You are not logged in.

Pages: 1

#1 2025-07-04 20:54:20

Loukl
Member
Registered: 2025-07-04
Posts: 7

Trouble getting the expected performance out of an AMD GPU

Here is my current Setup:

Both Arch Linux x86_64 6.15.4-zen2-1-zen and Arch Linux x86_64 linux-6.15.4.arch2-1
Plasma 6.4.2 (wayland)
CPU: AMD Ryzen Threadripper 3960X
Mainboard: ASUS ROG Zenith II Extreme
GPU: AMD ATI Radeon RX 7900 XTX (resizable BAR enabled)
Screen: 2560x1440@165Hz
4x DDR4 unbuffered ECC RAM at 3200MT/s to saturate all 4 Mem channels

Some applications/games seem to not actually use the GPU to an expected extent.
For example:
Unigine Superposition (4K preset/1080p) happily spikes the GPU to almost 3GHz and ~350W wich is to be expected. Even on 1080p (Medium or above) the GPU still draws about 200W while clocking at ~1.6GHz. In both cases the GPU usage is 80% or above.
If I however run some software like vkcube, vkmark or glmark2 for example the GPU is stuck at 1.2GHz, which still is more than idle, but wont draw more than 70W. Don't get me wrong, low power usage is great but since the GPU only shows a usage of 60% or below combined with the fact that LACT shows NO throttling but TEMP or POWER related throttling when running Superposition, it suggests that something is not working as it should.

This shows particularly annoying in games like CS2 with frame rates usually around 100 (lower than the 165 refresh rate with either V-sync on or off) but occasional drops to sub 20 mostly when looking at an enemy.

I've had this problem for a couple of years now and only upgraded from a 6900XT some months ago in the hopes I just needed some more power. However the problem still persisted and the performance increase from 6900XT to 7900XTX was not noticeable at all in those applications.

So here are things I already tried:

  • Rendering in Blender via HIP in either OpenGL or Vulkan mode gives expected Usage and power draw.

  • Setting the NUMA nodes in BIOS (CPU has multiple chiplets wich are connected to RAM and PCIe via a central chiplet AFAIK so this should not change too much) and running via "numactl --cpunodebind=3 --localalloc %command%" (6 Threads / 3 Cores) does not seem to affect performance but depending on the application does lower CPU usage wich is to be expected when reducing cores ig

  • Reinstalling archlinux did not help either

  • Switching from vulkan-radeon to amdvlk improves FPS by +20% and raises GPU Usage to roughly 75% with 100W power draw at 1.9GHz with vkcube. CS2 also experiences a slight improvement in performance but for some reason does not output sound.

  • Forcing the GPU min Freq to 2GHz with LACT does not affect performance or power draw while running the affected software and forcing VRAM min Freq does not even affect VRAM clocks

  • CS2 actually shows both behaviors: while neither being in game nor the main screen do utilize the GPU to full extend, inspecting a weapon or going into advanced video settings does

Some logs:

dmesg mentioning amdgpu (The edid override is just adjusting the blank times to fix the VRAM clock being stuck at max)

