Arch Linux

TjaBBe

Member

Registered: 2014-03-25

Posts: 4

Poor bumblebee performance

Hey all,

Lately I have been getting pretty frustrated about a problem with bumblebee on my T440S with a GeForce GT 730M discrete graphics card which I just can't seem to find the sollution for.

When I start a tool to test the graphics adapter (eg. glxspheres64) using optirun or primusrun with vsync disabled everything seems fine. Both primus/optirun report that they are using the GeForce GT 730M/PCIe/SSE2 for rendering and the program starts running. However, the performance is far lower then when running the same program on the intel gfx card (with vsync disabled):

With the integrated intel card I get over 200 fps:

# vblank_mode=0 glxspheres64 
Polygons in scene: 62464
ATTENTION: default value of option vblank_mode overridden by environment.
ATTENTION: default value of option vblank_mode overridden by environment.
Visual ID of window: 0x20
Context is Direct
OpenGL Renderer: Mesa DRI Intel(R) Haswell Mobile 
252.882853 frames/sec - 250.700979 Mpixels/sec
206.765815 frames/sec - 204.981840 Mpixels/sec
205.554419 frames/sec - 203.780896 Mpixels/sec
210.329552 frames/sec - 208.514829 Mpixels/sec
206.154247 frames/sec - 204.375548 Mpixels/sec

While the discrete NVIDIA card only gets at 100 FPS:

# vblank_mode=0 primusrun glxspheres64 
ATTENTION: default value of option vblank_mode overridden by environment.
ATTENTION: default value of option vblank_mode overridden by environment.
Polygons in scene: 62464
ATTENTION: default value of option vblank_mode overridden by environment.
ATTENTION: default value of option vblank_mode overridden by environment.
Visual ID of window: 0x20
Context is Direct
OpenGL Renderer: GeForce GT 730M/PCIe/SSE2
ATTENTION: default value of option vblank_mode overridden by environment.
ATTENTION: default value of option vblank_mode overridden by environment.
103.070044 frames/sec - 102.180756 Mpixels/sec
103.300532 frames/sec - 102.409255 Mpixels/sec
103.345781 frames/sec - 102.454114 Mpixels/sec
102.641088 frames/sec - 101.755501 Mpixels/sec
103.338942 frames/sec - 102.447334 Mpixels/sec
103.389334 frames/sec - 102.497291 Mpixels/sec
102.776195 frames/sec - 101.889442 Mpixels/sec
103.280539 frames/sec - 102.389434 Mpixels/se

I see some stuff going on in the logging while running a program with optirun but as of yet I haven't be able to trace it back to something useful:

