Amiga.org

Amiga computer related discussion => Amiga Hardware Issues and discussion => Topic started by: Piru on February 03, 2012, 08:46:28 PM

Title: MPlayer benchmarks
Post by: Piru on February 03, 2012, 08:46:28 PM
Yet Another Benchmark: This time decoding a H264 video clip with MPlayer (suggested by Fab (http://amigaworld.net/modules/newbb/viewtopic.php?mode=viewtopic&topic_id=35053&forum=33&start=240&viewmode=flat&order=0#650877)).

(http://sintonen.fi/pics/mplayer_benchmark.png)

As predicted X1000 PA6T fared better in this test.
Title: Re: MPlayer benchmarks
Post by: Karlos on February 03, 2012, 08:52:13 PM
Just an observation. Not sure if this was intentional but the options seem to imply this was a test of stream decode performance only, since the audio and video output options are both null.
Title: Re: MPlayer benchmarks
Post by: Piru on February 03, 2012, 08:55:17 PM
Quote from: Karlos;679025
Just an observation. Not sure if this was intentional but the options seem to imply this was a test of stream decode performance only, since the audio and video output options are both null.

The video is decoded with these options.
Title: Re: MPlayer benchmarks
Post by: jorkany on February 03, 2012, 09:15:54 PM
It's telling that all the focus is on the general processing benchmarks instead of "The Power of X" that the X1000 was supposed to deliver. If OS4 supported Xena, Xorro, etc. I expect we would be seeing a lot more posts devoted to that. Those #2 cores must be feeling awfully lonely about now!
Title: Re: MPlayer benchmarks
Post by: Karlos on February 03, 2012, 09:18:00 PM
Quote from: Piru;679027
The video is decoded with these options.


It is, but the frames are just discarded. Whenever I've used mplayer as a benchmark previously, it was always inclusive of rendering them to the display, which is far more important if you are trying to judge whether a system is capable of playing back the video or not (which is what I was doing, rather than basic benchmarking).

Early indications from the RageMem benchmarks suggested that there are problems with VRAM write speeds. At a guess, I'd assume these were driver related at this stage given the hardware is PCIe based. However, if anybody wanted to give the system a real drubbing, which I'm sure plenty of people do, there's your ideal opportunity. Quick, before anybody notices that and fixes it :lol:

Seriously though, right now, the PA6T is uncharted territory for existing software. It's a 64-bit processor running in 32-bit mode. To me, that raises a lot of questions. Are those old optimisations we're used to for the 32-bit machines still valid? Or are they a hindrance? In my personal experience, the PPC can be a huge pain in the arse when it comes to micro optimizing. For anybody unsure about this, Apple's PPC optimized memcpy() implementation is fascinating reading.

We're basing all out our conclusions so far on applications compiled for earlier PPC processors. I'd be interested to see what software actually optimized for PA6T (even in 32-bit mode) is capable of.
Title: Re: MPlayer benchmarks
Post by: Rob on February 03, 2012, 09:34:00 PM
@Piru

I ran the Mplayer and Lame tests on My 1Ghz XE and got 184 seconds and 30 seconds respectively.  The tests were done Ram: to Ram:.

You're welcome to add them to your graphs if you wish.
Title: Re: MPlayer benchmarks
Post by: Tuxedo on February 03, 2012, 10:24:49 PM
Quote from: Rob;679032
@Piru

I ran the Mplayer and Lame tests on My 1Ghz XE and got 184 seconds and 30 seconds respectively.  The tests were done Ram: to Ram:.

You're welcome to add them to your graphs if you wish.


mmm...
to me seems a bit too much time for that...
Have you done the tests at a "celan" start(i.e. without every commodities)?

You have to get tiomes near to Peg2@1000 imho...
Title: Re: MPlayer benchmarks
Post by: Rob on February 03, 2012, 10:41:46 PM
Quote from: Tuxedo;679044
mmm...
to me seems a bit too much time for that...
Have you done the tests at a "celan" start(i.e. without every commodities)?

You have to get tiomes near to Peg2@1000 imho...

