Quadcore or server processor for x264?

[dvd_maniac]

Well then, I currently encode using an AMD X2 4400+. If I built a machine with comparable components using a conroe C2D E6600 2.4Ghz with the 4MB level2 cache what type of increase should I expect? Is it better to get the 4MB cache over the 2MB one for encoding? Does Ram timings play a huge role with encoding?

[bkman]

Well then, I currently encode using an AMD X2 4400+. If I built a machine with comparable components using a conroe C2D E6600 2.4Ghz with the 4MB level2 cache what type of increase should I expect?

You are looking at a rough increase of about 35% in encoding speed, if you don't overclock. Not sure how much the cache and ram timings will affect things, but probably no more than 5-10% together based on some benchmarks I have seen.

[Inventive Software]

I'm now looking for a Pentium D as a cheap dual-core system around Xmas (I'm a student, I'm short of money, I'm realistic), so how does that compare? Does the long pipeline hurt it?

[bkman]

I'm now looking for a Pentium D as a cheap dual-core system around Xmas (I'm a student, I'm short of money, I'm realistic), so how does that compare? Does the long pipeline hurt it?

The architecture of Pressler cores doesn't hurt x264 performance as much as Xvid performance, but it does still hurt compared to the C2D. You get roughly half the performance per clock compared to C2D, so that should give you a reference for the value of a Pentium D purchase.

[Blue_MiSfit]

Dude you can build a basic, complete core 2 duo system for encoding for a tad over $600 shipped (just the tower).

E6300, Antec case + Power Supply, Biostar 945G motherboard, 1GB DDR2 800, 250GB, DVDRW

The only REAL difference between a Pentium D system and Core 2 Duo system is the processor - the cheapest C2D is $180 and the cheapest Pentium D is $95 for a crappy 2.66 GHz. Extra $100 for nearly 2x the performance (more if you overclock)?

To me it's a no brainer.

~MiSfit

[foxyshadis]

I'm now looking for a Pentium D as a cheap dual-core system around Xmas (I'm a student, I'm short of money, I'm realistic), so how does that compare? Does the long pipeline hurt it?

Are you just swapping out a LGA775 P4 for a P-D? That's the only thing I could see making sense, you can keep the motherboard and everything else. If you buy a whole system, make it X2 or C2D, which is only a $50-100 premium over P-D for a large performance and heat/noise benefit.

At Dell, the budget X2 systems are actually less than the P-Ds, but that's only because they're scamming huge profits on Intel cpus. ($100+ over OEM list on P-Ds, $150+ over for C2Ds.)

[Inventive Software]

No, this is an upgrade from a Celeron 800 MHz which is my current system to Pentium D or C2D.

Currently I have about £290 for a case with 400W PSU, 1GB RAM, C2D supported motherboard and C2D E6300. It's not cheap, but that's essentially what I need. I have a graphics card and cables. If anybody can help me get that lower than £290, please post links! :D

EDIT: With a Pentium D 805 it's around £200, which is more sensible, but I'd really like C2D for the overclocking prowess and purely for the fact it's faster!

[swaaye]

FYI, I got the base model E6300 up to 2.5 GHz on stock voltage on a cheapie uATX mobo (ASUS P5L-MX). It's amazing how well C2D clocks. This thing probably will go a good bit higher but the mobo's PCIe lock doesn't work and SATA stops working if I go any higher. I had to stick a PCI NIC in just to get this far cuz the onboard PCIe GiGE NIC went out at only 300 FSB.

All I know is the thing slaughters my 2.6 GHz Opteron 165 at home, and that thing heats up like no tomorrow at the 1.45V or so it needs for that clock. It's also on a uber overclocker board (DFI NF4 Ultra-D) with a $120 600W PSU to keep things stable. The uATX C2D has a 3 yr old $50 420W Thermaltake that isn't even 24-pin native, lol. :)

So I'd say the platform is anything but more expensive. I have a hard time seeing any AMD advantage at all right now.

[Sharktooth]

Well, AMD's 4x4 is coming (4 cores and 4 graphic processors) along with a new shiny new set of Athlon64 FX CPUs (new socket F, higher frequencies and larger cache). Soon we'll see the first benchmarks and comparative results.

http://www.tcmagazine.com/comments.php?shownews=12860&catid=2

[Dust Signs]

For those of you who understand German, here is an article with a little more details: http://www.computerbase.de/news/hardware/prozessoren/amd/2006/november/neue_informationen_amds_4x4-plattform/

Dust Signs

[EagleDM]

For the record, here at MaximoPC we tried 3 variants of C2D

The E6300, the E6400 and E6600


The 6300 and 6400 has 2Mb Cache and E660 is the Conroe variant with 4Mb cache (The 2Mb variant is called Allendale and is pretty good too)


All of those processors can run (WITH NO INCREASE IN VOLTAGE) to at easy 3Ghz, that means, almost ANY Core2Duo arquitecture (including the basic models) should do AT LEAST 3Ghz


People I know, fans of overclocking could do almost 3.5Ghz with the lowest variant (E6300) and some achieved 3.7Ghz with the E6600 variant

I must say (because I did the test) at 3Ghz (no matter what variant you have) expect to have so BRUTAL encoding speed, C2D arquitecture really begins to speed up on the 3Ghz barrier, this processors are born for speed and although it performance very good for the stock, when they reach 3Ghz these are monsters, and, since almost ALL (if not all) can reach 3Ghz without any kinds of problem (not even messing up with the voltage of the CPU) you have to spend more time on the motherboard and memory.