mrt 26 15:33:38 illidan kernel: bbswitch: enabling discrete graphics
mrt 26 15:33:39 illidan kernel: thinkpad_acpi: EC reports that Thermal Table has changed
mrt 26 15:33:39 illidan logger[1900]: ACPI group/action undefined: ibm/hotkey / LEN0068:00
mrt 26 15:33:39 illidan slim[358]: removing GPU device /sys/devices/pci0000:00/0000:00:1c.4/0000:04:00.0/drm/card1 /dev/dri/card1
mrt 26 15:33:39 illidan slim[358]: xf86: found device 1
mrt 26 15:33:39 illidan kernel: vgaarb: device changed decodes: PCI:0000:04:00.0,olddecodes=none,decodes=none:owns=none
mrt 26 15:33:39 illidan kernel: [drm] Initialized nvidia-drm 0.0.0 20130102 for 0000:04:00.0 on minor 1
mrt 26 15:33:39 illidan kernel: NVRM: loading NVIDIA UNIX x86_64 Kernel Module  334.21  Thu Feb 27 15:55:45 PST 2014
mrt 26 15:33:39 illidan acpid[303]: client connected from 1907[0:996]
mrt 26 15:33:39 illidan acpid[303]: 1 client rule loaded
mrt 26 15:33:39 illidan kernel: nvidia 0000:04:00.0: irq 66 for MSI/MSI-X
mrt 26 15:33:39 illidan kernel: ACPI Warning: \_SB_.PCI0.PEG_.VID_._DSM: Argument #4 type mismatch - Found [Buffer], ACPI requires [Package] (20131115/nsarguments-95)
mrt 26 15:33:39 illidan kernel: ACPI Warning: \_SB_.PCI0.PEG_.VID_._DSM: Argument #4 type mismatch - Found [Buffer], ACPI requires [Package] (20131115/nsarguments-95)
mrt 26 15:33:39 illidan kernel: ACPI Warning: \_SB_.PCI0.PEG_.VID_._DSM: Argument #4 type mismatch - Found [Buffer], ACPI requires [Package] (20131115/nsarguments-95)
mrt 26 15:33:39 illidan kernel: ACPI Warning: \_SB_.PCI0.PEG_.VID_._DSM: Argument #4 type mismatch - Found [Buffer], ACPI requires [Package] (20131115/nsarguments-95)
mrt 26 15:33:39 illidan kernel: ACPI Warning: \_SB_.PCI0.PEG_.VID_._DSM: Argument #4 type mismatch - Found [Buffer], ACPI requires [Package] (20131115/nsarguments-95)
mrt 26 15:33:39 illidan kernel: ACPI Warning: \_SB_.PCI0.PEG_.VID_._DSM: Argument #4 type mismatch - Found [Buffer], ACPI requires [Package] (20131115/nsarguments-95)
mrt 26 15:33:39 illidan kernel: ACPI Warning: \_SB_.PCI0.PEG_.VID_._DSM: Argument #4 type mismatch - Found [Buffer], ACPI requires [Package] (20131115/nsarguments-95)
mrt 26 15:33:39 illidan kernel: ACPI Warning: \_SB_.PCI0.PEG_.VID_._DSM: Argument #4 type mismatch - Found [Buffer], ACPI requires [Package] (20131115/nsarguments-95)
mrt 26 15:33:43 illidan kernel: ACPI Error: Field [TBF3] at 270336 exceeds Buffer [NULL] size 262144 (bits) (20131115/dsopcode-236)
mrt 26 15:33:43 illidan kernel: ACPI Error: Method parse/execution failed [\_SB_.PCI0.PEG_.VID_.GETB] (Node ffff88032cc81af0), AE_AML_BUFFER_LIMIT (20131115/psparse-536)
mrt 26 15:33:43 illidan kernel: ACPI Error: Method parse/execution failed [\_SB_.PCI0.PEG_.VID_._ROM] (Node ffff88032cc81ac8), AE_AML_BUFFER_LIMIT (20131115/psparse-536)
mrt 26 15:33:43 illidan kernel: ACPI Warning: \_SB_.PCI0.PEG_.VID_._DSM: Argument #4 type mismatch - Found [Buffer], ACPI requires [Package] (20131115/nsarguments-95)
mrt 26 15:33:43 illidan acpid[303]: client connected from 1907[0:996]
mrt 26 15:33:43 illidan acpid[303]: 1 client rule loaded
mrt 26 15:33:43 illidan kernel: thinkpad_acpi: asked for hotkey mask 0x0070ffbf, but firmware forced it to 0x0070ffbb
mrt 26 15:33:44 illidan bumblebeed[302]: [ 1003.375115] [WARN][XORG] (WW) Unresolved symbol: fbGetGCPrivateKey
mrt 26 15:33:44 illidan bumblebeed[302]: [ 1003.375132] [WARN][XORG] (WW) NVIDIA(0): Unable to get display device for DPI computation.
mrt 26 15:33:44 illidan bumblebeed[302]: [ 1003.375135] [ERROR][XORG] (EE) NVIDIA(0): Failed to initiate mode change.
mrt 26 15:33:44 illidan bumblebeed[302]: [ 1003.375138] [ERROR][XORG] (EE) NVIDIA(0): Failed to complete mode change

The last two lines seem to be of interest, but I haven't been able to find anything on it yet.

As you can see in the logging I'm using the nvidia driver, nouveau is uninstalled and blacklisted in /etc/modprobe.d.

Does anyone have a suggestion on where I can look to further troubleshoot this?

Thanks!

mrunion

Member

From: Jonesborough, TN

Registered: 2007-01-26

Posts: 1,938

Website

Re: Poor bumblebee performance

Primusrun will try to match your display framerate. For example, with

optirun glxspheres64

on my Acer w/ nVidia GT730M, I get about 2200-2500 fps. With primusrun I get 60.

---

Matt

"It is very difficult to educate the educated."

