x264 progress since 2005-05-03

Audionut · 22nd September 2008, 05:39

I got bored and thought i'd see what progress has been made.

2005-05-03

Basically the first build with configure

[code]x264 -p1 -B 3000 -b3 -m5 --ref 3 --weightb --progress

fps 11.51 bitrate 2999.6 psnr avg 45.21 psnr global 44.10

Code:

x264 -p2 -B 3000 -b3 -m5 --ref 3 --weightb --progress

fps 12.69 bitrate 3007.0 psnr avg 45.99 psnr global 45.59 size 39,209,448 bytes

From here I grabbed a build every 12 months and used a slightly different command line.
2005-09-20 Snapshot

[code]x264 -p1 -B 3000 -b3 -m5 --ref 3 --weightb --8x8dct --threads 3 --progress

fps 15.49 bitrate 3087.1 psnr avg 45.96 psnr global 45.53

Code:

x264 -p2 -B 3000 -b3 -m5 --ref 3 --weightb --8x8dct --threads 3 --progress

fps 17.20 bitrate 3007.0 psnr avg 46.00 psnr global 45.70 size 39,264,011 bytes

2006-09-22 Snapshot

[code]x264 -p1 -B 3000 -b3 -m5 --ref 3 --weightb --8x8dct --threads 3 --progress

fps 20.11 bitrate 3065.41 psnr avg 46.031 psnr global 45.604 ssim 0.9866237

Code:

x264 -p2 -B 3000 -b3 -m5 --ref 3 --weightb --8x8dct --threads 3 --progress

fps 22.50 bitrate 3005.40 psnr avg 46.086 psnr global 45.801 ssim 0.9868645 size 39,251,104 bytes

2007-09-21 Snapshot

[code]x264 -p1 -B 3000 -b3 -m5 --ref 3 --weightb --8x8dct --threads 3 --progress

fps 30.62 bitrate 3062.99 psnr avg 46.067 psnr global 45.632 ssim 0.9867144

[code]x264 -p2 -B 3000 -b3 -m5 --ref 3 --weightb --8x8dct --threads 3 --progress

fps 33.19 bitrate 3006.82 psnr avg 46.129 psnr global 45.849 ssim 0.9869905 size 39,244,564 bytes

2008-09-22 Snapshot

[code]x264 -p1 -B 3000 -b3 -m5 --ref 3 --weightb --8x8dct --threads 3 --aq-mode 0 --progress

fps 39.82 bitrate 3057.87 psnr avg 46.068 psnr global 45.629 ssim 0.9866864

[code]x264 -p2 -B 3000 -b3 -m5 --ref 3 --weightb --8x8dct --threads 3 --aq-mode 0 --progress

fps 43.07 bitrate 3009.90 psnr avg 46.134 psnr global 45.854 ssim 0.9849855 size 39,284,796 bytes

Conclusion: Very nice. The last few years have brought an extra 10 fps each year with an increase in quality regarding metrics. I included files from the last two years as ssim has dropped a little this year.

I currently get about 10fps on my dual-core with my settings.
If I glaze through the looking glass at 12 months time, I see, a hardware upgrade, I've held off on quad-core for the core i7.

With that and the hopefully usual fps increase I should be hitting about 30fps. Woot.

Audionut · 22nd September 2008, 05:40

Then I thought I would try to enable all the fancy features as they became available.

2005-09-20

Code:

x264 -p1 -B 3000 -b3 --b-pyramid -r3 -A all -w --me esa -m6 -8 --threads 3 --frames 1000

Code:

fps 2.88 bitrate 3212.2 psnr avg 46.56 psnr global 46.17

2006-09-22

Code:

x264 -p1 -B 3000 -b3 --b-pyramid -r3 -A all --direct auto -w --b-rdo --me esa -m7 --mixed-refs --bime -8 -t2 --threads 3 --thread-input --frames 1000

Code:

fps 3.28 bitrate 3230.92 psnr avg 46.822 psnr global 46.455 ssim 0.9874579

2007-09-21

Code:

x264 -p1 -B 3000 -b3 --b-pyramid -r3 -A all --direct auto -w --b-rdo --me esa -m7 --mixed-refs --bime -8 -t2 --threads 3 --thread-input --frames 1000

Code:

fps 4.37 bitrate 3198.10 psnr avg 46.926 psnr global 46.541 ssim 0.9878487

2008-09-22

Code:

x264 -p1 -B 3000 -b3 --b-adapt 2 --b-pyramid -r3 -A all --direct auto -w --b-rdo --me tesa -m7 --psy-rd 1.0:1.0 --mixed-refs --bime -8 -t2 --threads 3 --thread-input --frames 1000

Code:

fps 5.52 bitrate 3020.05 psnr avg 44.977 psnr global 44.436 ssim 0.9841742

Conclusion: Slower settings yet more speed. I included this and last years files as with AQ and psy-rd you can't trust metrics.

