Nice J2ME benchmark aimed at cpu characteristics a

[magic]

Just wanted you all know there's a new J2ME benchmark out there..

Analyzes cpu speed and characteristics. It also detects a bunch of extension apis (jsr120, jsr135, jsr82, Nokia UI Api, Vodafone Api etc). It's been a really good help for me when trying to optimize inner loops.

Here it is:

http://maciek.tv/benchpress

cheers

[ribot]

Very nice, and open source!

Submitted benchmark data from two of my handsets so far.

[magic]

Thank you!

The more devices tested the merrier! =)

I see now that I have to add a function that will let the user enter the name of the phone tested - some phones don't identify themselves correctly (or at all).

Some interesting conclusions can be drawn from the tested phones so far:

Nokia Series 40 phones have very different cpu speeds. I was thinking they all would be in the same league. Not so. For example: Nokia 6100 - 106 610ms and Nokia 6230 - 18 185ms. (Numbers are time to complete all tests, lower is better).

Division is sloow on all phones.

Exception catching can KILL performance on some phones, especially high-end ones.

Multiplication is usally just as fast as bitwise shifting.

If anyone has a phone that's not in the list - please do run the test. Together we could create a good cpu-speed database for most common phones.

cheers,
  Maciek

Games published by our own members! Check 'em out!

[Matzon]

I see now that I have to add a function that will let the user enter the name of the phone tested - some phones don't identify themselves correctly (or at all).

Indeed - I noticed that when I submitted a v500 - pretty slow phone - not that I notied it though... our games run fine on it.. *ponder*

[wooD]

I just ran it on two of my phones, a Samsung e317 on Cingular and got an io exception on the upload. Bummer.

The Sanyo 8100 on sprint completed and uploaded, but I don't see it in the results. Probably has to be approved on the server side?

Very cool app. Thanks.

[wooD]

Okay, the e317 worked the second time. What a dog for a brand new phone!  I do like the phone though and while my game is a little slower on it (and the j2me implementaion has serious bugs!), it's been very good as a phone so far.

Since I see these results immediatly i guess the other phone I ran didn't actually complete. I'll have to rerun it and see if it works the second time.

I think you need a retry loop on the networking code. I've found adding a retry on my networked apps seems to have fixed most of my issues. Annoying, but better than losing data.

[magic]

Sounds interesting, thanks for the feedback!

If I get the time this weekend I'm planning on doing the following:

* Add a dialog on the phone where user can enter phone model, in case phone doesn't identify itself.

* Retry network upload several times

* Add a feature on the site where users (=j2me developers) can add comments of several kinds (bug report, workaround, general comment etc).

/Maciek

[mlk]

An exit button at the end would be good.

[xdebugx]

If possible I would like to a see a feature that could tell you how many bytecodes per millisecond the devices run at.  It would help those who don't have the phones that others have tune the emulators to simulate those speeds.

[Abuse]

If possible I would like to a see a feature that could tell you how many bytecodes per millisecond the devices run at.  It would help those who don't have the phones that others have tune the emulators to simulate those speeds.

Such a statistic is meaningless
The slowest operation in any game is almost always rendering, and as such is done almost entirely in native code.

There is no simple metric that can be used to compare device speeds (and calibrate emulators to said speeds)

Even JBenchmarks extremely abstract metrics, are still horribly inaccurate on a great many devices.

Games published by our own members! Check 'em out!

[Linuxhippy]

The slowest operation in any game is almost always rendering, and as such is done almost entirely in native code.Even JBenchmarks extremely abstract metrics, are still horribly inaccurate on a great many devices.

Well, there are sooo many j2me benchmarks which only measure graphic-performance but e.g. I've developed an appliation which really depends on the cpu-performance and its nice to see some results for this kind of task!

lg Clemens

[munch]

Magic, the link is not available. Any other links? Really wanna see that benchmark. Could be quite usefful.

[Juriy]

Hmm, such a pity, I got nothing but "404 not found" message. Is this project dead? I would really like to test it with some of my phones! Could someone please post some version of this benchmark? Maybe you was lucky enough to download it before http://maciek.tv/benchpress is down?

[keldon85]

One thing about divides is that they tend not to be hardware supported in many portable platforms; even the Nintendo GBA had no hardware divides. The DS accesses division maths via a co-processor (it appears through the documentation)

Hmm, such a pity, I got nothing but "404 not found" message. Is this project dead

That's what I thought too, the only access to the site is with a password using his girlfriends name!