Power machine for x264 encoding

fangorn · 14th February 2009, 14:45

Hi,

For the last year I only built energy efficient low budget machines. So my knowledge of the performance sector is pretty outdated.

What I want to do is transcode HD Video to h.264 using x264, ffmpeg and mencoder (under Linux if this is of any interest). My current machine is >2 years and (for this task) as fast as the continental drift. As it is a perfectly good machine for all other tasks, I would be interested in upgrading.

For instance: Does someone here know of someone working on porting x264 to CUDA or the AMD version (the name fails me atm)? I would much prefer just putting in a GPU over building a whole new machine and would gladly help with testing and debugging.

If there is no such thing in the near future, I am looking for a replacement machine for my Athlon 64X2 2GHz with min. 3 times the power. So I think it will be an Intel machine atm. And with Intel my knowledge is even more outdated.

What I need is a stable machine (as in half a year uptime minimum). In that field I am not so satisfied with the Intel boxes we have at work. So tips are much appreciated.

I hope to get some helpful comments,
fangorn

jeffy · 14th February 2009, 15:02

Ideas: Let's start somewhere: Core i7 920, X58 chipset board, 3GB DDR3

Ma-Xell · 14th February 2009, 15:36

A fast Core2Quad might be cheaper than an i7 but socket 775 has no more future.

LoRd_MuldeR · 14th February 2009, 15:46

x264 benefits from Core i7 (Nehalem) greatly, as described here:
http://x264dev.multimedia.cx/?p=51

Quote:

Overall, the changes in Nehalem are extremely beneficial to x264 and have led to an enormous overall performance increase. Furthermore, since the primary speed increase is in SIMD, the more assembly code we write, the more of a boost Nehalem gets over previous processors.

That's why I'd go with a Core i7 nowadays. Anyway, a Core2 Quad isn't a bad choice either

(BTW: I don't expect that any GPGPU encoder can beat x264 speed-wise while retaining a similar quality at the same bitrate. At least not in the near future...)

Reimar · 14th February 2009, 16:51

Quote:

Originally Posted by fangorn

For instance: Does someone here know of someone working on porting x264 to CUDA or the AMD version (the name fails me atm)? I would much prefer just putting in a GPU over building a whole new machine and would gladly help with testing and debugging.

I am not aware of such plans and generally I'd expect the results to not be good enough to give you better performance per dollar than a new CPU.

Quote:

Originally Posted by fangorn

If there is no such thing in the near future, I am looking for a replacement machine for my Athlon 64X2 2GHz with min. 3 times the power. So I think it will be an Intel machine atm. And with Intel my knowledge is even more outdated.

As others have said, i7 is vastly (EDIT: well, vastly may be an exaggeration, not sure) more powerful, but if you do not like rebuilding, you could also check which CPUs your mainboard supports.
I have about the same CPU as you and mine can handle e.g. the Phenoms (which from what I know are not really impressive in performance though, but 3 times faster than your current CPU is not that hard to do). Particularly if the Phenom IIs work in yours they might be the cheapest way to get the performance you want.

fangorn · 14th February 2009, 17:23

My Board unluckily only supports a Phenom X4 with 2,2 GHz max.

This would be an option, but it is hard to get nowadays and the benefit would be not so big (x2).

If this has to make sense I will go for Core i7. Any suggestions for Mainboards? Normally I prefer Asus (as already said, I prefer stable systems, so overclocking is not an option) and Hardware that is min. 6 months old.

fangorn · 14th February 2009, 17:27

Quote:

Originally Posted by LoRd_MuldeR

That's why I'd go with a Core i7 nowadays. Anyway, a Core2 Quad isn't a bad choice either

Of what difference are we talking here in benefit? Is it percent or factors? Just to decide if it is worth the price.

LoRd_MuldeR · 14th February 2009, 17:28

Quote:

Originally Posted by fangorn

My Board unluckily only supports a Phenom X4 with 2,2 GHz max.

This would be an option, but it is hard to get nowadays and the benefit would be not so big (x2).

Moving from 2 cores to 4 cores (at roughly same clock speed) means doubled performance for x264, as it scales very well!

Quote:

Originally Posted by fangorn

If this has to make sense I will go for Core i7. Any suggestions for Mainboards? Normally I prefer Asus (as already said, I prefer stable systems, so overclocking is not an option) and Hardware that is min. 6 months old.

