Performance or graphical issues on older GPUs/iGPUs? Please attach your log here

Alpha 5 is the "potato PC" update. It brings options below console quality (lower internal resolution, lighter shadows, shorter draw distance and more), support for older GPUs and integrated graphics, and a Vulkan 1.2 path where we can. To aim it at the hardware you actually have, we need reports from it.

Please comment here if Savage Wind has any of these problems on your older GPU or integrated graphics (Intel UHD / Iris, AMD Vega / Radeon iGPUs, GTX 900 / 1000, RX 400 / 500 and the like):

Performance: low fps, stutter or long hitches.
Graphics: missing or wrong textures, black or flickering areas, wrong colours, broken shadows, water or effects, corrupted geometry.
It doesn't start because your GPU only has Vulkan 1.2.

Include:

savagewind.log of that run, attached (drag the file into the comment). On Windows it is next to savagewind.exe; on Linux, in ~/.local/share/savagewind/. It already names your GPU, driver and settings.
Your CPU, GPU and RAM, and for a laptop, whether it was plugged in.
Where and how it happens: the place, the resolution and the fps you see (the performance overlay shows them). In game, the console (F12 or the key left of 1) has a report command. It copies your version, hardware, settings and exact place to the clipboard: paste it here.
For graphical issues, a screenshot or a short video, and whether it happens everywhere or only in some places.
Optional, very useful for performance: one run started with SAVAGEWIND_VK_PROF=2. On Windows, open a command prompt in the Savage Wind folder, run set SAVAGEWIND_VK_PROF=2, then savagewind.exe. Play a minute where it's slow and attach that log.

Use the latest release (Alpha 4.1) and game version 1.9.0. Crashes and other bugs still go in their own issue through the bug report form.

Thanks for helping us make it playable everywhere!

https://github.com/user-attachments/files/34230111/savagewind.log
Rx 560 4gb, I7-4770 16gb ram
Every place has worse perfomance now with Alpha 4
Linux Cachy OS

First of all, huge thanks to the team for the dedication to this project

For now, perfomance on my RX 580 is pretty poor. Heavy stuttering throughout and low FPS.

System specs: Ryzen 5 3600, RX 580, 16GB Ram
Perfomance: Heavy stuttering and low fps across all areas of the game

Log file

savagewind.log

Hi guys,

Geforce 1660ti
CPU intel 9400F
32Go Memory
SSD
Menu flawless, but once spawn in game , FPS drop to 5

logs attached.

savagewind.log

crash at "*** presents"