[  +0.000000] Command line: BOOT_IMAGE=/vmlinuz-linux-zen root=UUID=a2e22b4d-7352-445f-b6ad-74605e3088db rw rootfstype=btrfs verbose loglevel=3 verbose amdgpu.ppfeaturemask=0xfff7ffff drm.edid_firmware=DP-2:edid/AOC_Q24G2_CVT-RB-OVERRIDE.bin
[  +0.000000] Kernel command line: BOOT_IMAGE=/vmlinuz-linux-zen root=UUID=a2e22b4d-7352-445f-b6ad-74605e3088db rw rootfstype=btrfs verbose loglevel=3 verbose amdgpu.ppfeaturemask=0xfff7ffff drm.edid_firmware=DP-2:edid/AOC_Q24G2_CVT-RB-OVERRIDE.bin
[  +0.009015] [drm] amdgpu kernel modesetting enabled.
[  +0.000189] amdgpu: Virtual CRAT table created for CPU
[  +0.000171] amdgpu: Topology: Add CPU node
[  +0.000091] amdgpu: Overdrive is enabled, please disable it before reporting any bugs unrelated to overdrive.
[  +0.000884] amdgpu 0000:4e:00.0: enabling device (0006 -> 0007)
[  +0.005449] amdgpu 0000:4e:00.0: amdgpu: detected ip block number 0 <soc21_common>
[  +0.000005] amdgpu 0000:4e:00.0: amdgpu: detected ip block number 1 <gmc_v11_0>
[  +0.000002] amdgpu 0000:4e:00.0: amdgpu: detected ip block number 2 <ih_v6_0>
[  +0.000002] amdgpu 0000:4e:00.0: amdgpu: detected ip block number 3 <psp>
[  +0.000002] amdgpu 0000:4e:00.0: amdgpu: detected ip block number 4 <smu>
[  +0.000003] amdgpu 0000:4e:00.0: amdgpu: detected ip block number 5 <dm>
[  +0.000002] amdgpu 0000:4e:00.0: amdgpu: detected ip block number 6 <gfx_v11_0>
[  +0.000002] amdgpu 0000:4e:00.0: amdgpu: detected ip block number 7 <sdma_v6_0>
[  +0.000002] amdgpu 0000:4e:00.0: amdgpu: detected ip block number 8 <vcn_v4_0>
[  +0.000002] amdgpu 0000:4e:00.0: amdgpu: detected ip block number 9 <jpeg_v4_0>
[  +0.000002] amdgpu 0000:4e:00.0: amdgpu: detected ip block number 10 <mes_v11_0>
[  +0.000034] amdgpu 0000:4e:00.0: amdgpu: Fetched VBIOS from VFCT
[  +0.000003] amdgpu: ATOM BIOS: 113-V520GAME-0OC
[  +0.010764] amdgpu 0000:4e:00.0: amdgpu: CP RS64 enable
[  +0.017947] amdgpu 0000:4e:00.0: vgaarb: deactivate vga console
[  +0.000004] amdgpu 0000:4e:00.0: amdgpu: Trusted Memory Zone (TMZ) feature not supported
[  +0.000033] amdgpu 0000:4e:00.0: amdgpu: MEM ECC is not presented.
[  +0.000002] amdgpu 0000:4e:00.0: amdgpu: SRAM ECC is not presented.
[  +0.000009] amdgpu 0000:4e:00.0: amdgpu: VRAM: 24560M 0x0000008000000000 - 0x00000085FEFFFFFF (24560M used)
[  +0.000002] amdgpu 0000:4e:00.0: amdgpu: GART: 512M 0x00007FFF00000000 - 0x00007FFF1FFFFFFF
[  +0.000174] [drm] amdgpu: 24560M of VRAM memory ready
[  +0.000002] [drm] amdgpu: 64377M of GTT memory ready.
[  +0.000216] amdgpu 0000:4e:00.0: amdgpu: Found VCN firmware Version ENC: 1.24 DEC: 9 VEP: 0 Revision: 11
[  +0.000069] amdgpu 0000:4e:00.0: amdgpu: Found VCN firmware Version ENC: 1.24 DEC: 9 VEP: 0 Revision: 11
[  +0.071844] amdgpu 0000:4e:00.0: amdgpu: reserve 0x1300000 from 0x85fc000000 for PSP TMR
[  +0.147467] amdgpu 0000:4e:00.0: amdgpu: RAP: optional rap ta ucode is not available
[  +0.000003] amdgpu 0000:4e:00.0: amdgpu: SECUREDISPLAY: securedisplay ta ucode is not available
[  +0.000033] amdgpu 0000:4e:00.0: amdgpu: smu driver if version = 0x0000003d, smu fw if version = 0x00000040, smu fw program = 0, smu fw version = 0x004e8100 (78.129.0)
[  +0.000004] amdgpu 0000:4e:00.0: amdgpu: SMU driver if version not matched
[  +0.188570] amdgpu 0000:4e:00.0: amdgpu: SMU is initialized successfully!
[  +0.072250] snd_hda_intel 0000:4e:00.1: bound 0000:4e:00.0 (ops amdgpu_dm_audio_component_bind_ops [amdgpu])
[  +0.333162] amdgpu: HMM registered 24560MB device memory
[  +0.001910] kfd kfd: amdgpu: Allocated 3969056 bytes on gart
[  +0.000017] kfd kfd: amdgpu: Total number of KFD nodes to be created: 1
[  +0.000101] amdgpu: Virtual CRAT table created for GPU
[  +0.000773] amdgpu: Topology: Add dGPU node [0x744c:0x1002]
[  +0.000003] kfd kfd: amdgpu: added device 1002:744c
[  +0.000014] amdgpu 0000:4e:00.0: amdgpu: SE 6, SH per SE 2, CU per SH 8, active_cu_number 96
[  +0.000006] amdgpu 0000:4e:00.0: amdgpu: ring gfx_0.0.0 uses VM inv eng 0 on hub 0
[  +0.000002] amdgpu 0000:4e:00.0: amdgpu: ring comp_1.0.0 uses VM inv eng 1 on hub 0
[  +0.000002] amdgpu 0000:4e:00.0: amdgpu: ring comp_1.1.0 uses VM inv eng 4 on hub 0
[  +0.000002] amdgpu 0000:4e:00.0: amdgpu: ring comp_1.2.0 uses VM inv eng 6 on hub 0
[  +0.000002] amdgpu 0000:4e:00.0: amdgpu: ring comp_1.3.0 uses VM inv eng 7 on hub 0
[  +0.000002] amdgpu 0000:4e:00.0: amdgpu: ring comp_1.0.1 uses VM inv eng 8 on hub 0
[  +0.000002] amdgpu 0000:4e:00.0: amdgpu: ring comp_1.1.1 uses VM inv eng 9 on hub 0
[  +0.000002] amdgpu 0000:4e:00.0: amdgpu: ring comp_1.2.1 uses VM inv eng 10 on hub 0
[  +0.000002] amdgpu 0000:4e:00.0: amdgpu: ring comp_1.3.1 uses VM inv eng 11 on hub 0
[  +0.000002] amdgpu 0000:4e:00.0: amdgpu: ring sdma0 uses VM inv eng 12 on hub 0
[  +0.000002] amdgpu 0000:4e:00.0: amdgpu: ring sdma1 uses VM inv eng 13 on hub 0
[  +0.000001] amdgpu 0000:4e:00.0: amdgpu: ring vcn_unified_0 uses VM inv eng 0 on hub 8
[  +0.000002] amdgpu 0000:4e:00.0: amdgpu: ring vcn_unified_1 uses VM inv eng 1 on hub 8
[  +0.000002] amdgpu 0000:4e:00.0: amdgpu: ring jpeg_dec uses VM inv eng 4 on hub 8
[  +0.000002] amdgpu 0000:4e:00.0: amdgpu: ring mes_kiq_3.1.0 uses VM inv eng 14 on hub 0
[  +0.001726] amdgpu 0000:4e:00.0: amdgpu: Using BACO for runtime pm
[  +0.000631] amdgpu 0000:4e:00.0: [drm] Registered 4 planes with drm panic
[  +0.000002] [drm] Initialized amdgpu 3.63.0 for 0000:4e:00.0 on minor 1
[  +0.013264] fbcon: amdgpudrmfb (fb0) is primary device
[  +0.017825] amdgpu 0000:4e:00.0: [drm] fb0: amdgpudrmfb frame buffer device