I have good experience with Gigabyte boards, I'm currently running a GA-P35-DS3R. Also it seems X58 is the chipset of choice for Core i7 currently.

So maybe that one:
http://www.alternate.de/html/product...l3=Sockel+1366

Or if you prefer Asus:
http://www.alternate.de/html/product...l3=Sockel+1366

LoRd_MuldeR · 14th February 2009, 17:45

Quote:

Originally Posted by fangorn

Of what difference are we talking here in benefit? Is it percent or factors? Just to decide if it is worth the price.

Some benchmarks:
http://www.anandtech.com/cpuchipsets...px?i=3448&p=18
http://www.techarp.com/showarticle.a...tno=603&pgno=8

And we don't know if they used a recent version of x264, that already contains the Core i7 optimizations.
One benchmarks says they used x264 r819, so a recent x264 version would run even faster on a Core i7 machine!

fangorn · 14th February 2009, 17:52

Thanks for the tips, I have to adjust at the Intel prices at the moment

For the price of Processor and Board alone I built two AMD boxes last year. But I'm in the performance league now. We wanna build a machine to replace four or more of the boxes I am used to.

DJ Bobo · 14th February 2009, 18:35

Uuh...

Quote:

I am looking for a replacement machine for my Athlon 64X2 2GHz with min. 3 times the power. So I think it will be an Intel machine atm

Funny conclusion. You didn't hear about the AMD Phenom II CPUs, did you? The 3GHz version will be at least 3 times faster than your current X2.
The Phenom II is much cheaper than the Core i7 CPUs (~210€ for the Phenom II @3GHz but ~270€ for the Core i7 @2,67GHz). It also runs much cooler (15°C less than the Core i7 using the same cooler) and consumes less energy (20W less than the Core i7 when fully loaded)

jeffy · 14th February 2009, 18:50

Phenom II: another graph: http://www.anandtech.com/cpuchipsets...px?i=3492&p=14

fangorn · 14th February 2009, 18:53

Yes I have heard of the PhenomII. But I have also heard that the fastest of them compares - in some benchmarks - to the slowest i7.

I am an AMD guy by heart, but I am a pragmatic engineer also. When I can get say 50% more performance for 20% more money I take the chance.

All depends on benchmarks comparing i7 and Phenom2, preferrably at same rate, while performing x264 encoding. Everything else I could do with the machine I have now. Actually I will build a new machine anyway and use the old one when I don't have to do heavy encoding. I have never dropped working hardware (besides harddrives

)

@Jeffy
Thanks, that is what I was looking for.

gigah72 · 14th February 2009, 19:43

i'd expect the i7-920 to beat the PII-X4 940 (in x264) by 30-40% @ 50-60% more the price of the amd (cpu+board+ram). you need to choose what's more important to you.

fangorn · 14th February 2009, 20:49

Thanks, I will have something to think of till they start delivering on monday.

fangorn · 15th February 2009, 10:41

One last question. Not exactly hardware related though.

At work we compile our binaries with ifc/icc. Has somebody here experience with compiling x264/mencoder/ffmpeg/vlc, ... with ICC? If it works is there a noticable speed increase over GCC compiled binaries? I know you can patch ICC compiles so they run optimized on AMD too, but I never used it.

Reimar · 15th February 2009, 13:00

Quote:

Originally Posted by fangorn

At work we compile our binaries with ifc/icc. Has somebody here experience with compiling x264/mencoder/ffmpeg/vlc, ... with ICC?

It works fine for ffmpeg, but you might have to search a bit for a version that does not miscompile. Note that it's relevant that you use Linux, only Linux ICC supports gcc-syntax, you would not be able to use Windows-ICC.

Quote:

Originally Posted by fangorn

If it works is there a noticable speed increase over GCC compiled binaries?

For some things, mostly H.264 decoding. I think that only applies to 32 bit binaries though, compiling 64 bit binaries reduces register pressure and gcc does a much better job, and the code is faster anyway (e.g. since also the system libraries can use MMX and SSE unconditionally).
x264 probably has more than enough hand-optimized assembler that the compiler will almost never make a difference.
I admit I am only guessing, it has been ages since I actually tested that stuff.

Quote:

Originally Posted by fangorn

I know you can patch ICC compiles so they run optimized on AMD too, but I never used it.