Intel(R) Core(TM) i5-7500 CPU @ 3.40GHz
NVIDIA GeForce GT 730
8GB DDR4

[savagewind] Savage Wind 0.4.1-alpha.4 (commit 428d9cc, tree 976539b)
[paths] data: C:\Users\USER\Downloads\SavageWind-0.4.1-alpha.4-windows\Savage Wind
[settings] loaded savagewind.ini
[window] display 1: 270DA23F-B 1920x1080 (primary)
[window] display 2: E1941 768x1366
[window] display 1: desktop 1920x1080 (density 1.00, 60.00 Hz), 18 fullscreen sizes; offered: 1920x1080 640x360 960x540 1280x720 1280x800 1680x1050 640x480 800x600 1024x768 1280x1024
[window] window 1280x720
[vk] application version 0.4.104 (Savage Wind 0.4.1-alpha.4)
[vk] gpu 0: NVIDIA GeForce GT 730 (api 1.2.175)
[vk] gpu 1: Intel(R) HD Graphics 630 (api 1.3.215)
[vk] GPU overlap on (the game's frame fence waited one frame late; SAVAGEWIND_GPU_OVERLAP=0: off)
[vk] global memory the HUD reads: copied at translation (final image)
[vk] shader global memory: one binding (storage buffer limit 1023 MB), the HUD's reads through copies
[vk] using Intel(R) HD Graphics 630
[vk] view format lists on
[vk] GPU checkpoints on (a device loss logs the pass the GPU was in; SAVAGEWIND_VK_CHECKPOINTS=0: off)
[vk] driver: Intel Corporation Intel driver (id 5), ASTC sampled by the GPU
[vk] maintenance5 not supported
[vk] graphics pipeline libraries off
[vk] attachment feedback loops off (target copies)
[vram] blocks ahead of need: off (SAVAGEWIND_VRAM_HEADROOM=1: on); video memory 4041 MB, budget 3637 MB (VK_EXT_memory_budget), in use 8 MB
[vk] pipeline cache: cache/vk_pipelines.bin, 0.1 MB
[vk] swapchain 1280x720, 4 images, present mode immediate (VSync mailbox; surface has immediate, fifo), window, display 60.00 Hz
[pso] pipeline list: 22 pipelines (0 damaged and 0 repeated records dropped)
[pso] pre-warm: 22 pipelines on 2 threads
[vk] recorder thread on
[pso] pre-warm done in 10 ms: 22 pipelines built, 0 built first by draws, 0 skipped (shaders missing or state unsupported); pipeline cache 22 hits, 0 misses
[vk] presents on the calling thread
[vk] swapchain 1280x720, 4 images, present mode fifo (VSync fifo relaxed; surface has immediate, fifo), window, display 60.00 Hz
[gamelib] extracted/swgame.dll (game version 1.9.0, 87 sites)
[gamelib] target: -march=x86-64-v3
[gamelib] main 7000000000-7001e98000 125144 functions
[gamelib] subsdk0 7001e98000-7002504000 13936 functions
[gamelib] sdk 7002504000-7003159000 40516 functions
loaded main at 7000000000 size 1e98000 syms 510
loaded subsdk0 at 7001e98000 size 66c000 syms 10111
loaded sdk at 7002504000 size c55000 syms 23066
relocated main (223994 packed relative, 50 imports bound to host stubs)
relocated subsdk0 (3424 packed relative, 2 imports bound to host stubs)
relocated sdk (15013 packed relative, 0 imports bound to host stubs)
[cpu] 4 cores, 4 logical processors: all logical processors
main thread: prio 44, stack 0x100000, program 01007ef00011e000
nn::init::Start at 7002614ee0 (init, init, fini)
init sdk (DT_INIT 7002504400, 83 init_array entries)
[sm] GetService lm
[ipc] control 3 on lm (handle 103)
[ipc] control 0 on lm (handle 103)
[sm] GetService apm
[sm] GetService appletOE
[ipc] control 3 on appletOE (handle 105)
[ipc] control 0 on appletOE (handle 105)
[am] ICommonStateGetter cmd 900 (unhandled, ok; logged once)
[am] ISelfController cmd 91 (unhandled, ok; logged once)
[kernel] CreateThread entry=700264c470 stack=72000eefe0 prio=28 core 3 -> 10a
[am] IWindowController cmd 10 (unhandled, ok; logged once)
[am] ISelfController cmd 13 (unhandled, ok; logged once)
[am] ISelfController cmd 16 (unhandled, ok; logged once)
[sm] GetService aoc:u
[sm] GetService ectx:aw
[ipc] no HLE implementation for service 'ectx:aw', using logging stub
[ipc] control 3 on ectx:aw (handle 10c)
[ipc] control 0 on ectx:aw (handle 10c)
[sm] GetService pctl
[ipc] control 3 on pctl (handle 10d)
[ipc] control 0 on pctl (handle 10d)
[sm] GetService pctl:s
[ipc] control 3 on pctl:s (handle 10e)
[ipc] control 0 on pctl:s (handle 10e)
[ipc] close session pctl:s (handle 10e)
[sm] GetService pctl:a
[ipc] control 3 on pctl:a (handle 10e)
[ipc] control 0 on pctl:a (handle 10e)
[ipc] close session pctl:a (handle 10e)
[sm] GetService pctl:r
[ipc] control 3 on pctl:r (handle 10e)
[ipc] control 0 on pctl:r (handle 10e)
[ipc] close session pctl:r (handle 10e)
[pctl] IParentalControlService cmd 1 (ok)
[sm] GetService time:u
[ipc] control 3 on time:u (handle 10e)
[ipc] control 0 on time:u (handle 10e)
[time] shared memory mapped at 7021f86000
init subsdk0 (DT_INIT 7001e98020, 45 init_array entries)
init main (DT_INIT 700000e8e0, 1040 init_array entries)
[options] title menu: Downloadable Content left out of 3 button lists
[lm] D:/home/Cafe/U-King/trunk/Game/Lib/sead/engine/library/modules/src/basis/nin/seadRawPrintNin.cpp | PrintStringImpl | | MainThread | 0000002e