Andrew Korn Ran the test on his Peg2 1/ghz and got 173 seconds under OS4.1.  Peg has a better northbridge and DDR ram.
Title: Re: MPlayer benchmarks
Post by: buzz on February 03, 2012, 11:03:15 PM
I was interested in how it compared to a couple of pcs here.

6-7 year old laptop with pentium-m @ 2ghz - 43.481s

desktop pc with i5-2500k single thread - 12.891s
desktop pc with i5-2500k threads=4 - 4.596s
desktop pc with i5-2500k threads=8 - 4.070s

of course ffmpeg libs have plenty of x86 optimisations.
Title: Re: MPlayer benchmarks
Post by: Fab on February 03, 2012, 11:11:44 PM
Quote from: Karlos;679025
Just an observation. Not sure if this was intentional but the options seem to imply this was a test of stream decode performance only, since the audio and video output options are both null.


The point was to test the CPU/Memory only, because i know the gfx subsystem on X1000 is still suboptimal and incomplete (no overlay output is available, which would mean only cgx_wpa output driver could be used).
Title: Re: MPlayer benchmarks
Post by: freqmax on February 03, 2012, 11:13:18 PM
Now FPGA synthesizis run is something I would like to see ;)
Title: Re: MPlayer benchmarks
Post by: Karlos on February 04, 2012, 12:07:45 AM
Quote from: Fab;679063
The point was to test the CPU/Memory only, because i know the gfx subsystem on X1000 is still suboptimal and incomplete (no overlay output is available, which would mean only cgx_wpa output driver could be used).


Fair enough.
Title: Re: MPlayer benchmarks
Post by: Miky060 on February 04, 2012, 12:59:28 PM
Quote from: Tuxedo;679044
mmm...
to me seems a bit too much time for that...
Have you done the tests at a "celan" start(i.e. without every commodities)?

You have to get tiomes near to Peg2@1000 imho...


Why did you expect similar results? Better hardware + better OS = better results.

Let's riassume..


  (http://img202.imagevenue.com/loc79/th_360006333_MPlayer_122_79lo.jpg) (http://img202.imagevenue.com/img.php?image=360006333_MPlayer_122_79lo.jpg)
Title: Re: MPlayer benchmarks
Post by: amigadave on February 04, 2012, 01:19:11 PM
Quote from: Piru;679023
Yet Another Benchmark: This time decoding a H264 video clip with MPlayer (suggested by Fab (http://amigaworld.net/modules/newbb/viewtopic.php?mode=viewtopic&topic_id=35053&forum=33&start=240&viewmode=flat&order=0#650877)).

(http://sintonen.fi/pics/mplayer_benchmark.png)

As predicted X1000 PA6T fared better in this test.

That is an encouraging test result, since the Peg2 is regarded as the fastest computer to run OS4.x on previous to the X1000 (unless there are some test results from the SAM460ex that I have not seen yet that show otherwise?), and the X1000 is more than twice as fast than the tested Peg2 in this test.

It is also faster than the 1.67GHz G4 PowerBook, which will shortly be the fastest non-overclocked, or third party accelerated, supported MorphOS hardware.  

I know the X1000 can't compete with modern x86 systems, but the results of testing so far are as good, or better than I had expected, as my expectations were realistic and not overly inflated.

I am satisfied that it appears A-Eon & Varisys have done a good job designing and building this system.
Title: Re: MPlayer benchmarks
Post by: klx300r on February 04, 2012, 03:33:09 PM
Quote from: amigadave;679125
...
I know the X1000 can't compete with modern x86 systems, but the results of testing so far are as good, or better than I had expected, as my expectations were realistic and not overly inflated.

I am satisfied that it appears A-Eon & Varisys have done a good job designing and building this system.


ah finally a voice of reason in all these benchmark threads:):drink:
Title: Re: MPlayer benchmarks
Post by: Fab on February 04, 2012, 04:17:54 PM
@amigadave

The thing is that in practice, it's more useful with output display, and in that case, the advantage of the X1000 might be shadowed by unoptimized video output drivers or slow bus speed (in CPU mode that is, in DMA mode, PCIe would show its real speed).