ICC binaries usually run about as well on AMD hardware as on Intel. Yes, I heard about that one issue there was, but I suspect I may have been only the Windows-Compilers that had it...

Dark Shikari · 15th February 2009, 13:06

Quote:

Originally Posted by Reimar

For some things, mostly H.264 decoding. I think that only applies to 32 bit binaries though, compiling 64 bit binaries reduces register pressure and gcc does a much better job

ICC seems to be equally better on 32-bit and 64-bit in my experience.

Quote:

Originally Posted by Reimar

x264 probably has more than enough hand-optimized assembler that the compiler will almost never make a difference.

x264 is up to 3-4% faster with ICC, depending on settings.

fangorn · 15th February 2009, 15:33

@Reimar
I was talking about Linux. As well at work as at home. I think I have a virtual machine with windows somewhere, but I don't know where atm.

And our tests with IFC 9 linux showed exactly that issue. Intel Core2 Processor 5 to 6 times faster than AMD Athlon 64 X2 at same cycle rate! Just because the binaries checked the processor ID and used completely unoptimized code if not Intel. For testers I patched one binary and the difference was nearly gone. I don't know if they correct more modern versions of the compilers. I havent followed the news concerning this. As this is only relevant for one of our programs that does heavy interpolation calculations, we just use the Intel machines to do that.

@All
Thanks again, you all were very helpful.

blubberbirne · 15th February 2009, 20:30

i prefer i7

because i have i7

but look at this results.....

First a Q9550

Code:

x264 HD BENCHMARK RESULTS 
Please copy/paste everything below the line into the forum post to report your data 
^^^^^^^^^^^^^^^^^^ 
 
Results for x264.exe v0.58.819 
encoded 1442 frames, 61.85 fps, 3886.74 kb/s
encoded 1442 frames, 60.84 fps, 3887.22 kb/s
encoded 1442 frames, 58.78 fps, 3886.56 kb/s
encoded 1442 frames, 57.75 fps, 3888.99 kb/s
encoded 1442 frames, 20.55 fps, 3966.41 kb/s
encoded 1442 frames, 18.98 fps, 3966.38 kb/s
encoded 1442 frames, 18.75 fps, 3969.14 kb/s
encoded 1442 frames, 18.62 fps, 3966.03 kb/s
 
Results for x264.exe v0.59.1101 
encoded 1442 frames, 76.14 fps, 3976.48 kb/s
encoded 1442 frames, 72.10 fps, 3976.48 kb/s
encoded 1442 frames, 71.37 fps, 3976.48 kb/s
encoded 1442 frames, 71.49 fps, 3976.48 kb/s
encoded 1442 frames, 20.91 fps, 3935.63 kb/s
encoded 1442 frames, 20.41 fps, 3935.83 kb/s
encoded 1442 frames, 20.30 fps, 3935.23 kb/s
encoded 1442 frames, 20.49 fps, 3935.87 kb/s
 
 
System Details 
-------------- 
Name			Intel Core 2 Quad Q9550
Codename		Yorkfield
Specification		Intel(R) Core(TM)2 Quad  CPU   Q9550  @ 2.83GHz
Core Stepping		C1
Technology		45 nm
Stock frequency		2833 MHz
Core Speed		2833.6 MHz (8.5 x 333.4 MHz)
FID range		6.0x - 8.5x
 
Northbridge		Intel P45 rev. A3
Southbridge		Intel 82801JR (ICH10R) rev. 00
 
CAS#			5.0
RAS# to CAS#		5
RAS# Precharge		5
Cycle Time (tRAS)	18
Command Rate		2T
Memory Frequency	400.0 MHz (5:6)
Memory Type		DDR3
Memory Size		3328 MBytes
Channels		Dual (Symmetric)
 
Windows Version		Microsoft Windows XP Professional  Service Pack 3 (Build 2600) 
 
max VID			1.250 V
Voltage sensor 0	1.13 Volts [0x8D] (CPU VCORE)
Number of processors	1
Number of threads	4
Number of threads	4 (max 4)
L2 cache		2 x 6144 KBytes, 24-way set associative, 64-byte line size
Instructions sets	MMX, SSE, SSE2, SSE3, SSSE3, SSE4.1, EM64T
Package			Socket 775 LGA (platform ID = 4h)
 