[am] ICommonStateGetter cmd 66 (unhandled, ok; logged once)
[am] ISelfController cmd 14 (unhandled, ok; logged once)
[am] ISelfController cmd 11 (unhandled, ok; logged once)
[am] ISelfController cmd 12 (unhandled, ok; logged once)
[sm] GetService fsp-srv
[ipc] control 3 on fsp-srv (handle 111)
[ipc] control 0 on fsp-srv (handle 111)
[ipc] control 2 on fsp-srv (handle 111)
[ipc] control 2 on fsp-srv (handle 112)
[input] mouse held in the window (camera)
[fs] romfs image extracted/romfs: 14431 MB, laid out in 4.31 s
[fs] Inventory clip layout: root 1280x720, pane 10.0 210.0 700.0x400.0, previous item x 56.0
[fs] HUD letters: 12 button letters use the icon font (archive 6348559 -> 6571550 bytes, pack +224 KB, 432 ms)
[fs] voices: 1 files bigger than the game's own, 1 ResourceSizeTable entries grown (17 KB in all)
[fs] overlay: 3 romfs files replaced
[fs] romfs index: 41062 files
[fs] romfs reads: 1 (0 KB)
[fs] romfs reads: 2 (0 KB)
[fs] romfs reads: 4 (36 KB)
[vk] Intel(R) HD Graphics 630, swapchain 1280x720
[nvn] device initialized
[nvn] window origin mode 1
[nvn] window with 2 textures
[kernel] CreateThread entry=700264c470 stack=72067c8fe0 prio=30 core -2 -> 118
[kernel] CreateThread entry=700264c470 stack=7211299fe0 prio=48 core -2 -> 119
[kernel] CreateThread entry=700264c470 stack=7221afdfe0 prio=43 core -2 -> 11a
[sm] GetService hid
[ipc] control 3 on hid (handle 11b)
[input] GetNpadStyleSet id 0
[input] GetNpadState id 0 lr 7000eb1afc
[motion] six-axis sensor handles for npad 0 (style 0x1): 1
[input] GetNpadStyleSet id 0x1
[input] GetNpadStyleSet id 0x2
[input] GetNpadStyleSet id 0x3
[kernel] CreateThread entry=700264c470 stack=721f8d1fe0 prio=44 core -2 -> 11e
[am] GetDesiredLanguage -> es-419
[am] GetDesiredLanguage -> es-419
[sm] GetService acc:u0
[ipc] control 3 on acc:u0 (handle 11e)
[ipc] control 0 on acc:u0 (handle 11e)
[acc] acc:u0 cmd 160 (ok)
[svc] aoc:u cmd 50 (ok)
[kernel] CreateThread entry=700264c470 stack=7205929fe0 prio=44 core -2 -> 11f
[kernel] CreateThread entry=700264c470 stack=722836afe0 prio=44 core -2 -> 120
[kernel] CreateThread entry=700264c470 stack=723bdc8fe0 prio=44 core -2 -> 121
[kernel] CreateThread entry=700264c470 stack=721de3efe0 prio=45 core -2 -> 122
[kernel] CreateThread entry=700264c470 stack=7229b47fe0 prio=47 core -2 -> 123
[kernel] CreateThread entry=700264c470 stack=72101ccfe0 prio=47 core -2 -> 124
[kernel] CreateThread entry=700264c470 stack=7238f90fe0 prio=48 core -2 -> 125
[kernel] CreateThread entry=700264c470 stack=7228709fe0 prio=48 core -2 -> 126
[kernel] CreateThread entry=700264c470 stack=723f7a0fe0 prio=48 core -2 -> 127
[kernel] CreateThread entry=700264c470 stack=7225f17fe0 prio=48 core -2 -> 128
[fs] romfs reads: 8 (2566 KB)
[aoc] ListAddOnContent(0, 2): 0 of 0
[input] GetNpadState id 0 lr 7000eb1afc
[input] GetNpadState id 0 lr 7000eb1afc
[input] GetNpadState id 0 lr 7000eb1afc
[input] GetNpadState id 0 lr 7000eb1afc
[input] GetNpadState id 0 lr 7000eb1afc
[fs] romfs reads: 16 (62036 KB)
[nvn] command lists run on the submission thread
[audren-sdk] OpenAudioRenderer 48000 Hz, 240 samples, 30 mix buffers, 96 voices, 14 effects, device 1, mode 1, work 0x68000 -> 0
[kernel] CreateThread entry=700264c470 stack=723f42cfe0 prio=28 core 3 -> 129
[kernel] CreateThread entry=700264c470 stack=72251eefe0 prio=28 core 3 -> 12a
[kernel] CreateThread entry=700264c470 stack=72170b8fe0 prio=31 core 1 -> 12b
[audren] device 48000 Hz, 480 frames/period; queue target 1200 samples
[kernel] CreateThread entry=700264c470 stack=721b66efe0 prio=32 core 1 -> 12c
[sm] GetService audren:u
[ipc] control 3 on IAudioDevice (handle 12e)
[ipc] close session IAudioDevice (handle 12e)
[ipc] control 3 on IAudioDevice (handle 12e)