numcatl --hardware

available: 8 nodes (0-7)
node 0 cpus: 0 1 2 24 25 26
node 0 size: 0 MB
node 0 free: 0 MB
node 1 cpus: 3 4 5 27 28 29
node 1 size: 0 MB
node 1 free: 0 MB
node 2 cpus: 6 7 8 30 31 32
node 2 size: 32067 MB
node 2 free: 18757 MB
node 3 cpus: 9 10 11 33 34 35
node 3 size: 32253 MB
node 3 free: 20465 MB
node 4 cpus: 12 13 14 36 37 38
node 4 size: 32205 MB
node 4 free: 27353 MB
node 5 cpus: 15 16 17 39 40 41
node 5 size: 32229 MB
node 5 free: 26198 MB
node 6 cpus: 18 19 20 42 43 44
node 6 size: 0 MB
node 6 free: 0 MB
node 7 cpus: 21 22 23 45 46 47
node 7 size: 0 MB
node 7 free: 0 MB
node distances:
node     0    1    2    3    4    5    6    7 
   0:   10   11   12   12   12   12   12   12 
   1:   11   10   12   12   12   12   12   12 
   2:   12   12   10   11   12   12   12   12 
   3:   12   12   11   10   12   12   12   12 
   4:   12   12   12   12   10   11   12   12 
   5:   12   12   12   12   11   10   12   12 
   6:   12   12   12   12   12   12   10   11 
   7:   12   12   12   12   12   12   11   10