Temperature sensor 0	43°C (109°F) [0x2B] (SYSTIN)
Temperature sensor 1	72°C (161°F) [0x90] (CPUTIN)
Temperature sensor 2	12°C (52°F) [0x1E9] (AUXTIN)
Temperature sensor 0	90°C (193°F) [0xF] (core #0)
Temperature sensor 1	90°C (193°F) [0xF] (core #1)
Temperature sensor 2	100°C (211°F) [0x5] (core #2)
Temperature sensor 3	102°C (215°F) [0x3] (core #3)
Temperature sensor 0	53°C (127°F) [0x35] (GPU Core)

And here my i7 920

Code:

x264 HD BENCHMARK RESULTS 
Please copy/paste everything below the line into the forum post to report your data 
^^^^^^^^^^^^^^^^^^ 
 
Results for x264.exe v0.58.819 
encoded 1442 frames, 73.10 fps, 3895.56 kb/s
encoded 1442 frames, 71.76 fps, 3896.01 kb/s
encoded 1442 frames, 72.46 fps, 3896.01 kb/s
encoded 1442 frames, 71.63 fps, 3894.31 kb/s
encoded 1442 frames, 29.61 fps, 3986.59 kb/s
encoded 1442 frames, 29.72 fps, 3986.64 kb/s
encoded 1442 frames, 29.75 fps, 3986.64 kb/s
encoded 1442 frames, 29.53 fps, 3983.05 kb/s
 
Results for x264.exe v0.59.1101 
encoded 1442 frames, 85.37 fps, 3976.27 kb/s
encoded 1442 frames, 86.49 fps, 3976.27 kb/s
encoded 1442 frames, 85.15 fps, 3976.27 kb/s
encoded 1442 frames, 86.40 fps, 3976.27 kb/s
encoded 1442 frames, 33.94 fps, 3938.61 kb/s
encoded 1442 frames, 33.89 fps, 3937.93 kb/s
encoded 1442 frames, 33.95 fps, 3938.65 kb/s
encoded 1442 frames, 33.89 fps, 3938.68 kb/s
 
 
System Details 
-------------- 
Name			Intel Processor
Codename		Bloomfield
Specification		Intel(R) Core(TM) i7 CPU         920  @ 2.67GHz (Engineering Sample)
Core Stepping		
Technology		45 nm
Core Speed		2797.9 MHz (21.0 x 133.2 MHz)
 
Northbridge		Intel X58 rev. 12
Southbridge		Intel 82801JR (ICH10R) rev. 00
 
Memory Frequency	1332.3 MHz ()
Memory Type		DDR3
Memory Size		6142 MBytes
 
Windows Version		Microsoft Windows Vista (6.0) Ultimate Edition  Service Pack 1 (Build 6001) 
 
max VID			0.825 V
Voltage sensor 0	1.20 Volts [0x4B] (CPU VCORE)
Number of processors	1
Number of threads	8
Number of threads	8 (max 16)
L2 cache		4 x 256 KBytes, 8-way set associative, 64-byte line size
Instructions sets	MMX, SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, EM64T
Package			Socket 1366 LGA (platform ID = 1h)
 
Temperature sensor 0	35°C (94°F) [0x23] (TMPIN0)
Temperature sensor 1	46°C (114°F) [0x2E] (TMPIN1)
Temperature sensor 2	51°C (123°F) [0x33] (TMPIN2)
Temperature sensor 0	62°C (143°F) [0x2B] (core #0)
Temperature sensor 1	63°C (145°F) [0x2A] (core #1)
Temperature sensor 2	60°C (139°F) [0x2D] (core #2)
Temperature sensor 3	57°C (134°F) [0x30] (core #3)
Temperature sensor 0	43°C (109°F) [0x2B] (GPU Core)

Second pass is more than 50% percent faster

14th February 2009, 14:45	#1 \| Link
fangorn Registered User Join Date: Jun 2008 Posts: 42	Power machine for x264 encoding Hi, For the last year I only built energy efficient low budget machines. So my knowledge of the performance sector is pretty outdated. What I want to do is transcode HD Video to h.264 using x264, ffmpeg and mencoder (under Linux if this is of any interest). My current machine is >2 years and (for this task) as fast as the continental drift. As it is a perfectly good machine for all other tasks, I would be interested in upgrading. For instance: Does someone here know of someone working on porting x264 to CUDA or the AMD version (the name fails me atm)? I would much prefer just putting in a GPU over building a whole new machine and would gladly help with testing and debugging. If there is no such thing in the near future, I am looking for a replacement machine for my Athlon 64X2 2GHz with min. 3 times the power. So I think it will be an Intel machine atm. And with Intel my knowledge is even more outdated. What I need is a stable machine (as in half a year uptime minimum). In that field I am not so satisfied with the Intel boxes we have at work. So tips are much appreciated. I hope to get some helpful comments, fangorn