[ipc] close session IAudioDevice (handle 12e)
[ipc] control 3 on IAudioDevice (handle 12e)
[ipc] close session IAudioDevice (handle 12e)
[ipc] control 3 on IAudioDevice (handle 12e)
[ipc] close session IAudioDevice (handle 12e)
[kernel] CreateThread entry=700264c470 stack=723a5f5fe0 prio=46 core -2 -> 12e
[ipc] close session IAudioDevice (handle 12f)
[ipc] close session IAudioDevice (handle 12f)
[kernel] CreateThread entry=700264c470 stack=721287ffe0 prio=36 core -2 -> 12f
[kernel] CreateThread entry=700264c470 stack=722e2eafe0 prio=47 core -2 -> 132
[kernel] CreateThread entry=700264c470 stack=7239227fe0 prio=44 core -2 -> 133
[kernel] CreateThread entry=700264c470 stack=7220d54fe0 prio=44 core -2 -> 134
[kernel] CreateThread entry=700264c470 stack=721583dfe0 prio=44 core -2 -> 135
[kernel] CreateThread entry=700264c470 stack=720e9b6fe0 prio=44 core -2 -> 136
[kernel] CreateThread entry=700264c470 stack=720e79dfe0 prio=44 core -2 -> 137
[kernel] CreateThread entry=700264c470 stack=720d87dfe0 prio=44 core -2 -> 138
[kernel] CreateThread entry=700264c470 stack=722be7bfe0 prio=48 core -2 -> 139
[kernel] CreateThread entry=700264c470 stack=7201decfe0 prio=48 core -2 -> 13a
[kernel] CreateThread entry=700264c470 stack=7222033fe0 prio=48 core -2 -> 13b
[kernel] CreateThread entry=700264c470 stack=7207325fe0 prio=48 core -2 -> 13c
[kernel] CreateThread entry=700264c470 stack=72377c6fe0 prio=46 core -2 -> 13d
[fs] mounted host directory save/save_0
[sm] GetService nfp:user
[ipc] control 3 on nfp:user (handle 13e)
[ipc] control 0 on nfp:user (handle 13e)
[ipc] control 2 on nfp:user (handle 13e)
[ipc] control 2 on nfp:user (handle 13f)
[ipc] control 2 on nfp:user (handle 140)
[kernel] CreateThread entry=700264c470 stack=720b62afe0 prio=45 core -2 -> 142
[kernel] CreateThread entry=700264c470 stack=72008d9fe0 prio=45 core -2 -> 143
[kernel] CreateThread entry=700264c470 stack=7224ce7fe0 prio=48 core -2 -> 144
[kernel] CreateThread entry=700264c470 stack=7204e9dfe0 prio=48 core -2 -> 145
[kernel] CreateThread entry=700264c470 stack=72344aefe0 prio=48 core -2 -> 146
[kernel] CreateThread entry=700264c470 stack=722f7fafe0 prio=45 core -2 -> 144
[time] StandardUserSystemClock::GetCurrentTime (host clock)
[kernel] CreateThread entry=700264c470 stack=7204c61fe0 prio=48 core -2 -> 147
[nvn] window present interval 2
[nvn] ReportCounter type 16
[nvn] ReportCounter type 0
[vk] color targets a pass doesn't draw into left out of it: off
[vk] pass barriers hold back only fragment work where they can: off
[vk] cube-compatible arrays: every square array of 6+ layers
[vram] t=148.04 new block: 32 MB of memory type 0 (heap 0) in 0.2 ms on a game thread
[am] ICommonStateGetter cmd 54 (unhandled, ok; logged once)
[am] IApplicationFunctions cmd 67 (unhandled, ok; logged once)
[colour] LUT C:\Users\USER\Downloads\SavageWind-0.4.1-alpha.4-windows\Savage Wind\colour-luts\Savage_Wind.cube: 16^3
[vram] t=148.05 new block: 128 MB of memory type 1 (heap 0) in 0.3 ms on a game thread
[vk] samplers: 1 (driver limit 4000)
[vk] guest heap imported for shader global memory: 3120 MB
[am] ICommonStateGetter cmd 50 (unhandled, ok; logged once)
[vk] heap copy in video memory: off (not a discrete GPU, 4041 MB of video memory, heap 3120 MB, write watch yes)
[kernel] CreateThread entry=700264c470 stack=72307e5fe0 prio=48 core -2 -> 148
[vk] command counts: 0:1402 2:1294 3:123 4:3363 5:447 6:447 7:432 9:432 10:123 11:123 12:123 13:16438 15:123 17:108 18:108 19:540 20:1511 21:1511 22:1511 23:447 28:432 33:323 38:123 48:108 51:1618 54:540 61:432
[vk] t=151.0 frame 108 53.2 fps max 99ms (123 draws, 0 skipped, 216 passes, 0 dispatches, 8189 flushes, wait 1ms, compile 0 (warm 4), fbcopy 0 (+0 live), upload 0.1MB, sets 246 (0 shared), fbloop 0 (0 barriers), gmem waits 0 holds 0, uub 108 (107 same), gcopy 0 hit 153 miss (0.0 MB))