Audionut · 22nd September 2008, 10:04

2005-09-20

Code:

x264 -p1 -B 3000 -b3 --b-pyramid -r3 -A all -w --me esa -m6 -8 --threads 3 --frames 1000

Code:

fps 2.88 bitrate 3212.2 psnr avg 46.56 psnr global 46.17

2008-09-22

Code:

x264 -p1 -B 3000 -b3 --b-pyramid -r3 -A all -w --me esa -m6 -8 --threads 3 --aq-mode 0 --psy-rd 0.0:0.0 --frames 1000

Code:

fps 14.89 bitrate 3189.88 psnr avg 46.657 psnr global 46.266

5.17 times faster. So in 12 months time that becomes 25.65fps. In three years that's 76.98fps.

Inventive Software · 22nd September 2008, 12:37

Development will slow exponentially though, so I reckon it's safe to assume a 2x increase.

Avenger007 · 24th September 2008, 17:26

Maybe it's a good time to have another Doom9 codec shoot-out.
The last one was in 2005 http://www.doom9.org/index.html?/codecs-final-105-1.htm

ZombiePimp · 24th September 2008, 21:54

Quote:

Originally Posted by Avenger007

Maybe it's a good time to have another Doom9 codec shoot-out.
The last one was in 2005 http://www.doom9.org/index.html?/codecs-final-105-1.htm

I agree!

click2 · 25th September 2008, 03:53

Quote:

Originally Posted by Avenger007

Maybe it's a good time to have another Doom9 codec shoot-out.
The last one was in 2005 http://www.doom9.org/index.html?/codecs-final-105-1.htm

There's Dark_Shikari's codec comparison from July 08 which was conducted once psyRDO was implemented(?). It compares x264, Elecard HD, and Nero Recode for H.264, as well as the same source with Xvid, DivX, VC-1, VP7, and HCEnc.

Dark Shikari · 25th September 2008, 04:06

Quote:

Originally Posted by click2

There's Dark_Shikari's codec comparison from July 08 which was conducted once psyRDO was implemented(?). It compares x264, Elecard HD, and Nero Recode for H.264, as well as the same source with Xvid, DivX, VC-1, VP7, and HCEnc.

Things have improved considerably since then, so I really should redo the test. And this time, I'd add a few others, like Ateme.

Avenger007 · 25th September 2008, 04:12

Quote:

Originally Posted by click2

There's Dark_Shikari's codec comparison from July 08 which was conducted once psyRDO was implemented(?). It compares x264, Elecard HD, and Nero Recode for H.264, as well as the same source with Xvid, DivX, VC-1, VP7, and HCEnc.

I'm not sure if you can really call that a shoot-out. I was hoping for something much more detailed and transparent; something along the lines of the Fourth Annual MSU MPEG-4 AVC/H.264 Video Codec Comparison.

The Doom9 2005 codec shoot-out has been referenced all over the web. A new shoot-out should be thorough, transparent and as detailed as possible.

Dark Shikari · 25th September 2008, 04:13

Quote:

Originally Posted by Avenger007

I'm not sure if you can really call that a shoot-out. I was hoping for something much more detailed and transparent; something along the lines of the Fourth Annual MSU MPEG-4 AVC/H.264 Video Codec Comparison.

The Doom9 2005 codec shoot-out has been referenced all over the web. A new shoot-out should be thorough, transparent and as detailed as possible.

Yes, I didn't intend for it to be Doom9-level quality; if I wanted to do that I most definitely could though, it would just take more work. The main annoyance is that I like having tons and tons of encoders on comparisons, not just two or three of the best, and that really increases the workload of doing a comparison.

Avenger007 · 25th September 2008, 04:23

Having a new Doom9 codec shoot-out would act as a new benchmark. So the new benchmark can be referenced when having "tons and tons of encoders on comparisons", which can always be appended separately.

IgorC · 25th September 2008, 04:39

Quote:

Originally Posted by Dark Shikari

Things have improved considerably since then, so I really should redo the test. And this time, I'd add a few others, like Ateme.

I think it will be more interesting (competitive) since there are only two H.264 encoders wich have good psy implementation: Ateme and x264.
What version of Ateme encoder do you have? I saw Ateme got a new generation encoder. Old one can't be on par with current psy implementation of x264 but still not bad.

Dark Shikari · 25th September 2008, 04:53

Quote:

Originally Posted by IgorC

I think it will be more interesting (competitive) since there are only two H.264 encoders wich have good psy implementation: Ateme and x264.
What version of Ateme encoder do you have? I saw Ateme got a new generation encoder. Old one can't be on par with current psy implementation of x264 but still not bad.

The version I have access to is the July 2007 core.

Overall, I'd say its pretty decent, but falls into a lot of the same traps as many of the other encoders. Its significantly better than Mainconcept though.

bob0r · 26th September 2008, 00:29

Fact remains, thousands of happy x264 users and more coming every day. The x264 developers are doing a great job, I thank them all!

komarovsky · 26th September 2008, 01:19

Quote:

Originally Posted by bob0r

Fact remains, thousands of happy x264 users and more coming every day. The x264 developers are doing a great job, I thank them all!

i do too, they're great

LoRd_MuldeR · 26th September 2008, 01:30

Quote:

Originally Posted by bob0r

Fact remains, thousands of happy x264 users and more coming every day. The x264 developers are doing a great job, I thank them all!

