Exact build time is 3 mins 4.834 secs.
Curious. Which ARM CPU does the OMAP5 SoC use?
Cortex-A15 dual cores
No wonder the OMAP5 is so much more grunty.
Wonder what it would be like to create an OpenJDK 8 build using a lower end Beagleboard X15 (the one that is currently in the design phase).
Wonder what it would be like to create an OpenJDK 8 build using a lower end Beagleboard X15 (the one that is currently in the design phase).
The ARM performance of the AM572x will have a similar performance to the OMAP5. Both are Dual CortexA15 running at 1.5GHz. However, the DSP on the AM572x is faster than the DSP on the OMAP5 and the AM572x has two DSPs.
Regards,
John
Have uploaded a copy of the OpenJFX 1.8u40 runtime (GPL 2 license with class path exception). On the BBB Extract the lib directory from the 7z file and move the directory to the JRE installation directory.
Google Groups text editor has mysteriously decided to stop working. Here is the link to the OpenJFX runtime.
By default OpenJDK 8 uses the Zero JVM which doesn’t provide a high level of performance. Highly recommended to use the high performance JamVM JVM instead, which is comparable performance wise to the JVM used in Oracle’s Java SE Embedded 8. JamVM was originally designed to perform well on resource constrained embedded devices. Java 8 Lambda features can be used in a JavaFX program on an embedded device using JamVM, which the Oracle JVM doesn’t support (in the case of Java SE Embedded 8).
Do be aware with OpenJFX that JavaFX programs on the BBB will only run in X11 with DirectFB being specified as the renderer. DirectFB on the BBB with the current revision of OpenJFX that I was using had the bizarre GTK requirement, which prevents the framebuffer (no X11) from being used. Also a custom version of DirectFB must be built and installed. There is an ancient version of DirectFB in the Debian/Ubuntu repository but it doesn’t work with OpenJFX or JavaFX.
Some very useful info on changing JRE/JVMs with Debian/Ubuntu.
Basic overview of setting up a BBB running Ubuntu 14.04 to run a JavaFX program (rough draft):
- Add unofficial OpenJDK 8 PPA
- Update package list
- Install OpenJDK 8 JamVM package
- Make sure OpenJDK is installed by running java -version in a terminal (should see OpenJDK 1.8 being mentioned in the output)
- Build DirectFB 1.6
- Install DirectFB to /usr
- Extract OpenJFX 1.8u40 runtime file (7z) using p7zip and copy the lib directory to the OpenJDK 8 JRE directory
- Run Ensemble8 demo in X11 via LXDE to make sure that a JavaFX program will run properly:
java -Djavafx.platform=DirectFB -jamvm -jar /home/a_user/sample_dir/Ensemble8.jar
Would it be possible to have a JavaFX section in the BBB wiki on the ELinux website?
FYI OpenJFX 1.8u40 includes Dialog/Alert API. If you use the Oracle JDK then you will have to wait until around March 2015 to take advantage of the new API.
So over thanksgiving i let my quad wand do the same task:
omap5_uevm: (i was also building cloud9 during week, let's say 95% of
the time was openjdk)
Build needed 38:19:47, 6215244k disc space
wand quad: (i was out of state while this was running, so 100% core usage..)
Build needed 50:06:26, 6215460k disc space
The dual a15's come out to 30% faster then the quad a9's...
Regards,
I'm thinking debian might still have been using a mx53loco based box
for that 120hour run..
Regards,
Can't see the OpenJDK 8 build. I take it the build failed?
Have some fantastic news on the Ubuntu front. Various OpenJDK 8
packages via
an unofficial PPA are available for Ubuntu 12.04 (Precise) and 14.04
(Trusty).So it only took 38 hours 19 minutes.
So the Quad CortexA9 took 5 days (120 hours) and the OMAP5 took 38 hours
which is more then 3X performance. My bet is the build was running all
CPUs close to 100% for the entire build so this would be a good
comparison.So over thanksgiving i let my quad wand do the same task:
omap5_uevm: (i was also building cloud9 during week, let's say 95% of
the time was openjdk)
Build needed 38:19:47, 6215244k disc spacewand quad: (i was out of state while this was running, so 100% core
usage..)
Build needed 50:06:26, 6215460k disc spaceThe dual a15's come out to 30% faster then the quad a9's...
Interesting.
In this article, ARM says the CortexA15 is 2X performance compared to the
CortexA9, so 2xA15 = 4xA9; however, the OMAP5432 is running at 1.5MHz and
the Wandboard running at 1MHz, so given that, I would expect the OMAP5 to
be about 50% faster. Clearly Debian are running much slower hardware.
Regards,
John
Robert - What is the performance like with a Dual Core Cortex A9 compared to a Quad Core Cortex A9?
Robert - What is the performance like with a Dual Core Cortex A9 compared to a Quad Core Cortex A9?
With the quad wand, you also get 2gb of ram over the dual which only has 1gb… That alone is worth the price delta. For the price/performance it’s really hard to beat the wand quad.
Robert - thanks for the advise. Would it make sense economically to go for the Wandboard Quad over the upcoming Beagleboard X15 when it comes to running GUI programs?
3d: vivante will be reversed engineered before powervr. Freescale hasn’t released ‘armhf’ drivers whereas ti has…
The ‘x15’ will be supported by bb.org and ti… Who really supports the wand? All the info I put out is just because I replaced my failing pandas with quad wands…
Regards,