glxinfo -B #mesa-utils

name of display: :1
display: :1  screen: 0
direct rendering: Yes
Extended renderer info (GLX_MESA_query_renderer):
    Vendor: AMD (0x1002)
    Device: AMD Radeon RX 7900 XTX (radeonsi, navi31, LLVM 20.1.6, DRM 3.63, 6.15.4-zen2-1-zen) (0x744c)
    Version: 25.1.4
    Accelerated: yes
    Video memory: 24576MB
    Unified memory: no
    Preferred profile: core (0x1)
    Max core profile version: 4.6
    Max compat profile version: 4.6
    Max GLES1 profile version: 1.1
    Max GLES[23] profile version: 3.2
Memory info (GL_ATI_meminfo):
    VBO free memory - total: 19635 MB, largest block: 19635 MB
    VBO free aux. memory - total: 64096 MB, largest block: 64096 MB
    Texture free memory - total: 19635 MB, largest block: 19635 MB
    Texture free aux. memory - total: 64096 MB, largest block: 64096 MB
    Renderbuffer free memory - total: 19635 MB, largest block: 19635 MB
    Renderbuffer free aux. memory - total: 64096 MB, largest block: 64096 MB
Memory info (GL_NVX_gpu_memory_info):
    Dedicated video memory: 24576 MB
    Total available memory: 88953 MB
    Currently available dedicated video memory: 19635 MB
OpenGL vendor string: AMD
OpenGL renderer string: AMD Radeon RX 7900 XTX (radeonsi, navi31, LLVM 20.1.6, DRM 3.63, 6.15.4-zen2-1-zen)
OpenGL core profile version string: 4.6 (Core Profile) Mesa 25.1.4-arch1.2
OpenGL core profile shading language version string: 4.60
OpenGL core profile context flags: (none)
OpenGL core profile profile mask: core profile

OpenGL version string: 4.6 (Compatibility Profile) Mesa 25.1.4-arch1.2
OpenGL shading language version string: 4.60
OpenGL context flags: (none)
OpenGL profile mask: compatibility profile

OpenGL ES profile version string: OpenGL ES 3.2 Mesa 25.1.4-arch1.2
OpenGL ES profile shading language version string: OpenGL ES GLSL ES 3.20

vulkaninfo --summary #vulkan-tools

Vulkan Instance Version: 1.4.313


Instance Extensions: count = 20
-------------------------------
VK_EXT_acquire_xlib_display            : extension revision 1
VK_EXT_debug_report                    : extension revision 10
VK_EXT_debug_utils                     : extension revision 2
VK_EXT_direct_mode_display             : extension revision 1
VK_EXT_display_surface_counter         : extension revision 1
VK_EXT_swapchain_colorspace            : extension revision 5
VK_KHR_device_group_creation           : extension revision 1
VK_KHR_display                         : extension revision 23
VK_KHR_external_fence_capabilities     : extension revision 1
VK_KHR_external_memory_capabilities    : extension revision 1
VK_KHR_external_semaphore_capabilities : extension revision 1
VK_KHR_get_display_properties2         : extension revision 1
VK_KHR_get_physical_device_properties2 : extension revision 2
VK_KHR_get_surface_capabilities2       : extension revision 1
VK_KHR_portability_enumeration         : extension revision 1
VK_KHR_surface                         : extension revision 25
VK_KHR_wayland_surface                 : extension revision 6
VK_KHR_xcb_surface                     : extension revision 6
VK_KHR_xlib_surface                    : extension revision 6
VK_LUNARG_direct_driver_loading        : extension revision 1

