Adobe Flash HD
This time I used the 2009 JHU Commencement highlights video, after clicking the little HD button. The size of the video on-screen was approximately 896x504. Here's a table again:
|Test||BW (Mbps)||CPU (%)||Quality|
|S10/Firefox 3/Flash 9||25||40||Poor|
So even with HD YouTube, the enhancements help a lot, especially in reducing bandwidth used. However, this time it couldn't match MPlayer's playback quality.
According to a post in this forum thread, Flash enhancements for UNIX are coming in a future release. Hopefully that's sooner rather than later.
Soft Client on Linux
It works with Wine on Linux. I couldn't test it with Wine on Solaris 10 or OpenSolaris, and the way I tested it on Linux meant that I couldn't check if sound worked, but otherwise it seemed to work pretty well.
New GUI Firmware Options
There are at least two new options in the Advanced menu; these are ones that I took note of because it's not obvious to me what they do, and the documentation only lists them but does not describe them:
- Enable Fast Download
- Video input disable
REAL BONUS - new X extension: XRender
I have to say REAL here since last time I said BONUS, it turned out that it was in the SRSS 4.2 What's New list, but not in the SRS 5 What's New list (see the edit that I made to yesterday's blog entry). This time, I'm much more sure it's not obvious from the wiki docs. The only place it's mentioned is in the man page for utxconfig (and the corresponding version included on wikis.sun.com).
By default, the XRender extension is not turned on. You can turn it on for your token by running "utxconfig -n on", then logging out and back in. I have no idea how XRender support will affect performance of apps like Firefox on Sun Ray (Cairo uses XRender), but what I do know is that KDE 4 should now be usable! Over a year ago, I compiled the base libraries and desktop for KDE 4.0.5 (with QT 4.3.3). I logged in to see how well it was working, and I was not impressed. The desktop was very ugly and the Logout button didn't even work! (There's a screenshot linked to from this forum thread, but the link seems to be dead now.) I abandoned ship and waited. Today I pulled out that archived set of packages and it works great! As it turns out, the logout button fails if XRender isn't present.
A KDE developer had previously suggested to me compiling QT 4.5 with -graphicssystem raster to work around the lack of XRender. I never got a chance to do this, so I don't know if that would perform better than letting it use the new XRender extension. With the QT and KDE that I have now, running over my desktop icons rapidly and continuously results in Xnewt usage of 50% CPU and plasma process usage of about 35% CPU, as well as a bandwidth stream of 10 Mbps. Contrast that to about 30% CPU usage of the Xorg process when I used KDE 4 from the VirtualBox console (plasma CPU usage was comparable). It's possible a newer QT and KDE, even with XRender, would do better because of optimizations made to the code since I compiled it. I don't know whether raster would do better or worse though. I do know that someone on the SunRay-Users list had tried QT-raster on Linux and found it unusable since QT made the assumption that the color mask on the X server was RGB, instead of querying and determining that Sun Rays actually use BGR. It's possible this issue would not appear on Solaris.
To get a more objective idea of how XRender actually performed since one of the Sun Ray engineers had mentioned on the mailing list that the feature would be kept in for release if it performed well, I downloaded render_bench and ran it against both the Sun Ray and the X console running in VirtualBox. The first time I ran it on the Sun Ray, the X server crashed or reset for an unidentified reason before the test completed. The second time, the test completed without incident. I found that off-screen XRender and Imlib2 tests were comparable between the Sun Ray and console, but that for on-screen XRender, the Sun Ray took 5-12 times longer to complete the test. Subjectively, though, it still looks pretty good.
This ends my formal testing of EA 1. I will make another post if I find another hidden new feature, or have an update on something else I've already blogged about. Otherwise, we wait for EA 2!