I noticed my F@H PPD halved on my 5700XT by turning on >4G on my setup using amdgpu-pro 20.30-1109583.
Request: Smart Access Memory information
Collapse
X
-
vega56 BAR=8192M working, but not my old hd 7950.
hd 7950:
$ AMD_DEBUG=info glxinfo | grep vram
vram_size = 3072 MB
vram_vis_size = 256 MB
vram_type = 5
vram_bit_width = 384
has_dedicated_vram = 1
all_vram_visible = 0
# dmesg | grep BAR
[ 0.565360] pci 0000:0a:00.0: BAR 0: assigned to efifb
[ 4.215309] amdgpu 0000:0a:00.0: amdgpu: Fetched VBIOS from ROM BAR
[ 4.216290] [drm] Detected VRAM RAM=3072M, BAR=256M
b550 aorus pro:
above 4G decoding enabled and re-size bar support auto (only have auto, disabled)
Comment
-
-
Originally posted by marek View PostThe performance can decrease with Above 4G Decoding due to unknown reasons because this is all experimental and might not have been validated by hardware vendors. Some users report decreased performance.
I just tried it out with the superpostion benchmark and could not see a big difference for my system.
I currently run a RX 6900XT on a Ryzen 5800x below Arch 5.9.14 with current gnome/wayland/xwayland using the opensourse/mesa drivers. I cannot see a real difference between above 4g decoding on or off. But I think this benchmark might not be best suited to find out.
Code:$ dmesg | grep BAR [ 0.672903] pci 0000:27:00.0: VF(n) BAR2 space: [mem 0x00000000-0x0006ffff 64bit] (contains BAR2 for 7 VFs) [ 0.672920] pci 0000:27:00.0: VF(n) BAR4 space: [mem 0x00000000-0x0001bfff 64bit] (contains BAR4 for 7 VFs) [ 0.678504] pci 0000:2f:00.0: BAR 0: assigned to efifb [ 0.701208] pci 0000:27:00.0: BAR 9: assigned [mem 0xfc820000-0xfc88ffff 64bit] [ 0.701215] pci 0000:27:00.0: BAR 11: assigned [mem 0xfc890000-0xfc8abfff 64bit] [ 1.573731] [drm] Detected VRAM RAM=16368M, BAR=16384M
Code:$ AMD_DEBUG=info glxinfo | grep vram vram_size = 16384 MB vram_vis_size = 16368 MB vram_type = 9 vram_bit_width = 256 has_dedicated_vram = 1
Comment
-
-
Originally posted by agd5f View Post
Try disabling CSM or making sure you are using UEFI rather than legacy mode.(Finally find a chance to reboot my working machine
)
Code:AMD_DEBUG=info glxinfo | grep vram vram_size = 8192 MB vram_vis_size = 8176 MB vram_type = 9 vram_bit_width = 128 has_dedicated_vram = 1
Comment
-
-
@HerrLange
For found lower performance maybe vblank_mode=0 glxgears could help (even its not benchmark thing, result are changing by on/off 4g), 4g off=32k on=22k (ryzen 5600x (4.6 ghz) and vega 56).
Edit:
with new motherboard bios (ryzen 5600x (4.6 ghz) and vega 56) result:
4g on or off + resizebar disabled, give bar=256m and glxgears 29k
4g on + resizebar on, give bar=8192m, 26k
Before bios and kernel 5.10.x it was 10k difference 32k/22k, Now ony 3k with new mb bios and kernel 5.11-rc1.
Edit 2:
This sh.. b550 motherboard are clearly not ready to use.
I can have BAR=8192M with only 4g on resizebar disabled, early this day it was not (both csm off).
