PDA

View Full Version : [ubuntu] nouveau-firmware increases performance?



ventrical
December 21st, 2012, 02:37 PM
According to gtkperf my nVidia card runs the gtkperf tests 12 seconds faster with the nouveau-firmware graphics driver installed as oppossed to the nvidia-current.

See:

http://ubuntuforums.org/showpost.php?p=12415465&postcount=41

sgage
December 21st, 2012, 04:24 PM
According to gtkperf my nVidia card runs the gtkperf tests 12 seconds faster with the nouveau-firmware graphics driver installed as oppossed to the nvidia-current.

See:

http://ubuntuforums.org/showpost.php?p=12415465&postcount=41

Very interesting. I might just run my own test when I get a chance. My problems with nouveau aren't performance as such, though.

1) Nouveau seems to run hotter than nvidia-current on my machine.

2) I can not resume from suspend with nouveau, while nvidia-current works perfectly.

#1 has improved over the past year, but #2 is a showstopper for me, and I wish they'd fix it.

mc4man
December 21st, 2012, 04:31 PM
Very interesting. I might just run my own test when I get a chance. My problems with nouveau aren't performance as such, though.

1) Nouveau seems to run hotter than nvidia-current on my machine.

It may run a bit hotter as nouveau doesn't support 'powermizer' so your card is always at max performance (if your card uses powermizer under nvidia drivers

ventrical
December 21st, 2012, 04:34 PM
2 seconds to suspend, 3 seconds to unsuspend!! Nice stuff. Only problem is the network(eth0) will not enable. Thats a showstopper.

sgage
December 22nd, 2012, 03:03 PM
I did some checking with gtkperf.

With the nouveau drivers, I averaged around 8.4 seconds. Adding in the nouveau-firmware package took about 0.3 seconds off this time.

With the nvidia-current drivers, I averaged very marginally faster, around 7.8 seconds.

Still no resume from suspend with nouveau, so back to nvidia-current.

What were your actual scores with the two drivers? Are we running different versions of gtkperf? Our numbers seem out of whack...

Oh, my card is an Nvidia GeForce 8500 GT.

ventrical
December 22nd, 2012, 04:46 PM
gtkperf = version 0.40+ds-2

01:00.0 VGA compatible controller: NVIDIA Corporation GT218 [GeForce G210] (rev a2)

ventrical
December 22nd, 2012, 04:56 PM
I did some checking with gtkperf.

With the nouveau drivers, I averaged around 8.4 seconds. Adding in the nouveau-firmware package took about 0.3 seconds off this time.

With the nvidia-current drivers, I averaged very marginally faster, around 7.8 seconds.

Still no resume from suspend with nouveau, so back to nvidia-current.

What were your actual scores with the two drivers? Are we running different versions of gtkperf? Our numbers seem out of whack...

Oh, my card is an Nvidia GeForce 8500 GT.

I do not get network with either nouveau or current after suspend. Examining what mac4man had contributed and if gtkperf can be considered any type of benchmarking app then mac4mans suggested method would produce faster test results. I cannot validate the accuracy of that program (gtkperf), but , assuming it is not useless then I find the results convincing that there is an increase in Unity performance overall with the newer kernels .

However, it is one thing to notice a visual increase in speed and performance of Unity and it is another thing to authenticate those declarations. Running compares with older kernels and newer ones proves out that Unity *has* has had increase in performance according to gtkperf.

regards..

sgage
December 22nd, 2012, 07:19 PM
I do not get network with either nouveau or current after suspend. Examining what mac4man had contributed and if gtkperf can be considered any type of benchmarking app then mac4mans suggested method would produce faster test results. I cannot validate the accuracy of that program (gtkperf), but , assuming it is not useless then I find the results convincing that there is an increase in Unity performance overall with the newer kernels .

However, it is one thing to notice a visual increase in speed and performance of Unity and it is another thing to authenticate those declarations. Running compares with older kernels and newer ones proves out that Unity *has* has had increase in performance according to gtkperf.

regards..

I'm using the same gtkperf version as you. You said that your nouveau score was 12 seconds less than your n-c score, yet both of my scores were down around 8. Were you running the 'All Tests' with a 100 test rounds?

Incidentally, I get similar results in Gnome Shell or Unity.

With nouveau, I get the symptom of no networking on resume from suspend, plus no response to the mouse clicks. The cursor moves, but nothing happens when I click.

ventrical
December 22nd, 2012, 10:48 PM
Perhaps this link may help clarify what I was assuming.

http://ubuntuforums.org/showpost.php?p=12379429&postcount=34

Also .. allow me to clarify myself. When I mention performance in (Unity) Ubuntu-desktop, I am referring to the 'response' times when a Unity launcher icon is clicked or when the Dash Home>apps icons appear on the dashboard and the opening and closing of the Unity interactive GUI all together.

Part of this process depends on the hardware-motherboard, part depends on graphical adaptor card and, of course, another fraction of this depends on the kernel version and software drivers.

So as I am plodding along in this beta testing environ I had noticed visually that Unity response times were more 'snappy' than usual. I had simply set out to see if this may do with with the current version install of the kernel and noticed that the 3.7.rcs actually made Unity 'appear' faster. User_figaude suggested 'gtkperf' and so I tried that across compares of older kernels and the newer RCs. In my testing on two particular machines (one with GeForce 210/218 and one Intel graphics) gtkperf had shown that there are actual faster response time with the newer kernels.

However, when I tried the nouveau-firware drivers (for nvidia) it appeared yet even snappier . mac4man suggested to set Power mizer to Preferred maximum performance which resulted (according to gtkperf) that nvidia-current was even faster than the results presented by gtkperf using the nouveau-firmware nvidia drivers.

Adopting mac4mans test method and according to gtkperf results, nouveau-firmware does not appear to increase performance. I am not saying that this is a wrong assumption on my part because I had originally proposed this in the form of a question and until I find some other methods of testing then I would have to safely assume that the answer to my original question is no.

regards..

seenthelite
December 25th, 2012, 07:08 AM
VGA compatible controller GT216 [GeForce GT 230M]

Would not suspend with nouveau, so using nvidia-current.

ronacc
December 25th, 2012, 02:08 PM
8.26 seconds here with 3.8rc1 nvidia 304.41 driver and nvidia 7600gs card .

seenthelite
December 25th, 2012, 09:01 PM
GtkEntry - time: 0.02
GtkComboBox - time: 1.35
GtkComboBoxEntry - time: 0.48
GtkSpinButton - time: 0.08
GtkProgressBar - time: 0.08
GtkToggleButton - time: 0.32
GtkCheckButton - time: 0.08
GtkRadioButton - time: 0.13
GtkTextView - Add text - time: 0.66
GtkTextView - Scroll - time: 0.05
GtkDrawingArea - Lines - time: 0.69
GtkDrawingArea - Circles - time: 0.97
GtkDrawingArea - Text - time: 0.36
GtkDrawingArea - Pixbufs - time: 0.08

Total time: 5.38