331-gfx803.7z| 1. Install ROCm-5.3.0 2. `sudo dpkg -i hsa-rocr_1.7.0.50300-63.20.04_amd64.deb` 3. `7z x blender331-gfx803.7z` The `kernel_gfx803.fatbin` is on the `blender331-gfx803/3.3/scripts/addons/cycles/lib/`. --- My modification is very simple. 1. modify `intern/cycles/kernel/device/hip/compat.h`, change all of `__noinline__` to `__inline__` 2. modify `intern/cycles/device/hip/util.h`, change `9` to `8`, then cycles can use gfx803. --- I test [cube_diorama](https://www.blender.org/download/demo/bundles/bundles-3.0/asset-demo-bundle-3.0-cube-diorama.zip) successfully. But I cannot guaruntee we won't meet other problems. You could have a try with your own risk. My suggestion is just buy a ROCm offically supported card, like Vega64, RadeonVII, RDNA.   ## blender-3.4.1 and ROCm-5.4.1 |OS |linux|Python|ROCm |GPU |Blender| |--------------|-----|------|-----|-----|-------| |Ubuntu-20.04.6|5.15 |3.8.10|5.4.1|RX580|3.4.1 | I tried blender-3.4.1 with ROCm-5.4.1, at least cube demo can render properly. The modification is just same.   