You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I tested version 0.3.1 and didn't encounter any graphical glitches or bugs, though the performance wasn't great. Following the release of version 4.0, there was a massive performance boost reaching up to 50 fps in Full HD but numerous graphical artifacts and glitches appeared in the menus, cutscenes, and during actual gameplay. The images show a comparison between version 0.3.1 and 0.4.0.
Hello, thanks, please refer to #332
We are aware Savage Wind doesn't work properly in older hardware and we want to consolidate all cases into one big thread.
Thanks!
Thanks, the 0.3.1 / 0.4.0 comparison is exactly what we needed. The sky slivers and the broken menu text look like a bug in the code Alpha 4 added for integrated graphics, not a hardware limit. Could you try Alpha 4.1 with these set one at a time (Command Prompt in the Savage Wind folder, set NAME=value, then savagewind.exe) and tell us which one fixes it: SAVAGEWIND_VK_LEAN_USAGE=0, SAVAGEWIND_VK_FULL_PASS_BARRIER=1, SAVAGEWIND_GPU_OVERLAP=0, SAVAGEWIND_VK_GSNAP=0, SAVAGEWIND_VK_NO_FEEDBACK_LOOP=1?
Thanks @aton27, your test pointed straight at the cause. On Intel integrated GPUs the attachment feedback loops draw wrong (sky slivers, shredded menu text), and SAVAGEWIND_VK_NO_FEEDBACK_LOOP=1 removes it, as it did for you. The next build uses the render target copies by default on Intel integrated GPUs, like 0.3.1 did, so you will not need the variable. You measured a 15 fps drop with the copies on your 2-core i3, so the cost is real; we are looking at making the copies cheaper separately. We have not found yet what is wrong in the loop path itself (we have no Intel GPU to test on). SAVAGEWIND_VK_FEEDBACK_LOOP=1 turns the loops back on for testing, if you want to try a future build with them.
Update: we found a barrier in the loop path that was missing VK_DEPENDENCY_FEEDBACK_LOOP_BIT_EXT, which a driver like Intel's may need. The next build has it. It still uses the copies by default on Intel integrated GPUs, to be safe. When that build is out, please run once with SAVAGEWIND_VK_FEEDBACK_LOOP=1 (loops on) and tell us if the flicker is gone. If it is, we turn the loops back on for Intel and you get the frame rate back.
Activity
Revivedx commentedon Nov 1, 2026
Hello, thanks, please refer to #332
We are aware Savage Wind doesn't work properly in older hardware and we want to consolidate all cases into one big thread.
Thanks!
Revivedx commentedon Nov 1, 2026
Thanks, the 0.3.1 / 0.4.0 comparison is exactly what we needed. The sky slivers and the broken menu text look like a bug in the code Alpha 4 added for integrated graphics, not a hardware limit. Could you try Alpha 4.1 with these set one at a time (Command Prompt in the Savage Wind folder,
set NAME=value, thensavagewind.exe) and tell us which one fixes it:SAVAGEWIND_VK_LEAN_USAGE=0,SAVAGEWIND_VK_FULL_PASS_BARRIER=1,SAVAGEWIND_GPU_OVERLAP=0,SAVAGEWIND_VK_GSNAP=0,SAVAGEWIND_VK_NO_FEEDBACK_LOOP=1?aton27 commentedon Nov 1, 2026
I ran all the tests, and the only one that resolved the issue was the "VK No Feedback Loop" test. However, there was a 15 FPS drop during that test.
teste5(VK No Feedback Loop).log
Revivedx commentedon Nov 2, 2026
Thanks @aton27, your test pointed straight at the cause. On Intel integrated GPUs the attachment feedback loops draw wrong (sky slivers, shredded menu text), and
SAVAGEWIND_VK_NO_FEEDBACK_LOOP=1removes it, as it did for you. The next build uses the render target copies by default on Intel integrated GPUs, like 0.3.1 did, so you will not need the variable. You measured a 15 fps drop with the copies on your 2-core i3, so the cost is real; we are looking at making the copies cheaper separately. We have not found yet what is wrong in the loop path itself (we have no Intel GPU to test on).SAVAGEWIND_VK_FEEDBACK_LOOP=1turns the loops back on for testing, if you want to try a future build with them.Update: we found a barrier in the loop path that was missing
VK_DEPENDENCY_FEEDBACK_LOOP_BIT_EXT, which a driver like Intel's may need. The next build has it. It still uses the copies by default on Intel integrated GPUs, to be safe. When that build is out, please run once withSAVAGEWIND_VK_FEEDBACK_LOOP=1(loops on) and tell us if the flicker is gone. If it is, we turn the loops back on for Intel and you get the frame rate back.Bug-fixing day (2026-11-02): Linux recorder thread, Intel iGPU, AMD W…