TjaBBe

Member

Registered: 2014-03-25

Posts: 4

Re: Poor bumblebee performance

I know, that's why I run "vblank_mode=0" in front for the test (ofcourse normally, vsync is not a bad thing) which disables the syncing. Without that option primusrun runs at 60fps here aswell.

To illustrate, optirun performs even worse (85 fps):

 $ optirun glxspheres64 
Polygons in scene: 62464
Visual ID of window: 0x20
Context is Direct
OpenGL Renderer: GeForce GT 730M/PCIe/SSE2
85.696671 frames/sec - 84.957280 Mpixels/sec
85.767265 frames/sec - 85.027265 Mpixels/sec
85.727955 frames/sec - 84.988294 Mpixels/sec

mrunion

Member

From: Jonesborough, TN

Registered: 2007-01-26

Posts: 1,938

Website

Re: Poor bumblebee performance

Sorry -- missed the vblank_mode completely!

I have no other ideas, then.

---

Matt

"It is very difficult to educate the educated."

kjw

Member

Registered: 2014-03-19

Posts: 4

Re: Poor bumblebee performance

Not sure what is going on but I have a T440s and I can confirm that I also see about half the frame rate using optirun / nvidia than with intel.

Weird thing is, here are my results for 32/64 bit glxspheres and intel / optirun (all with vblank_mode=0):

glxspheres64               200fps
optirun glxspheres64   80fps
glxspheres32               60fps
optirun glxspheres32   85fps

Why the huge difference between 32 and 64 bit I don't know.

Anyone find any performance difference between optirun and primusrun on the T440s?

ypnos

Member

Registered: 2007-12-30

Posts: 59

Re: Poor bumblebee performance

I can confirm the problem on my T440s. Here also primus doesn't work correctly. It is a bit slower than optirun and all GL displays are not drawn (either black or old content from GPU memory). Anybody have an idea about that problem?

tobias_

Member

Registered: 2012-09-01

Posts: 25

Re: Poor bumblebee performance

Maybe we should open a bumblebee ticket for this. I stumbled upon this thread because I have the same Issue on my T440s

Maybe it's related to this: http://forums.lenovo.com/t5/Linux-Discu … -p/1401235

I can imagine that something is put into a low power mode, such that the performance is so bad. Kernel also outputs

[37772.144931] nvidia 0000:04:00.0: irq 64 for MSI/MSI-X
[37772.151336] ACPI Warning: \_SB_.PCI0.PEG_.VID_._DSM: Argument #4 type mismatch - Found [Buffer], ACPI requires [Package] (20131115/nsarguments-95)
[37772.151419] ACPI Warning: \_SB_.PCI0.PEG_.VID_._DSM: Argument #4 type mismatch - Found [Buffer], ACPI requires [Package] (20131115/nsarguments-95)
[37772.151478] ACPI Warning: \_SB_.PCI0.PEG_.VID_._DSM: Argument #4 type mismatch - Found [Buffer], ACPI requires [Package] (20131115/nsarguments-95)
[37772.151537] ACPI Warning: \_SB_.PCI0.PEG_.VID_._DSM: Argument #4 type mismatch - Found [Buffer], ACPI requires [Package] (20131115/nsarguments-95)
[37772.151594] ACPI Warning: \_SB_.PCI0.PEG_.VID_._DSM: Argument #4 type mismatch - Found [Buffer], ACPI requires [Package] (20131115/nsarguments-95)
[37772.151806] ACPI Warning: \_SB_.PCI0.PEG_.VID_._DSM: Argument #4 type mismatch - Found [Buffer], ACPI requires [Package] (20131115/nsarguments-95)
[37772.151996] ACPI Warning: \_SB_.PCI0.PEG_.VID_._DSM: Argument #4 type mismatch - Found [Buffer], ACPI requires [Package] (20131115/nsarguments-95)
[37772.152055] ACPI Warning: \_SB_.PCI0.PEG_.VID_._DSM: Argument #4 type mismatch - Found [Buffer], ACPI requires [Package] (20131115/nsarguments-95)
[37776.292576] ACPI Error: Field [TBF3] at 270336 exceeds Buffer [NULL] size 262144 (bits) (20131115/dsopcode-236)
[37776.292582] ACPI Error: Method parse/execution failed [\_SB_.PCI0.PEG_.VID_.GETB] (Node ffff88032cb54af0), AE_AML_BUFFER_LIMIT (20131115/psparse-536)
[37776.292588] ACPI Error: Method parse/execution failed [\_SB_.PCI0.PEG_.VID_._ROM] (Node ffff88032cb54ac8), AE_AML_BUFFER_LIMIT (20131115/psparse-536)
[37776.299385] ACPI Warning: \_SB_.PCI0.PEG_.VID_._DSM: Argument #4 type mismatch - Found [Buffer], ACPI requires [Package] (20131115/nsarguments-95)