Instance Layers: count = 8
--------------------------
VK_LAYER_AMD_switchable_graphics_32 AMD switchable graphics layer                  1.4.313  version 1
VK_LAYER_AMD_switchable_graphics_64 AMD switchable graphics layer                  1.4.313  version 1
VK_LAYER_FROG_gamescope_wsi_x86_64  Gamescope WSI (XWayland Bypass) Layer (x86_64) 1.3.221  version 1
VK_LAYER_MANGOHUD_overlay_x86_64    Vulkan Hud Overlay                             1.3.0    version 1
VK_LAYER_VALVE_steam_fossilize_32   Steam Pipeline Caching Layer                   1.3.207  version 1
VK_LAYER_VALVE_steam_fossilize_64   Steam Pipeline Caching Layer                   1.3.207  version 1
VK_LAYER_VALVE_steam_overlay_32     Steam Overlay Layer                            1.3.207  version 1
VK_LAYER_VALVE_steam_overlay_64     Steam Overlay Layer                            1.3.207  version 1

Devices:
========
GPU0:
        apiVersion         = 1.4.313
        driverVersion      = 2.0.349
        vendorID           = 0x1002
        deviceID           = 0x744c
        deviceType         = PHYSICAL_DEVICE_TYPE_DISCRETE_GPU
        deviceName         = AMD Radeon RX 7900 XTX
        driverID           = DRIVER_ID_AMD_OPEN_SOURCE
        driverName         = AMD open-source driver
        driverInfo         = 2025.Q2.1 (LLPC)
        conformanceVersion = 1.4.1.3


Feel free to ask for any further information that might be useful since I clearly got no clue where the problem might lie.

Offline

#2 2025-07-04 21:33:02

seth
Member
Registered: 2012-09-03
Posts: 65,700

Re: Trouble getting the expected performance out of an AMD GPU

How's the performance w/ https://wiki.archlinux.org/title/Steam/ … _emulation ?
Also is this 165Hz or https://wiki.archlinux.org/title/Variable_refresh_rate and can you disable the latter in the monitor OSD?

Plasma 6.4.2 (wayland)

also certainly test the performance on X11 (ideally uncomposited, SHIFT+alt+F12 toggles it though I guess steam will likely unredirect automagically, it's better to know wink

How to upload text · How to boot w/o GUI · Disable Windows Fast-Start! · Fix your xinitrc

Offline

#3 2025-07-04 21:47:08

Loukl
Member
Registered: 2025-07-04
Posts: 7

Re: Trouble getting the expected performance out of an AMD GPU

seth wrote:

-> everything runs native, no Wine/Proton in use. Tried it anyways but did not do anything just as expected.

seth wrote:

Also is this 165Hz or https://wiki.archlinux.org/title/Variable_refresh_rate and can you disable the latter in the monitor OSD?

-> It's a Q24G2A/BK with Adaptive-Sync set to always in KDE. However I did reach 180 FPS in CS2 for a brief moment and mangohud reports 5k-6k FPS in vkcube so I believe it is not limiting FPS. Furthermore I did my tests in full screen and windowed modes.

I will get back to you after testing under X11

Offline

#4 2025-07-04 22:17:19

Loukl
Member
Registered: 2025-07-04
Posts: 7

Re: Trouble getting the expected performance out of an AMD GPU

Tested with X11 (i3) and had the same results under amdvlk.
vulkan-radeon did a little better on X11 than wayland, but only reached close to the performance I already had with amdvlk.

Offline

#5 2025-07-04 22:28:32

seth
Member
Registered: 2012-09-03
Posts: 65,700

Re: Trouble getting the expected performance out of an AMD GPU

https://bbs.archlinux.org/viewtopic.php?id=305440
https://bbs.archlinux.org/viewtopic.php?id=306409
https://bbs.archlinux.org/viewtopic.php … 4#p2248394

Try to disable VRR. In the monitor. Just a hunch.

How to upload text · How to boot w/o GUI · Disable Windows Fast-Start! · Fix your xinitrc

Offline

#6 2025-07-04 22:38:03

Loukl
Member
Registered: 2025-07-04
Posts: 7

Re: Trouble getting the expected performance out of an AMD GPU

Disabling VRR, G-Sync and Adaptive Sync in the Monitor OSD seems to do nothing.

