We've just released the beta version of Equalizer 1.2. This release has many new features, most notably:
- Automic local and remote configuration using the GPU-SD library
- Initial release of Sequel, a simplification and utility layer on top of Equalizer
- Runtime failure tolerance detecting hardware and software failures
- Tile compounds for fill-limited rendering such as direct volume rendering and interactive raytracing
And in the Collage network library:
- Distributed single-producer, multi-consumer queue
- RDMA-based connection class for InfiniBand (Linux only)
- Support push-based object distribution
Now is the time to extensively test this release and port your 1.0 applications. This version is source-code compatible with 1.0 for all official API calls. If you find any issues, please report them on this mailing list or on https://github.com/Eyescale/Equalizer/issues.
The full release notes are available on http://www.equalizergraphics.com/documents/RelNotes/RelNotes_1.1.7.html
Cheers,
Stefan.
_______________________________________________
eq-dev mailing list
eq-...@equalizergraphics.com
http://www.equalizergraphics.com/cgi-bin/mailman/listinfo/eq-dev
http://www.equalizergraphics.com
I have done some checks with the 1.2-beta package, Bino, and a
work-in-progress application. The applications build fine without
changes. Many thanks for the stable API :)
With Bino I see a performance problem that I did not see with Equalizer
1.0.1. On my home system (Ubuntu 11.10 amd64, NVIDIA 9600 with the
290.10 drivers) I only get up to 20 fps in Bino's benchmark mode. In
normal mode, that is too slow to keep up, so Bino drops all the frames.
With 1.0.1, I get > 150 fps in benchmark mode, and no frame drops occur
in normal mode.
I did not yet have time to investigate this issue in detail; I just did
a quick test with the attached simple configuration file and a 1920x540
(left-right stereo) video file.
Martin
On 8. Jan 2012, at 21:39, Martin Lambers wrote:
> With Bino I see a performance problem that I did not see with Equalizer
> 1.0.1. On my home system (Ubuntu 11.10 amd64, NVIDIA 9600 with the
> 290.10 drivers) I only get up to 20 fps in Bino's benchmark mode. In
> normal mode, that is too slow to keep up, so Bino drops all the frames.
> With 1.0.1, I get > 150 fps in benchmark mode, and no frame drops occur
> in normal mode.
This is weird, since afair nothing fundamental has changed which can cause this. Does this also happen with other applications? Do you see in the statistics where the time is spend?
Once you'll isolated it a bit further I can definitely look into it. It's definitely a show-stopper, and first I hear about it.
On Mon, 9 Jan 2012 13:31:42 +0100, Stefan Eilemann wrote:
> > With Bino I see a performance problem that I did not see with
> > Equalizer 1.0.1. On my home system (Ubuntu 11.10 amd64, NVIDIA 9600
> > with the 290.10 drivers) I only get up to 20 fps in Bino's
> > benchmark mode. In normal mode, that is too slow to keep up, so
> > Bino drops all the frames. With 1.0.1, I get > 150 fps in benchmark
> > mode, and no frame drops occur in normal mode.
>
> This is weird, since afair nothing fundamental has changed which can
> cause this. Does this also happen with other applications? Do you see
> in the statistics where the time is spend?
>
> Once you'll isolated it a bit further I can definitely look into it.
> It's definitely a show-stopper, and first I hear about it.
Finally I get around to trying git bisect :)
I started with Equalizer current HEAD (bad) and release-1.0.1 (good).
Git bisect tells me that 9ab87c0bac94d1962d21491b96cec9b7b3ad8c6c
makes the difference (log attached). Before this, I get > 70 FPS with
my current test video, after it I get only ~20 FPS.
But I must admit that I don't understand the impact of this commit and
why it causes the drop in frame rate. I only see this happen with Bino,
not with my other application. So maybe Bino does something wrong here?
Martin
Hmmm, I tried to use current HEAD with
9ab87c0bac94d1962d21491b96cec9b7b3ad8c6c reverted, but that did not
work - still ~20 FPS. Maybe I did something wrong when solving the
conflicts that 'git revert' caused. Or maybe the real cause is
something different...
Martin
FYI I moved this information to https://github.com/Eyescale/Equalizer/issues/81