Has anyone tried running Xorg with nvidia driver directly? Maybe the performance issues are not because of bumblebee.

Last edited by tobias_ (2014-03-29 20:55:04)

Menabrea

Member

Registered: 2014-03-30

Posts: 16

Re: Poor bumblebee performance

I have the exact same problem with a 630M on a lenovo E530.
I've also sent an email with a bug report to bumblebee developers.

Last edited by Menabrea (2014-03-30 14:11:30)

TjaBBe

Member

Registered: 2014-03-25

Posts: 4

Re: Poor bumblebee performance

Not sure what is going on but I have a T440s and I can confirm that I also see about half the frame rate using optirun / nvidia than with intel.

Weird thing is, here are my results for 32/64 bit glxspheres and intel / optirun (all with vblank_mode=0):

glxspheres64               200fps
optirun glxspheres64   80fps
glxspheres32               60fps
optirun glxspheres32   85fps

Why the huge difference between 32 and 64 bit I don't know.

Anyone find any performance difference between optirun and primusrun on the T440s?

I didn't test with 32 bit versions but I did compare primusrun (without vsync) and optirun. While optirun also gets me at about 85 fps primusrun gives 105fps, which is still way too low.

Interesting to see that the problem seems to be related to the specific laptop model/type. I'll try and have another look at this tomorrow.

Menabrea

Member

Registered: 2014-03-30

Posts: 16

Re: Poor bumblebee performance

I mailed a bumblebee developer with a bug report, still waiting for a response.
Anyway on a TS E530 I7-3612QM nvidia 630M 2GB

vblank_mode=0  glxspheres64  60fps
optirun     glxspheres64           140fps
primusrun glxspheres64           60fps
optirun -c jpeg glxspheres64    170fps

Doesn't seems like using vblank_mode=0 on optirun or primusrun affects performance.

Last edited by Menabrea (2014-04-01 14:07:18)

JamesLens

Member

Registered: 2014-04-03

Posts: 17

Re: Poor bumblebee performance

Edit: No longer a problem, using nouveau drivers with PRIME instead of bumblebee.

Last edited by JamesLens (2014-04-06 16:32:27)

Menabrea

Member

Registered: 2014-03-30

Posts: 16

Re: Poor bumblebee performance

Edit: No longer a problem, using nouveau drivers with PRIME instead of bumblebee.

Thank you for your suggestion. I completley uninstalled primus, bumblebee bbswitch nvidia and nvidia-utils and installed nouveau drivers. I cannot loed nouveau with kms because if i do so udev freeze, and i cannot change udev with systemd hook because i use lvm2, which is not compatible.
However i load the nouveau module normalli, and in fact seems working:

01:00.0 VGA compatible controller: NVIDIA Corporation GF108M [GeForce GT 630M] (rev a1) (prog-if 00 [VGA controller])
	Subsystem: Lenovo Device 5001
	Flags: bus master, fast devsel, latency 0, IRQ 49
	Memory at e0000000 (32-bit, non-prefetchable) [size=16M]
	Memory at a0000000 (64-bit, prefetchable) [size=512M]
	Memory at c0000000 (64-bit, prefetchable) [size=32M]
	I/O ports at 5000 [size=128]
	Expansion ROM at e1000000 [disabled] [size=512K]
	Capabilities: [60] Power Management version 3
	Capabilities: [68] MSI: Enable+ Count=1/1 Maskable- 64bit+
	Capabilities: [78] Express Endpoint, MSI 00
	Capabilities: [b4] Vendor Specific Information: Len=14 <?>
	Capabilities: [100] Virtual Channel
	Capabilities: [128] Power Budgeting <?>
	Capabilities: [600] Vendor Specific Information: ID=0001 Rev=1 Len=024 <?>
	Kernel driver in use: nouveau
	Kernel modules: nouveau

But:

[root@box]# xrandr --listproviders
Providers: number : 1
Provider 0: id: 0x90 cap: 0xb, Source Output, Sink Output, Sink Offload crtcs: 4 outputs: 5 associated providers: 0 name:Intel

seems like xrandr can't find the dedicated graphic card. Any suggest?

