Announcement

Collapse
No announcement yet.

On Low-End GPUs, Nouveau Speeds Past The NVIDIA Driver

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • glisse
    replied
    Originally posted by Hephasteus View Post
    What's cpu bottleneck look like. Video cards are first order busmasters. All the cpu has to do is give it permission to take control of the bus. Granted a super slow cpu that is being banged up against the 100 percent usage barrier begins not responding to irq's. But game engines are all designed to keep the cpu and gpu out of that 100 percent usage barrier.
    Which makes a good point. Is the nouveau driver being as fast as the nvidia driver at that the same gpu usage point.
    CPU can be a bottleneck, GPU driver as to do a lot of stuff on the CPU and the way the driver is done and what the whole driver stack does, can have a huge impact on the CPU usage. Just compare the same GPU on a slow and fast CPU with any decent benchmark and you will see right away a difference that can in some case give you half the performance.

    Leave a comment:


  • Hephasteus
    replied
    Originally posted by Daekdroom View Post
    That's exactly what wouldn't happen. A low-end GPU on top of a high-end CPU would mostly likely - not - stop at the driver's CPU bottlenecks.
    What's cpu bottleneck look like. Video cards are first order busmasters. All the cpu has to do is give it permission to take control of the bus. Granted a super slow cpu that is being banged up against the 100 percent usage barrier begins not responding to irq's. But game engines are all designed to keep the cpu and gpu out of that 100 percent usage barrier.
    Which makes a good point. Is the nouveau driver being as fast as the nvidia driver at that the same gpu usage point.

    Leave a comment:


  • frej
    replied
    Poor benchmark

    The benchmark is clearly bandwidth limited, given the high resolutions for low end GPUS. It's likely that it doesn't tax the driver in any meaningful way.

    I'm sure it will perform worse in a lower resolution. Classic phoenix standard .

    Test the cards in a playable resolution, nobody will run in 1080p if they end up with 16fps.

    Leave a comment:


  • MuPuF
    replied
    Originally posted by davem View Post
    The main thing stopping me from switching to nouveau from the Nvidia blob is lack of VDPAU support. I run an old P4 and without GPU acceleration using VDPAU I can't watch 1080p movies. Is anyone aware of any work on going in nouveau to help change this? I might even be interested in contributing as much as I am able, if so.
    There is some Reverse Engineering done on it but seriously, do not expect VDPAU anytime soon.
    I would be looking forward to shader-based or openCL-based decoding if I were you, but I'm far from being competent on the matter.

    Leave a comment:


  • pingufunkybeat
    replied
    There is work on generic shader-based decoding going on, but it will be a while, and I don't know if anyone is working specifically on nvidia hardware.

    There is work for radeons, which runs XvMC already and is moving towards more challenging stuff, and once that is done, porting it over to nouveau should be relatively easy.

    The dedicated hardware decoder on the card will probably never work with open drivers. Unfortunately.

    Leave a comment:


  • davem
    replied
    The main thing stopping me from switching to nouveau from the Nvidia blob is lack of VDPAU support. I run an old P4 and without GPU acceleration using VDPAU I can't watch 1080p movies. Is anyone aware of any work on going in nouveau to help change this? I might even be interested in contributing as much as I am able, if so.

    Leave a comment:


  • not.sure
    replied
    Can't be repeated often enough:

    These benchmarks need to include the CPU usage!!!!!!

    Leave a comment:


  • crazycheese
    replied
    The reason there is advantage is dead simple. Obfuscation requires CPU time.
    But don't jump out of joy, amd loves proprietary same as nvidia.

    Leave a comment:


  • darkbasic
    replied
    Originally posted by iznogood View Post
    Is this true ? If yes then the test is useless.
    It's not the first time. That's why I DON'T use PTS to bench nexuiz.

    Originally posted by iznogood View Post
    Can you repeat it after disabling all not supported features ?
    No, they already told him the apache benchmark was completely useless without an external machine which generates requests but he never cared. They still publish USELESS apache benchmarks, he only cares about using PTS, it doesn't matter if the benchmark is a complete FAKE.

    Leave a comment:


  • bridgman
    replied
    Originally posted by smitty3268 View Post
    Can't this be done by using versioning?
    Yes, but then code for all the different versions has to be updated and tested, which significantly increases the workload and reduces the time available to spend on other improvements. Another nasty tradeoff.

    Leave a comment:

Working...
X