Offline

#7 2025-07-04 22:48:00

seth
Member
Registered: 2012-09-03
Posts: 65,700

Re: Trouble getting the expected performance out of an AMD GPU

Have you tried to enforce a higher power profile or performance level, https://wiki.archlinux.org/title/AMDGPU … nce_levels ?
There's a bug in 6.15, https://bbs.archlinux.org/viewtopic.php … 6#p2249146 - seems a bit very selective in your case - but it's a bug.

How to upload text · How to boot w/o GUI · Disable Windows Fast-Start! · Fix your xinitrc

Offline

#8 2025-07-04 22:56:17

Loukl
Member
Registered: 2025-07-04
Posts: 7

Re: Trouble getting the expected performance out of an AMD GPU

Yes, I did try enforcing different Performance Levels and even setting the clocks manually but it was to no use. Either it did not apply correctly or it did apply and raise clocks but did not affect FPS or power draw. Power Limit is raised to 378W and in Superposition the GPU does reach that high.

Offline

#9 2025-07-05 07:08:00

seth
Member
Registered: 2012-09-03
Posts: 65,700

Re: Trouble getting the expected performance out of an AMD GPU

Ok, let's get away from CS2 (might be game specific problem) and stuff like vkcube that just bores the shit out of your GPU
How do any of https://aur.archlinux.org/packages?O=0&K=unigine perform?

How to upload text · How to boot w/o GUI · Disable Windows Fast-Start! · Fix your xinitrc

Offline

#10 2025-07-05 16:52:51

Loukl
Member
Registered: 2025-07-04
Posts: 7

Re: Trouble getting the expected performance out of an AMD GPU

I did some Testing:

superposition, heaven and valley perform as expected (GPU clocks close to max and draws about 380W)

vulkan-radeon

[vertex] device-local=true: FPS: 12726 FrameTime: 0.079 ms
[vertex] device-local=false: FPS: 12880 FrameTime: 0.078 ms
[texture] anisotropy=0: FPS: 12660 FrameTime: 0.079 ms
[texture] anisotropy=16: FPS: 12619 FrameTime: 0.079 ms
[shading] shading=gouraud: FPS: 13030 FrameTime: 0.077 ms
[shading] shading=blinn-phong-inf: FPS: 12900 FrameTime: 0.078 ms
[shading] shading=phong: FPS: 12877 FrameTime: 0.078 ms
[shading] shading=cel: FPS: 12902 FrameTime: 0.078 ms
[effect2d] kernel=edge: FPS: 12969 FrameTime: 0.077 ms
[effect2d] kernel=blur: FPS: 13043 FrameTime: 0.077 ms
[desktop] <default>: FPS: 11802 FrameTime: 0.085 ms
[cube] <default>: FPS: 13083 FrameTime: 0.076 ms
[clear] <default>: FPS: 11819 FrameTime: 0.085 ms
=======================================================
                                   vkmark Score: 12716
=======================================================

vulkan-radeon + mangohud

[vertex] device-local=true: FPS: 6003 FrameTime: 0.167 ms
[vertex] device-local=false: FPS: 5107 FrameTime: 0.196 ms
[texture] anisotropy=0: FPS: 5099 FrameTime: 0.196 ms
[texture] anisotropy=16: FPS: 5112 FrameTime: 0.196 ms
[shading] shading=gouraud: FPS: 5123 FrameTime: 0.195 ms
[shading] shading=blinn-phong-inf: FPS: 5142 FrameTime: 0.194 ms
[shading] shading=phong: FPS: 5130 FrameTime: 0.195 ms
[shading] shading=cel: FPS: 5122 FrameTime: 0.195 ms
[effect2d] kernel=edge: FPS: 5139 FrameTime: 0.195 ms
[effect2d] kernel=blur: FPS: 5149 FrameTime: 0.194 ms
[desktop] <default>: FPS: 4925 FrameTime: 0.203 ms
[cube] <default>: FPS: 5152 FrameTime: 0.194 ms
[clear] <default>: FPS: 4909 FrameTime: 0.204 ms
=======================================================
                                   vkmark Score: 5162
=======================================================

