Kepler: Nouveau Is ~20% The Speed Of NVIDIA's Blob
Phoronix: Kepler: Nouveau Is ~20% The Speed Of NVIDIA's Blob
With the NVIDIA GeForce GTX 650 Linux review out of the way, I carried out some benchmarks comparing the performance of the open-source "Nouveau" driver to NVIDIA's official closed-source Linux graphics driver on this Kepler-based GK107 GPU. Similar to other NVIDIA GeForce 400/500/600 GPUs, if using the Nouveau driver as found by default on Ubuntu Linux and others, you can expect the performance to be less than 20% that of the official NVIDIA Linux driver...
Originally Posted by phoronix
Stay mad Nvididiots.
Why don't you underclock the card when it's on the binary blob so that we can compare the performance of the drivers in an equal setting?
He did that some time ago, and nouveau did quite well.
Originally Posted by boogerlad
Of course, bottlenecks and small delays tend to punish you more at higher clock speeds (larger proportion of each frame is spent waiting for the CPU), so the only real comparison would be to have them both running at full speed.
what the point?
Originally Posted by boogerlad
The point is to compare clock for clock perfomance, which was done last year, where Nouveau almost perfomed the same as the binary drivers.
Originally Posted by Andrecorreia
This almost provides an idea of how well Nouveau would perform if it was capable of reclocking, compared to the nVidia's drivers.
Except it doesn't.
Originally Posted by UraniumDeer
Because, as someone else already pointed out, most of the bottlenecks don't show up until you start rendering at high FPS speeds. Meaning underclocking automatically throws away 90% of the optimization work the binary drivers have worked on.
Which is the reason I didn't say that it 'almost provides an idea', and not 'it shows how they will compare once Nouveau reach the same clock as nVidia drivers'.
Originally Posted by smitty3268
I really do hope that nVidia pushes some documentation through on this soon, though they've stated that that's some of the hardest documentation for them to release.
Yeah these benchmarks are borderline useless w/o clock info.
From the log:
nouveau [ DRM] 3 available performance level(s)
nouveau [ DRM] 1: core 162MHz shader 324MHz memory 648MHz voltage 30mV
nouveau [ DRM] 2: core 405MHz shader 810MHz memory 1080MHz voltage 30mV
nouveau [ DRM] 3: core 1250MHz shader 2500MHz memory 1080MHz voltage 40mV
nouveau [ DRM] c:
nouveau [ DRM] MM: using COPY for buffer copies
Unfortunately doesn't say what it used, but it has to be either 1 or 2, which in the latter case would be a factor of 3, and putting nouveau at ~60% of the blob.
In fact it doesn't matter, and even I was criticizing Phoronix for many benchmarks, this time it doesn't matter. Because for an user makes little difference when the driver is so crippled. It may be good to know that it uses less watts and it has "just" 60% of the "native" binary closed driver, but in the meantime it makes most of things you will use, like Gnome Shell (or KDE4) and your game playing not practical. Isn't it so?
Originally Posted by not.sure