Re: x264 HD Benchmark

[graysky]

Originally Posted by torrentthief

How about you host build 819 x264.exe and get users to submit those results too, add another column to the spreadsheet with "build 819" as title and add a new build every 2 months as this will showcase what new cpu's can do as multithreading and sse3 improvements in x264 each month are large.

here is build 819 sharktooth patched: sendspace.com/file/c1ps6j (368kb rarred)

Overclocked result with build 819:

1st pass avg: 74.43 fps (8.7% faster)
2nd pass avg: 21.34 fps (12.37% faster)

Also you have written Q9540 instead of Q9450 several times in your spreadsheet, could you change them please.

I hope you can add a column for build 819 as people use this spreadsheet to decide what cpu to buy if they do lots of x264 encoding, older cpu's wont get much faster with new builds but new cpu's like my 45nm quad will.


---------- RUN1PASS1.LOG
encoded 1442 frames, 74.43 fps, 3888.27 kb/s

---------- RUN2PASS1.LOG
encoded 1442 frames, 74.43 fps, 3887.23 kb/s

---------- RUN3PASS1.LOG
encoded 1442 frames, 74.31 fps, 3889.34 kb/s

---------- RUN4PASS1.LOG
encoded 1442 frames, 74.55 fps, 3888.84 kb/s

---------- RUN1PASS2.LOG
encoded 1442 frames, 21.34 fps, 3963.26 kb/s

---------- RUN2PASS2.LOG
encoded 1442 frames, 21.30 fps, 3963.56 kb/s

---------- RUN3PASS2.LOG
encoded 1442 frames, 21.37 fps, 3962.75 kb/s

---------- RUN4PASS2.LOG
encoded 1442 frames, 21.36 fps, 3962.82 kb/s

How about it all... I like the idea of including data for new releases of x264.exe since it is under heavy development and always improving. As TT pointed out, the current version does give slightly higher numbers.

One downside to doing this is that when a new version is placed in the rar file for download, all the data generated on the "old" version becomes invalid since you can't compare back. Another argument in favor of doing this would be that people would get a more accurate idea how a given system can perform encoding a given 720p clip.

I guess the real question is what do people want out of the benchmark. Do they want a standard by which systems can be compared to each other in relative terms or do they want more of a dual purpose benchmark that does what I just mentioned as well as shows what the current version of the x264 can do on a given platform.

Another downside I can see to constantly updating is that the table would get HUGE and it's already too big in my opinion (lots of data in there). I guess I'm kinda talking myself out of doing this i.e. just keeping it w/ version 0.58.474 even though it's a tad slower than the latest and greatest builds. We have amassed a good deal of data points already...

Discussion please!

[torrentthief]

not sure what you mean by:

"One downside to doing this is that when a new version is placed in the rar file for download, all the data generated on the "old" version becomes invalid since you can't compare back."

You can keep the current rar which is 40mb or similar hosted but have a 2nd rar (450kb ish) with build 819 of x264.exe, the user can run the test with the old build which creates a results.txt or similar file, then they can overwrite the old x264.exe with build 819 and re-run the test and create results2.txt and upload them both here.

A column on the spreadsheet called build 819 would give users a good idea of what a current cpu/chipset can do as quad core cpu's and sse3/4.1 will improve drastically for x264 as there is huge development for it. If you compare a single core AMD 3000+ i doubt you would see the performance increases that i got:

1st pass avg: 77.41 fps (9.55% faster)
2nd pass avg: 22.25 fps (12.71% faster)

If like me you encode x264 sports from tv captures, 13% faster is a large improvement if your encode takes 2hrs.

Every 2 months on the 1st of each month add the latest sharktooth patched x264 build to the benchmark table. Rather simple really, only a bit of extra data entry needed.

We really need to use 1080p h264 video instead of 720p mpeg2 source as 440*6 gave virtually identical fps as 440*7.5 thats 660mhz difference in cpu speed, make a 1080p h264 80mb.h264 file and this will give a far more accurate comparison of processors instead of their fsb speeds which is currently clearly does as there was only 1fps difference in my 1st pass and 0.5fps difference in my 2nd pass yet 2.64ghz should be alot slower than 3.3ghz.

[Adrian Wong]

Yeah, maybe it would be good to have separate data for multiple versions of the benchmark. Not necessary in the same table, we can always split it up. Meaning once there's a new version, we just keep the old version's data aside (still available) and start on the data of the new version?

However, I do agree that the data entry takes up a lot of time. That is really our fault (not graysky's). We had promised him a database-run system and we have not delivered it. I will look into getting something done about this. So sorry, graysky!!

[graysky]

OK guys, I'll bite. I'm in the process of updating it now. The new version will have two major changes:

-Includes CPUz and greps through the output to make collecting system specs easy for you and me!