I wanted to test in game, i figured out a problem, maybe its a bios problem, with 4g: on/resizebar: on i have half fps on csgo, full fps with 4g: on/resizebar:disabled.
bios 4g on resizebar on csm off:
# dmesg | grep BAR
[ 0.609557] pci 0000:0c:00.0: BAR 0: assigned to efifb
[ 4.499440] [drm] Detected VRAM RAM=8176M, BAR=8192M
csgo: around 330 fps
Gpu sound very quiet
gpu stats ingame kernel 5.9 debian Or 5.11-rc1:
GFX Clocks and Power:
167 MHz (MCLK)
832 MHz (SCLK)
1269 MHz (PSTATE_SCLK)
700 MHz (PSTATE_MCLK)
800 mV (VDDGFX)
56.0 W (average GPU)
GPU Temperature: 46 C
GPU Load: 84 %
gpu stats ingame kernel 5.9 debian Or 5.11-rc1:
GFX Clocks and Power:
167 MHz (MCLK)
607 MHz (SCLK)
1269 MHz (PSTATE_SCLK)
700 MHz (PSTATE_MCLK)
800 mV (VDDGFX)
29.0 W (average GPU)
GPU Temperature: 42 C
GPU Load: 81 %
-------------------------------------------------------------------------------
bios 4g on resizebar disabled csm off:
# dmesg | grep BAR
[ 0.586413] pci 0000:0c:00.0: BAR 0: assigned to efifb
[ 5.387805] amdgpu 0000:0c:00.0: BAR 2: releasing [mem 0xfff0000000-0xfff01fffff 64bit pref]
[ 5.387808] amdgpu 0000:0c:00.0: BAR 0: releasing [mem 0xffe0000000-0xffefffffff 64bit pref]
[ 5.387829] pcieport 0000:0b:00.0: BAR 15: releasing [mem 0xffe0000000-0xfff01fffff 64bit pref]
[ 5.387832] pcieport 0000:0a:00.0: BAR 15: releasing [mem 0xffe0000000-0xfff01fffff 64bit pref]
[ 5.387834] pcieport 0000:00:03.1: BAR 15: releasing [mem 0xffe0000000-0xfff01fffff 64bit pref]
[ 5.387843] pcieport 0000:00:03.1: BAR 15: assigned [mem 0x900000000-0xbffffffff 64bit pref]
[ 5.387846] pcieport 0000:0a:00.0: BAR 15: assigned [mem 0x900000000-0xbffffffff 64bit pref]
[ 5.387849] pcieport 0000:0b:00.0: BAR 15: assigned [mem 0x900000000-0xbffffffff 64bit pref]
[ 5.387852] amdgpu 0000:0c:00.0: BAR 0: assigned [mem 0xa00000000-0xbffffffff 64bit pref]
[ 5.387863] amdgpu 0000:0c:00.0: BAR 2: assigned [mem 0x900000000-0x9001fffff 64bit pref]
[ 5.387944] [drm] Detected VRAM RAM=8176M, BAR=8192M
csgo: around 660 fps
Gpu sound heavy
gpu stats ingame kernel 5.11-rc1:
GFX Clocks and Power:
800 MHz (MCLK)
1154 MHz (SCLK)
1269 MHz (PSTATE_SCLK)
700 MHz (PSTATE_MCLK)
843 mV (VDDGFX)
92.0 W (average GPU)
GPU Temperature: 52 C
GPU Load: 71 %Last edited by kripteks; 28 December 2020, 05:04 PM.
Comment
-
-
Originally posted by marek View PostI have Ryzen 3900X but the chipset is only X370 and I can't find the option to enable SAM in the BIOS. I guess it depends on the motherboard vendor. (mine is Asus)
- Updated AMD AGESA ComboAm4v2PI 1.1.0.0 Patch D
- Support Re-size BAR function to enhance GPU performance.
after enabling just the resize BAR option my output is
vram_size = 8192 MB
vram_vis_size = 256 MB
vram_type = 5
vram_bit_width = 512
has_dedicated_vram = 1
after enabling both "Resize BAR" and "Above 4G memory/Crypto" I get
vram_size = 8192 MB
vram_vis_size = 8192 MB
vram_type = 5
vram_bit_width = 512
has_dedicated_vram = 1
running a benchmark in CSGO on de_dust2 I get very similar performance with each configuration on my Ryzen 3500X + R9 390 machine running Arch+KDE, leaning like 1% better for fully enabled (595-615 FPS @ 128-tick 1024x768 4xMSAA medium settings)
Comment
-
-
Originally posted by HerrLange View Post
Do you have some benchmarks showing this? Is this true for all situations or only for spcial cases?
If you get no visible performance improvement in some app, it's because SAM doesn't help in that specific app. However, you should never get a performance regression.
Comment
-
Comment