No doubt on that

A detailed codec shoot-out would be very interesting anyway. And maybe it could help to make x264 even more popular and/or reveal parts that still have room for improvement...

22nd September 2008, 05:39	#1 \| Link
Audionut Registered User Join Date: Nov 2003 Posts: 1,281	x264 progress since 2005-05-03 I got bored and thought i'd see what progress has been made. 2005-05-03 Basically the first build with configure [code]x264 -p1 -B 3000 -b3 -m5 --ref 3 --weightb --progress fps 11.51 bitrate 2999.6 psnr avg 45.21 psnr global 44.10 Code: x264 -p2 -B 3000 -b3 -m5 --ref 3 --weightb --progress fps 12.69 bitrate 3007.0 psnr avg 45.99 psnr global 45.59 size 39,209,448 bytes From here I grabbed a build every 12 months and used a slightly different command line. 2005-09-20 Snapshot [code]x264 -p1 -B 3000 -b3 -m5 --ref 3 --weightb --8x8dct --threads 3 --progress fps 15.49 bitrate 3087.1 psnr avg 45.96 psnr global 45.53 Code: x264 -p2 -B 3000 -b3 -m5 --ref 3 --weightb --8x8dct --threads 3 --progress fps 17.20 bitrate 3007.0 psnr avg 46.00 psnr global 45.70 size 39,264,011 bytes 2006-09-22 Snapshot [code]x264 -p1 -B 3000 -b3 -m5 --ref 3 --weightb --8x8dct --threads 3 --progress fps 20.11 bitrate 3065.41 psnr avg 46.031 psnr global 45.604 ssim 0.9866237 Code: x264 -p2 -B 3000 -b3 -m5 --ref 3 --weightb --8x8dct --threads 3 --progress fps 22.50 bitrate 3005.40 psnr avg 46.086 psnr global 45.801 ssim 0.9868645 size 39,251,104 bytes 2007-09-21 Snapshot [code]x264 -p1 -B 3000 -b3 -m5 --ref 3 --weightb --8x8dct --threads 3 --progress fps 30.62 bitrate 3062.99 psnr avg 46.067 psnr global 45.632 ssim 0.9867144 [code]x264 -p2 -B 3000 -b3 -m5 --ref 3 --weightb --8x8dct --threads 3 --progress fps 33.19 bitrate 3006.82 psnr avg 46.129 psnr global 45.849 ssim 0.9869905 size 39,244,564 bytes 2008-09-22 Snapshot [code]x264 -p1 -B 3000 -b3 -m5 --ref 3 --weightb --8x8dct --threads 3 --aq-mode 0 --progress fps 39.82 bitrate 3057.87 psnr avg 46.068 psnr global 45.629 ssim 0.9866864 [code]x264 -p2 -B 3000 -b3 -m5 --ref 3 --weightb --8x8dct --threads 3 --aq-mode 0 --progress fps 43.07 bitrate 3009.90 psnr avg 46.134 psnr global 45.854 ssim 0.9849855 size 39,284,796 bytes Conclusion: Very nice. The last few years have brought an extra 10 fps each year with an increase in quality regarding metrics. I included files from the last two years as ssim has dropped a little this year. I currently get about 10fps on my dual-core with my settings. If I glaze through the looking glass at 12 months time, I see, a hardware upgrade, I've held off on quad-core for the core i7. With that and the hopefully usual fps increase I should be hitting about 30fps. Woot. Last edited by Audionut; 23rd October 2011 at 17:35.

22nd September 2008, 12:37	#4 \| Link
Inventive Software Turkey Machine Join Date: Jan 2005 Location: Lowestoft, UK (but visit lots of places with bribes [beer]) Posts: 1,953	Development will slow exponentially though, so I reckon it's safe to assume a 2x increase. __________________ On Discworld it is clearly recognized that million-to-one chances happen 9 times out of 10. If the hero did not overcome huge odds, what would be the point? Terry Pratchett - The Science Of Discworld

24th September 2008, 17:26	#5 \| Link
Avenger007 Bruce Wayne Join Date: Dec 2007 Posts: 283	Maybe it's a good time to have another Doom9 codec shoot-out. The last one was in 2005 http://www.doom9.org/index.html?/codecs-final-105-1.htm

25th September 2008, 04:23	#11 \| Link
Avenger007 Bruce Wayne Join Date: Dec 2007 Posts: 283	Having a new Doom9 codec shoot-out would act as a new benchmark. So the new benchmark can be referenced when having "tons and tons of encoders on comparisons", which can always be appended separately.

26th September 2008, 00:29	#14 \| Link
bob0r Pain and suffering Join Date: Jul 2002 Posts: 1,337	Fact remains, thousands of happy x264 users and more coming every day. The x264 developers are doing a great job, I thank them all!