[perf] NVML: no device
[perf] PDH GPU engine counters (any vendor)
[perf] PDH 164 engine instances; e.g. "pid_1216_luid_0x00000000_0x0000D959_phys_0_eng_0_engtype_3D"; 3D sum 0.0
[kernel] CreateThread entry=700264c470 stack=720a413fe0 prio=48 core -2 -> 149
[event] Demo169_0
[vk] command counts: 0:2936 2:2710 3:359 4:7139 5:1037 6:1037 7:904 9:904 10:359 11:359 12:359 13:34610 15:359 17:226 18:226 19:1130 20:3163 21:3163 22:3163 23:1037 28:904 33:677 38:359 48:226 51:3388 54:1130 61:904
[vk] t=153.0 frame 226 59.0 fps max 25ms (236 draws, 0 skipped, 236 passes, 0 dispatches, 236 flushes, wait 0ms, compile 0 (warm 2), fbcopy 0 (+0 live), upload 0.1MB, sets 472 (0 shared), fbloop 0 (0 barriers), gmem waits 0 holds 0, uub 118 (118 same), gcopy 0 hit 472 miss (0.1 MB))
[vram] t=154.94 new block: 256 MB of memory type 0 (heap 0) in 0.9 ms on a game thread
[vk] linear render target 1x1 format 0x25 stride 128 at 72e7da7d60: copied back to guest memory
[vk] linear render target 1x1 format 0x2e stride 128 at 72e7da7de0: copied back to guest memory
[fs] romfs reads: 32 (208296 KB)
[vk] texture #3061 (nvn fmt 0x49, 32x32x32 target 2) can't be drawn into as format 95: left out of the pass
[vk] texture #3061 (nvn fmt 0x49, 32x32x32 target 2) can't be drawn into as format 95: left out of the pass
[vk] texture #3061 (nvn fmt 0x49, 32x32x32 target 2) can't be drawn into as format 95: left out of the pass
[vk] texture #3061 (nvn fmt 0x49, 32x32x32 target 2) can't be drawn into as format 95: left out of the pass
[vk] texture #3061 (nvn fmt 0x49, 32x32x32 target 2) can't be drawn into as format 95: left out of the pass
[vk] texture #3061 (nvn fmt 0x49, 32x32x32 target 2) can't be drawn into as format 95: left out of the pass
[vk] texture #3061 (nvn fmt 0x49, 32x32x32 target 2) can't be drawn into as format 95: left out of the pass
[vk] texture #3061 (nvn fmt 0x49, 32x32x32 target 2) can't be drawn into as format 95: left out of the pass
[vk] texture #3061 (nvn fmt 0x49, 32x32x32 target 2) can't be drawn into as format 95: left out of the pass
[vk] texture #3061 (nvn fmt 0x49, 32x32x32 target 2) can't be drawn into as format 95: left out of the pass
[vk] texture #3061 (nvn fmt 0x49, 32x32x32 target 2) can't be drawn into as format 95: left out of the pass
[vk] texture #3061 (nvn fmt 0x49, 32x32x32 target 2) can't be drawn into as format 95: left out of the pass
[vk] texture #3061 (nvn fmt 0x49, 32x32x32 target 2) can't be drawn into as format 95: left out of the pass
[vk] texture #3061 (nvn fmt 0x49, 32x32x32 target 2) can't be drawn into as format 95: left out of the pass
[vk] texture #3061 (nvn fmt 0x49, 32x32x32 target 2) can't be drawn into as format 95: left out of the pass
[vk] texture #3061 (nvn fmt 0x49, 32x32x32 target 2) can't be drawn into as format 95: left out of the pass
[vk] texture #3061 (nvn fmt 0x49, 32x32x32 target 2) can't be drawn into as format 95: left out of the pass
[vk] texture #3061 (nvn fmt 0x49, 32x32x32 target 2) can't be drawn into as format 95: left out of the pass
[vk] texture #3061 (nvn fmt 0x49, 32x32x32 target 2) can't be drawn into as format 95: left out of the pass
[vk] texture #3061 (nvn fmt 0x49, 32x32x32 target 2) can't be drawn into as format 95: left out of the pass
[vk] linear render target 192x192 format 0x6 stride 384 at 72dc239000: copied back to guest memory
[vk] linear render target 1x1 format 0x2e stride 128 at 72dbebdee0: copied back to guest memory
[vk] linear render target 1x1 format 0x2e stride 128 at 72dbebdf60: copied back to guest memory
[vk] GPU overlap: a one-off readback in submission 8646, waited for 8646 of 8646