Menabrea

Member

Registered: 2014-03-30

Posts: 16

Re: Poor bumblebee performance

Ok, i fixed the first problem just installing the 'modesetting' package.
Now:

[user@box ~]$ xrandr --listproviders
Providers: number : 2
Provider 0: id: 0x48 cap: 0xb, Source Output, Sink Output, Sink Offload crtcs: 4 outputs: 5 associated providers: 0 name:Intel
Provider 1: id: 0x107 cap: 0x2, Sink Output crtcs: 2 outputs: 1 associated providers: 0 name:modesetting

But when i try, following the wiki https://wiki.archlinux.org/index.php/PRIME :

[user@box ~]$ xrandr --setprovideroffloadsink 0x107 0x48
X Error of failed request:  BadValue (integer parameter out of range for operation)
  Major opcode of failed request:  140 (RANDR)
  Minor opcode of failed request:  34 ()
  Value in failed request:  0x107
  Serial number of failed request:  16
  Current serial number in output stream:  17

And i have now two other problems:
-The laptop does not anymore suspend when i close it
-The nvidia card is always so the battery last one third than before

JamesLens

Member

Registered: 2014-04-03

Posts: 17

Re: Poor bumblebee performance

Hey,

I'm not sure if installing modesetting was the correct thing to do, because for me the output of xrandr --listproviders was

Provider 0: id: 0x8e cap: 0xb, Source Output, Sink Output, Sink Offload crtcs: 3 outputs: 3 associated providers: 1 name:Intel
Provider 1: id: 0x63 cap: 0x7, Source Output, Sink Output, Source Offload crtcs: 2 outputs: 2 associated providers: 1 name:nouveau

where the name of the second provider is nouveau rather than modesetting.
The suggested thing in the wiki to do if only one provider appears is to "Delete/move /etc/X11/xorg.conf file and any other files relating to GPUs in /etc/X11/xorg.conf.d/", and also I know that you have to delete all possible nvidia related config files, otherwise its a mess. Have you tried that? From what I read nvidia automatically makes a xorg.conf file, but I never had to deal with it since I went straight for nouveau.
Also, nouveau needs to be started in boot, so in your /etc/mkinitcpio.conf you have to add nouveau to MODULES. Mine looks like MODULES="i915 nouveau", so the intel drivers are loaded too with i915 and the boot screen looks nicer. I think you must have done this though since the kernel uses the nouveau driver.
PRIME should automatically manage power, so the nvidia card won't always be on when properly installed.
Hope this helps!

Last edited by JamesLens (2014-04-09 00:48:35)

Menabrea

Member

Registered: 2014-03-30

Posts: 16

Re: Poor bumblebee performance

Hey,

I'm not sure if installing modesetting was the correct thing to do, because for me the output of xrandr --listproviders was

Provider 0: id: 0x8e cap: 0xb, Source Output, Sink Output, Sink Offload crtcs: 3 outputs: 3 associated providers: 1 name:Intel
Provider 1: id: 0x63 cap: 0x7, Source Output, Sink Output, Source Offload crtcs: 2 outputs: 2 associated providers: 1 name:nouveau

where the name of the second provider is nouveau rather than modesetting.
The suggested thing in the wiki to do if only one provider appears is to "Delete/move /etc/X11/xorg.conf file and any other files relating to GPUs in /etc/X11/xorg.conf.d/", and also I know that you have to delete all possible nvidia related config files, otherwise its a mess. Have you tried that? From what I read nvidia automatically makes a xorg.conf file, but I never had to deal with it since I went straight for nouveau.
Also, nouveau needs to be started in boot, so in your /etc/mkinitcpio.conf you have to add nouveau to MODULES. Mine looks like MODULES="i915 nouveau", so the intel drivers are loaded too with i915 and the boot screen looks nicer. I think you must have done this though since the kernel uses the nouveau driver.
PRIME should automatically manage power, so the nvidia card won't always be on when properly installed.
Hope this helps!