amdvlk

[vertex] device-local=true: FPS: 10033 FrameTime: 0.100 ms
[vertex] device-local=false: FPS: 9127 FrameTime: 0.110 ms
[texture] anisotropy=0: FPS: 10169 FrameTime: 0.098 ms
[texture] anisotropy=16: FPS: 9991 FrameTime: 0.100 ms
[shading] shading=gouraud: FPS: 9745 FrameTime: 0.103 ms
[shading] shading=blinn-phong-inf: FPS: 9724 FrameTime: 0.103 ms
[shading] shading=phong: FPS: 9970 FrameTime: 0.100 ms
[shading] shading=cel: FPS: 9841 FrameTime: 0.102 ms
[effect2d] kernel=edge: FPS: 9673 FrameTime: 0.103 ms
[effect2d] kernel=blur: FPS: 9146 FrameTime: 0.109 ms
[desktop] <default>: FPS: 9891 FrameTime: 0.101 ms
[cube] <default>: FPS: 9791 FrameTime: 0.102 ms
[clear] <default>: FPS: 18775 FrameTime: 0.053 ms
=======================================================
                                   vkmark Score: 10452
=======================================================

amdvlk + mangohud

[vertex] device-local=true: FPS: 6644 FrameTime: 0.151 ms
[vertex] device-local=false: FPS: 6337 FrameTime: 0.158 ms
[texture] anisotropy=0: FPS: 6428 FrameTime: 0.156 ms
[texture] anisotropy=16: FPS: 6428 FrameTime: 0.156 ms
[shading] shading=gouraud: FPS: 6440 FrameTime: 0.155 ms
[shading] shading=blinn-phong-inf: FPS: 6443 FrameTime: 0.155 ms
[shading] shading=phong: FPS: 6425 FrameTime: 0.156 ms
[shading] shading=cel: FPS: 6466 FrameTime: 0.155 ms
[effect2d] kernel=edge: FPS: 6432 FrameTime: 0.155 ms
[effect2d] kernel=blur: FPS: 6441 FrameTime: 0.155 ms
[desktop] <default>: FPS: 6447 FrameTime: 0.155 ms
[cube] <default>: FPS: 6419 FrameTime: 0.156 ms
[clear] <default>: FPS: 5997 FrameTime: 0.167 ms
=======================================================
                                   vkmark Score: 6411
=======================================================


Some Games I tested:
Helldivers 2 - 60%, 1.3GHz, 110W, 90FPS (limited at 90), Proton experimental 10.0-20250630 and 80%, 1.3GHz, 120W 120FPS (game limits to 144)
Sea of Thieves - 65%, 1.3GHz, 110W, 50-140FPS depending on if I look at something or just the sky, Proton experimental 10.0-20250630
Splitgate - 75%, 1.3GHz, 120W, ~300FPS up to 360FPS (Game limits to 360)
The Witcher 3: Wild Hunt - 80%, 2GHz, 180W, 144FPS (Game limits to 144), Proton experimental 10.0-20250630
SUPERHOT - 80%, 1.3GHz, 130W, 300FPS (Ingame), 80%, 2.7GHz, 250W 980FPS (in Menu)
Palworld - 92%, 2.7GHz, 320W, 115FPS GE-Proton9-10-2-glebf6827 (unlimited), 80%, 2.3GHz, 200W, 90FPS (limited at 90)
Hitman - 50%, 1.25GHz, 100W, 90FPS wine-8.0-2754
Satisfactory - DX12Mode, 80%, 2.8GHz, 250W, 250FPS Proton GE 10-8

However basemark(aur) shows some interesting results:

official Test Vulkan1.0: 27189, avg. 272, min. 101, max. 673, 2.7GHz, ~380W
official Test opengl4.5: 15586, avg. 156, min. 70, max. 397, initially on 2.6GHz and 250W but drops to 1.6GHz / 1.3GHz and <150W after the first scene

custom(1280x720 vulkan) avg. 482, min. 63, max. 762, Score 48160
custom(3840x2160 vulkan) avg. 272, min. 109, max. 633, Score 27241
-> big difference between 720p and 4K

custom (1280x720 opengl) avg. 166, min. 98, max. 429, Score 16583
custom (3840x2160 opengl) avg. 155, min. 62, max. 426, Score 15544
-> No big difference between 720p and 4K