*** exception c0000005 accessing 0x298
host pc 7ffc2660fab2 (igvk64.dll+5cfab2)
host lr 0 (?)
host backtrace (guest functions are f_):
#0 00007ff6136bf5d8 savagewind+af5d8
#1 00007ff61361c0e1 savagewind+c0e1
#2 00007ff61361d703 savagewind+d703
#3 00007ff6136bf776 savagewind+af776
#4 00007ffc7341c507 KERNELBASE.dll+12c507
#5 00007ffc759558d8 ntdll.dll+a58d8
#6 00007ffc7593ce46 ntdll.dll+8ce46
#7 00007ffc7595296f ntdll.dll+a296f
#8 00007ffc75902554 ntdll.dll+52554
#9 00007ffc7595147e ntdll.dll+a147e
#10 00007ffc2660fab2 igvk64.dll+5cfab2
#11 00007ffc266042dd igvk64.dll+5c42dd
#12 00007ffc26588235 igvk64.dll+548235
#13 00007ff6137072c4 savagewind+f72c4
#14 00007ff61370300d savagewind+f300d
#15 00007ffc590b766d libwinpthread-1.dll+766d
#16 00007ffc736e1bb2 ucrtbase.dll+21bb2
#17 00007ffc75527374 KERNEL32.DLL+17374
#18 00007ffc758fcc91 ntdll.dll+4cc91
[window] focus lost (frame 336)
[input] mouse free
[window] focus gained (frame 336)
[input] mouse held in the window (camera)