Thank you for your response.
Actually, before installing modesetting i've tried to delete all xorg configuration files, but it didn't worked. I've uninstalled with pacman -Rsdd all nvidia proprietary drivers so they should not conflict.
The problem is that i can't add 'nouveau to MODULES="" (actually i haven't added neither i915) because if i do so it fails to boot. I've read somewhere that 'nouveau' may conflict with udev hook and that the solution is to replace udev with systemd. But systemd goes in conflict with lvm2 hook and the machine fails to boot again. (i'm on a luks+lvm setup). So i'll try just for curiosity if nouveau driver with bumblebee get a better performance.

JamesLens

Member

Registered: 2014-04-03

Posts: 17

Re: Poor bumblebee performance

I'm sorry I don't think I can provide additional help, I'm also on a luks + lvm setup and my mkinitcpio has the udev hook which works just fine with the nouveau module. I tried using nouveau with bumblebee first and that's what gave me poor performance with the video card. When I contacted the bumblebee devs they said that you're supposed to use nouveau with PRIME and that bumblebee is not intended to be used with nouveau.
Best of luck, hope you find a solution.

Last edited by JamesLens (2014-04-10 15:51:36)

Menabrea

Member

Registered: 2014-03-30

Posts: 16

Re: Poor bumblebee performance

I'm sorry I don't think I can provide additional help, I'm also on a luks + lvm setup and my mkinitcpio has the udev hook which works just fine with the nouveau module. I tried using nouveau with bumblebee first and that's what gave me poor performance with the video card. When I contacted the bumblebee devs they said that you're supposed to use nouveau with PRIME and that bumblebee is not intended to be used with nouveau.
Best of luck, hope you find a solution.

Thanks, when i try to boot with udev and nouveau it freeze ad udev loading, but maybe it's my mistake, can you please provide me yuor 'hooks' and 'modules' array? Maybe i did something wrong (i really hope so)

JamesLens

Member

Registered: 2014-04-03

Posts: 17

Re: Poor bumblebee performance

Sure.

MODULES="i915 nouveau"

HOOKS="base udev autodetect modconf block keyboard encrypt lvm2 filesystems fsck"

and then run it with mkinitcpio -p linux.

Menabrea

Member

Registered: 2014-03-30

Posts: 16

Re: Poor bumblebee performance

So it's really the same of mine

HOOKS="base udev autodetect modconf block keyboard encrypt lvm2 filesystems fsck"

The only change i think it's that i modified the encrypt hook to unlock two disks with the same key, but it shouldn't make any difference or affect the boot process.

So i'll try adding both i915 and nouveau and see what happens.

Menabrea

Member

Registered: 2014-03-30

Posts: 16

Re: Poor bumblebee performance

So, i partially solved the issue adding both i915 and nouveau. Bcause probably it didn't worked because i wrote nouveau only.
The framerate problem seems to be resolved, thus just partially:

[user@box ~]$ DRI_PRIME=1 glxspheres64
Polygons in scene: 62464
Visual ID of window: 0x68
Context is Direct
OpenGL Renderer: Gallium 0.4 on NVC1
362.989918 frames/sec - 374.939546 Mpixels/sec
352.083158 frames/sec - 363.673735 Mpixels/sec
359.412124 frames/sec - 371.243971 Mpixels/sec
353.928991 frames/sec - 365.580334 Mpixels/sec
355.441993 frames/sec - 367.143143 Mpixels/sec
354.700928 frames/sec - 366.377682 Mpixels/sec
352.993520 frames/sec - 364.614066 Mpixels/sec
357.759975 frames/sec - 369.537434 Mpixels/sec

But here's what is shown:

So yes, while the console show a good framerate, glxspheres seems completly broken. And if i try benchmarking with unigine-heaven it is even more laggy than with nvidia and i get some frequent 'black interference' and rendering problems.

Also, i can play at Amnesia The Dark Descent (using steam) at 1024x768 with maximum quality, while playing at native resolution of 1600x900 it is a bit laggy, but still well playable with medium quality.

Last edited by Menabrea (2014-04-10 19:34:15)