Tested on another setup with 2060 super:
custom (1280x720 opengl) avg. 168, min. 87, max. 408, Score 16836
custom (3840x2160 opengl) avg. 73, min. 34, max. 319, Score 7335
-> big difference between 720p and 4K

custom(1280x720 vulkan) avg. 289, min. 80, max. 797, Score 28915
custom(3840x2160 vulkan) avg. 78, min. 24, max. 688, Score 7757
-> big difference between 720p and 4K

So while vulkan behaves as expected, opengl only behaves as expected on the 2060 machine, not mine. opengl being less performant than vulkan was to be expected, but the noticeable difference with vulkan between 720p and 4K and missing one with opengl between 720p and 4K makes me believe that this is not related to cpu bottleneck.

On further note even tho it might not be related: audio in CS2 breaks randomly sometimes.
wireplumber-0.5.10-1
pipewire-1:1.4.6-1

Offline

#11 2025-07-05 20:43:15

seth
Member
Registered: 2012-09-03
Posts: 65,700

Re: Trouble getting the expected performance out of an AMD GPU

basemark might be capped, in doubt vync? (The max scores between both GPUs are pretty *very* much the same, whereas on vulkan the 7900 runs circles around the 2900)
Does anything change if you "export vblank_mode=0"?

But right now, the basmark oddity aside, CS2 (while I understand its your focus) remains an isolated problem? (as the GPU has no problems running at full capacity for the unigine demos)

Edit: 

pacman -Qs pulse

Last edited by seth (2025-07-05 20:43:44)

How to upload text · How to boot w/o GUI · Disable Windows Fast-Start! · Fix your xinitrc

Offline

#12 2025-07-05 23:47:22

Loukl
Member
Registered: 2025-07-04
Posts: 7

Re: Trouble getting the expected performance out of an AMD GPU

export vblank_mode=0 seems to have no effect. Same goes for moving the benchmark window to a 60Hz V-Sync screen.

output of "pacman -Qs pulse":

local/gtk-vnc 1.5.0-1
    VNC viewer widget for GTK
local/lib32-libcanberra 1:0.30+r2+gc0620e4-3
    A small and lightweight implementation of the XDG Sound Theme Specification (32-bit)
local/lib32-libpulse 17.0+r43+g3e2bb8a1e-1
    A featureful, general-purpose sound server (32-bit client libraries)
local/libcanberra 1:0.30+r2+gc0620e4-4
    A small and lightweight implementation of the XDG Sound Theme Specification
local/libpulse 17.0+r43+g3e2bb8a1e-1
    A featureful, general-purpose sound server (client library)
local/pipewire-pulse 1:1.4.6-1
    Low-latency audio/video router and processor - PulseAudio replacement
local/plasma-pa 6.4.2-1 (plasma)
    Plasma applet for audio volume management using PulseAudio
local/pulseaudio-qt 1.7.0-1
    Qt bindings for libpulse
local/qemu-audio-pa 10.0.0-7
    QEMU PulseAudio audio driver


I also did some tests with 1/2 and 1/4 CPU clocks via

cpupower frequency-set -u 1GHz

While helldivers FPS scaled proportionally, cs2 went from ~140 to ~70 to ~20 so not exactly linear but not too weird. However Sea of Thieves went from 80 to 40 to 40 making me doubt everything.
I'd give up and call it a CPU bottleneck if the other system running a 2060 super and a Ryzen 7 3700X (so pretty close to the TR 3960X not considering core count) wouldn't outperform my main setup by roughly +20% in some cases.

Offline

#13 2025-07-06 06:34:27

seth
Member
Registered: 2012-09-03
Posts: 65,700

Re: Trouble getting the expected performance out of an AMD GPU

local/pipewire-pulse 1:1.4.6-1

Audio isn't a conflict w/ pulseaudio.

Do you use the exactly same output configuration w/ both setups (same monitors attached the same way?)

And just for the lolz:

everything runs native, no Wine/Proton in use. Tried it anyways but did not do anything just as expected.

How does the wine/proton version of CS2 perform?

How to upload text · How to boot w/o GUI · Disable Windows Fast-Start! · Fix your xinitrc

Offline

Pages: 1

Board footer

Powered by FluxBB