Here is sample output from my system (pay no attention to the values for FPS, I was heavily using the machine while it ran but you get the idea):

x264 HD BENCHMARK RESULTS 
Please copy/paste everything below the line into the forum post to report your data 
^^^^^^^^^^^^^^^^^^ 
 
Results for x264.exe v0.58.747 
encoded 1442 frames, 60.95 fps, 3904.62 kb/s
encoded 1442 frames, 61.40 fps, 3904.62 kb/s
encoded 1442 frames, 60.36 fps, 3904.62 kb/s
encoded 1442 frames, 61.08 fps, 3904.62 kb/s
encoded 1442 frames, 16.76 fps, 3953.18 kb/s
encoded 1442 frames, 16.66 fps, 3953.18 kb/s
encoded 1442 frames, 16.67 fps, 3953.18 kb/s
encoded 1442 frames, 16.74 fps, 3953.18 kb/s
 
Results for x264.exe v0.59.819M 
encoded 1442 frames, 65.41 fps, 3889.68 kb/s
encoded 1442 frames, 65.18 fps, 3889.34 kb/s
encoded 1442 frames, 65.31 fps, 3890.65 kb/s
encoded 1442 frames, 65.36 fps, 3889.29 kb/s
encoded 1442 frames, 18.68 fps, 3962.71 kb/s
encoded 1442 frames, 18.70 fps, 3962.63 kb/s
encoded 1442 frames, 18.61 fps, 3962.45 kb/s
encoded 1442 frames, 18.67 fps, 3962.75 kb/s
 
 
System Details 
-------------- 
Name			Intel Core 2 Quad Q6600
Codename		Kentsfield
Specification		Intel(R) Core(TM)2 Quad CPU           @ 2.40GHz
Core Stepping		B3
Technology		65 nm
Stock frequency		2400 MHz
Core Speed		2999.8 MHz (9.0 x 333.3 MHz)
FID range		6.0x - 9.0x
 
Northbridge		Intel P965/G965 rev. C2
Southbridge		Intel 82801HB/HR (ICH8/R) rev. 02
 
CAS#			4.0
RAS# to CAS#		4
RAS# Precharge		4
Cycle Time (tRAS)	10
Command Rate		2T
Memory Frequency	333.3 MHz (1:1)
Memory Type		DDR2
Memory Size		4096 MBytes
Channels		Dual (Symmetric)
 
Windows Version		Microsoft Windows XP x64 Professional  Service Pack 2 (Build 3790) 
 
max VID			1.288 V
Voltage sensor 0	1.23 Volts [0x9A] (CPU VCORE)
Number of processors	1
Number of threads	4
Number of threads	4 (max 4)
L2 cache		2 x 4096 KBytes, 16-way set associative, 64-byte line size
Instructions sets	MMX, SSE, SSE2, SSE3, SSSE3, EM64T
Package			Socket 775 LGA (platform ID = 4h)
 
Temperature sensor 0	45°C (112°F) [0x2D] (SYSTIN)
Temperature sensor 1	48°C (118°F) [0x60] (CPUTIN)
Temperature sensor 2	127°C (260°F) [0xFE] (AUXTIN)
Temperature sensor 0	58°C (136°F) [0x2A] (core #0)
Temperature sensor 1	59°C (138°F) [0x29] (core #1)
Temperature sensor 2	58°C (136°F) [0x2A] (core #2)
Temperature sensor 3	60°C (139°F) [0x28] (core #3)
Temperature sensor 0	56°C (132°F) [0x38] (GPU Core)

-Included x264 0.59.819 and an option for users to run it and post the data as well.

I'll need some beta testers with XP and Vista.
I have tested it on XP x86 and XP x64 but I don't have Vista. Also, I'd like some feedback from you too even if you have XP.

I'll PM the link to interested parties, just post in this thread saying you're interested.

Here is a small graphic illustrating the 9-10 % difference on quads. I ran it on an older P4 today and the new build was 12-13 % faster, so whatever optimization is in there, it's NOT using SSE4.

[graysky]

@torrentthief - I tried to send you a PM with the URL to d/l this but there isn't an option to send you a PM! I can send you an email, but it said you have that option turned off...??

[Adrian Wong]

Originally Posted by graysky

@torrentthief - I tried to send you a PM with the URL to d/l this but there isn't an option to send you a PM! I can send you an email, but it said you have that option turned off...??

He must have turned off both PM and e-mail options.

[graysky]

Oh well, I made this version live now and updated the main page.

[torrentthief]

think its turned off by default, i just registered a few days ago and didnt change settings. I'll give the new version a try soon, killed my motherboard during overclocking, will have to wait till i get my replacement Put too many volts thought my motherboard

Good idea about the cpu-z creating the specs.txt file.

Will let you know about my results on vista x64 and xp pro sp2.

[Mac Daddy]

Dang on the mobo torrentthief sh*t happens sometimes hope ya get it sorted Bro

Graysky cool on the addition but to members using Vista might be an idea to disable UAC for the test for now

[graysky]

Originally Posted by Mac Daddy

Dang on the mobo torrentthief sh*t happens sometimes hope ya get it sorted Bro

Graysky cool on the addition but to members using Vista might be an idea to disable UAC for the test for now

Yeah, UAC disables CPU-Z from running... we're looking it a work-around now (runas doesn't do it).