Windows 10 22H2
Intel (R) UHD graphics 128mb (driver 32.0.101.7092)
11th Gen Intel(R) Core(TM) i3-1115G4 @ 3.00GHz 3.00Ghz
12gb RAM

Tested on version 0.4.1, at 720p resolution, with all settings turned off.

Following the guidance to resolve the issue of yellow lines appearing across the screen, as well as glitches in effects and menus, I used "VK No Feedback Loop." The problem was fixed by this test, though it resulted in a loss of 15.

teste5(VK No Feedback Loop).log

Poor performance on an RX 570; while the graphical artifact issue has been resolved, the game remains stuck at 30 fps at 1080p. A GTX 1050 or 1650 Ti runs the game better, despite being comparable to this RX card. Through some research, I discovered that this is due to issues with Vulkan on the Polaris architecture—the RX 400 and 500 series all suffer from this same abysmal performance. (Also, the shadows look strange; I’m not sure if it’s a settings issue or a bug.)

savagewind.log

crush still remains on 0.4.1-alpha.4
Intel(R) Core(TM) i7-8665U CPU @ 1.90GHz (2.11 GHz)
Intel(R) UHD Graphics 620 (128 MB)
Vulkan Instance Version: 1.3.301

savagewind.log

savagewind.log
here my log running with SAVAGEWIND_VK_PROF=2 with some command above that looks like increase performance SAVAGEWIND_VK_FEEDBACK_LOOP=1, SAVAGEWIND_VK_NO_GPL=1

my cpu on the game always sit at 100% ( doesnt happen while playing on emu ) while my gpu variate between 44-88%
fps max i can get on that test around 35 with lowest low 18
my vram also sit around 1,2gb

running on this recomp is actually getting more higher fps than cemu & eden emu, however the cpu sit at 100% all time make it inconsistent & resulting average fps is lower than i play on cemu & eden

@Revivedx

@Revivedx

I tested the test branch built from source (commit af31601, game version 1.9.0).

Setup: CachyOS, Ryzen 5 5600GT, Vega 7 iGPU (RADV Renoir), Mesa 26.2.4, 16GB RAM
Settings: 1080p output, FSR 1 Balanced 1.7x (scene 1129x635)
Env vars: [SAVAGEWIND_VK_RECORDER=0]

Graphics issue: Graphics issue: I get color smears/blotches over the image, and some of the game's own textures and icons flicker on screen. It happens everywhere, no matter where the camera is pointing.

[With SAVAGEWIND_VK_RECORDER=0 it still happens / it goes away.]

[I did not see this on the 0.4.1-alpha.4 AppImage.]

Performance on the Great Plateau is around 50+ FPS at 1080p [without any env vars]. The FPS in the video is lower because OBS was recording.

run.log

Constant buffers copied again while the GPU overlap is on (#332)

@viith0

@viith0 Thanks for testing test. The screenshots and the log were exactly what we needed.

The smears, the magenta and white world and the flickering textures came from a change in af31601. On integrated GPUs and CPUs with 4 cores or fewer, the GPU now read the game's constant buffers in place instead of from a copy. That raced with another optimization. The game writes the next frame's constants while the GPU still draws the current one, which is only safe while the GPU reads a copy. On your Vega 7 the GPU is slow enough that it lost the race almost every frame. We could reproduce it here too, less often.

The fix is on test (commit 7c3665d): constant buffers are copied again. Until you update, SAVAGEWIND_VK_UBO_DIRECT=0 gives the same result on af31601.

You don't need SAVAGEWIND_VK_RECORDER=0 on your PC. That one was for the Intel crash.

@Revivedx

run.log

Setup: CachyOS, Ryzen 5 5600GT, Vega 7 iGPU (RADV Renoir), 16GB RAM, Mesa 26.2.4

Thanks for the fix @Revivedx, happy to help

I tested the update to commit 7c3665d all the graphical glitches and color smears on the screen are completely gone. One problem remains, though, some of the water scattered around the Great Plateau is white. Not all of it, but a good share, as you can see in the screenshots below.

@viith0

Thanks @viith0, great to hear the smears are gone. The white water is tracked in #362 (it shows up on NVIDIA and other AMD cards too, but usually only now and then, while on your Vega 7 it is there every time, which makes your setup the best place to find it). Could you start the game from a terminal once with each of these set, one at a time, go to the same water, and tell us which ones make it look normal?

SAVAGEWIND_GPU_OVERLAP=0 ./Savage_Wind-*.AppImage
SAVAGEWIND_VK_RECORDER=0 ./Savage_Wind-*.AppImage
SAVAGEWIND_VK_HEAP_COPY=0 ./Savage_Wind-*.AppImage
SAVAGEWIND_SYNC_SHADERS=1 ./Savage_Wind-*.AppImage

The pos line from the console (F12) at one of those spots would also help us stand in the same place.