14th February 2009, 18:53	#13 \| Link
fangorn Registered User Join Date: Jun 2008 Posts: 42	Yes I have heard of the PhenomII. But I have also heard that the fastest of them compares - in some benchmarks - to the slowest i7. I am an AMD guy by heart, but I am a pragmatic engineer also. When I can get say 50% more performance for 20% more money I take the chance. All depends on benchmarks comparing i7 and Phenom2, preferrably at same rate, while performing x264 encoding. Everything else I could do with the machine I have now. Actually I will build a new machine anyway and use the old one when I don't have to do heavy encoding. I have never dropped working hardware (besides harddrives ) @Jeffy Thanks, that is what I was looking for. Last edited by fangorn; 14th February 2009 at 18:55.

15th February 2009, 15:33	#19 \| Link
fangorn Registered User Join Date: Jun 2008 Posts: 42	@Reimar I was talking about Linux. As well at work as at home. I think I have a virtual machine with windows somewhere, but I don't know where atm. And our tests with IFC 9 linux showed exactly that issue. Intel Core2 Processor 5 to 6 times faster than AMD Athlon 64 X2 at same cycle rate! Just because the binaries checked the processor ID and used completely unoptimized code if not Intel. For testers I patched one binary and the difference was nearly gone. I don't know if they correct more modern versions of the compilers. I havent followed the news concerning this. As this is only relevant for one of our programs that does heavy interpolation calculations, we just use the Intel machines to do that. @All Thanks again, you all were very helpful. Last edited by fangorn; 15th February 2009 at 15:41.

14th February 2009, 15:02	#2 \| Link
jeffy Registered User Join Date: Jan 2007 Posts: 943	Ideas: Let's start somewhere: Core i7 920, X58 chipset board, 3GB DDR3

14th February 2009, 15:36	#3 \| Link
Ma-Xell Somewhere Back in Time Join Date: Jul 2008 Location: Hungary Posts: 61	A fast Core2Quad might be cheaper than an i7 but socket 775 has no more future.

14th February 2009, 17:23	#6 \| Link
fangorn Registered User Join Date: Jun 2008 Posts: 42	My Board unluckily only supports a Phenom X4 with 2,2 GHz max. This would be an option, but it is hard to get nowadays and the benefit would be not so big (x2). If this has to make sense I will go for Core i7. Any suggestions for Mainboards? Normally I prefer Asus (as already said, I prefer stable systems, so overclocking is not an option) and Hardware that is min. 6 months old.

14th February 2009, 17:52	#10 \| Link
fangorn Registered User Join Date: Jun 2008 Posts: 42	Thanks for the tips, I have to adjust at the Intel prices at the moment For the price of Processor and Board alone I built two AMD boxes last year. But I'm in the performance league now. We wanna build a machine to replace four or more of the boxes I am used to.

14th February 2009, 18:50	#12 \| Link
jeffy Registered User Join Date: Jan 2007 Posts: 943	Phenom II: another graph: http://www.anandtech.com/cpuchipsets...px?i=3492&p=14

14th February 2009, 19:43	#14 \| Link
gigah72 Registered User Join Date: Feb 2002 Location: OF/Germany Posts: 111	i'd expect the i7-920 to beat the PII-X4 940 (in x264) by 30-40% @ 50-60% more the price of the amd (cpu+board+ram). you need to choose what's more important to you.

14th February 2009, 20:49	#15 \| Link
fangorn Registered User Join Date: Jun 2008 Posts: 42	Thanks, I will have something to think of till they start delivering on monday.

15th February 2009, 10:41	#16 \| Link
fangorn Registered User Join Date: Jun 2008 Posts: 42	One last question. Not exactly hardware related though. At work we compile our binaries with ifc/icc. Has somebody here experience with compiling x264/mencoder/ffmpeg/vlc, ... with ICC? If it works is there a noticable speed increase over GCC compiled binaries? I know you can patch ICC compiles so they run optimized on AMD too, but I never used it.