@Revivedx

@Revivedx I tested the four variables one at a time. I took all the screenshots at the same lake and the same spot, as you can see below. Only SAVAGEWIND_GPU_OVERLAP=0 fixed the white water.

Setup: test branch built from source (commit 7c3665d), CachyOS, Ryzen 5 5600GT, Vega 7 (RADV Renoir), Mesa 26.2.4

pos: swr:1 x=-891.79 y=221.69 z=1878.09 yaw=101.7 pitch=-27.5 hour=10.77
(I only captured pos once, since all the screenshots were taken from the same spot.)

SAVAGEWIND_GPU_OVERLAP=0 (water looks normal, the problem is fixed):

SAVAGEWIND_VK_RECORDER=0 (some white areas are still on the water, so it doesn't fully fix it):

SAVAGEWIND_VK_HEAP_COPY=0 (didn't fix it, the water is white just like before):

SAVAGEWIND_SYNC_SHADERS=1 (didn't fix it, the water is white just like before):

White water fix with the GPU overlap

@viith0

@viith0 Thank you, that test was the key. Only SAVAGEWIND_GPU_OVERLAP=0 fixing it pointed us straight at the cause, and with your pos we could reproduce it here: on an RTX 3060 with a 4K window (so the GPU is slower than the game, like your Vega 7) the lake at your spot came out white in 11 runs out of 14.

What happened: with the GPU overlap the game reads some of its render targets back one frame late. For a few tiny 1x1 targets (light and exposure values), a late read in the first moments after starting or loading put the water in a wrong state that stayed until the next load. Savage Wind now waits for those first reads (after each loading screen). The rest of the overlap stays on, so the frame rate doesn't change.

The fix is on test (commit 0e21a86). With it, the lake was normal in 8 runs out of 8 here at the same frame rate. Could you build it and check the same lake, without SAVAGEWIND_GPU_OVERLAP=0? If you also pass by water right after leaving a shrine, tell us how it looks there too (that's where another player saw it).

If it still shows up, SAVAGEWIND_VK_WAIT_TINY_READBACKS=-1 (wait for those reads always) would tell us whether it's the same cause, and the savagewind.log of that run helps.

@Revivedx

@Revivedx I tested the new commit (0e21a86) several times, both with old saves and by starting new games. All the water on the Great Plateau looks normal to me, including the water near shrines, like the small puddle near the exit of the Shrine of Resurrection and the water near the first shrine, Oman Au. I don't know if you were referring to a specific shrine, but the water nearby was normal.

But when starting new games I ran into another bug: when the camera gets very close to the structure where Link wakes up, the walls of the starting room become invisible. After you move the camera back a little, the walls return to normal.

run.log

very unusual performance for a rx 580 2048sp. barely 30fps on native 1080p

Alpha 5 is out, with options below console quality and many fixes for older GPUs and integrated graphics. Thank you to everyone who reported here: those logs are what made it possible. Please test Alpha 5 and report what still goes wrong on integrated or older GPUs in #589. We're closing this one as a duplicate so the new reports stay together.

Navigation Menu

Repository navigation

Performance or graphical issues on older GPUs/iGPUs? Please attach your log here #332

38 remaining items

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Performance or graphical issues on older GPUs/iGPUs? Please attach your log here #332

Description

Activity

NOOBTW00 commented on Nov 1, 2026

YowYow98282 commented on Nov 1, 2026

giiggss commented on Nov 1, 2026

sebas-ai commented on Nov 1, 2026

aton27 commented on Nov 1, 2026

champagnebaby23 commented on Nov 1, 2026

Ganon78 commented on Nov 1, 2026

38 remaining items

Shiranami6 commented on Nov 7, 2026

viith0 commented on Nov 7, 2026

Revivedx commented on Nov 7, 2026

viith0 commented on Nov 7, 2026

Revivedx commented on Nov 8, 2026

viith0 commented on Nov 8, 2026

Revivedx commented on Nov 8, 2026

viith0 commented on Nov 8, 2026

kleinergoo7 commented on Nov 8, 2026

Revivedx commented on Nov 9, 2026

Metadata

Metadata

Assignees

Labels

Projects

Milestone

Relationships

Development

Issue actions