My recomendation to anyone who is found of Overclocking and want to give those puppys a try, get the Asus P5B motherboard, it's like the bible for the Core2Duo amont overclocking fans, you will not get other motherboard with that potencial.

[Sharktooth]

Just a note. SSE speed is proportional to clock speed.
Pentium 4s were monster CPUs for what concerned SSE execution coz of the high clock rates.
C2D have rather low clock rates but have "brutal" SSE execution speed due to the 128 bit SSE execution. What would happen when Athlon 64s will get 128 bit SSE execution?
Those chips already run at higher clock speed than C2D. So what will be the Intel answer? As we all know C2D have a high overclock potential and i suspect intel is selling "underclocked" C2Ds coz, at this time, they're not interested in producing CPUs with more performance. So, in the future, when they need to compete again for the performance, they will sell the same CPUs with different multipliers (to obtain higher clock rates) at a higher price.
So, basically, enjoy the C2D overclocking now, coz there's no reason to not overclock those CPUs.

[bill_baroud]

Hello, sorry for the up from outerspace (well 15 days... ;)) but i just had my hand on a dual-woodcrest (xeon core 2 duo equivalent) 3Ghz server so i tried a really quick encoding to see how it would perform and i thought it would interest some of you.

x264 was build from the 29/11/2006 daily snapshot on a debian sarge.
Source is uncompressed 1920x1080 4:2:0, 217 frames. (ftp://ftp.ldv.e-technik.tu-muenchen.de/dist/test_sequences/1080p/)
Out of inspiration, I tried a Sharktooth's command line found in the HD-DVD challenge thread ;) but in 1 pass.
Tried 1, 2, 4 (number of cores) and 8 threads.

Results
1 thread : 1.09 fps
2 threads : 1.93 fps
4 threads : 3.23 fps (2.55 fps with ref 3, umh and subme 7)
8 threads : 3.25 fps

Detailed Results

model name : Intel(R) Xeon(R) CPU 5160 @ 3.00GHz (Woodcrest) x 2
cpu MHz : 2992.752
cache size : 4096 KB
bogomips : 5989.03



./x264 --keyint 15 --min-keyint 1 --vbv-maxrate 29400 --vbv-bufsize 9500 --qpmin 10 --level 4.1 --bframe 2 --b-rdo --bime --weightb --ref 3 --mixed-refs --direct auto --filter 0:0 --bitrate 12000 --stats "H264_12Mbps.log" --qcomp 0.75 --ipratio 1.10 --pbratio 1.33 --analyse "all" --8x8dct --me "hex" --subme 6 --no-fast-pskip --trellis 2 --threads 1 --progress -o ../output.264 ../blue_sky.yuv 1920x1080
x264 [warning]: width or height not divisible by 16 (1920x1080), compression will suffer.
x264 [info]: using cpu capabilities MMX MMXEXT SSE SSE2
x264 [info]: slice I:15 Avg QP:21.67 size:281868 PSNR Mean Y:43.43 U:43.70 V:44.68 Avg:43.65 Global:43.15
x264 [info]: slice P:98 Avg QP:21.84 size: 94491 PSNR Mean Y:42.68 U:42.36 V:43.18 Avg:42.69 Global:41.99
x264 [info]: slice B:104 Avg QP:24.09 size: 20499 PSNR Mean Y:41.68 U:41.00 V:42.27 Avg:41.63 Global:40.94
x264 [info]: mb I I16..4: 25.3% 41.6% 33.1%
x264 [info]: mb P I16..4: 7.3% 3.9% 0.9% P16..4: 44.2% 20.5% 7.8% 0.7% 0.3% skip:14.4%
x264 [info]: mb B I16..4: 0.1% 0.1% 0.0% B16..8: 17.8% 0.5% 1.2% direct:12.2% skip:68.1%
x264 [info]: final ratefactor: 21.24
x264 [info]: 8x8 transform intra:37.6% inter:53.0%
x264 [info]: direct mvs spatial:99.0% temporal:1.0%
x264 [info]: ref P 92.5% 5.1% 2.4%
x264 [info]: ref B 98.6% 1.1% 0.4%
x264 [info]: SSIM Mean Y:0.9803225
x264 [info]: PSNR Mean Y:42.250 U:41.800 V:42.846 Avg:42.248 Global:41.517 kb/s:14396.29

encoded 217 frames, 1.09 fps, 14397.37 kb/s

./x264 --keyint 15 --min-keyint 1 --vbv-maxrate 29400 --vbv-bufsize 9500 --qpmin 10 --level 4.1 --bframe 2 --b-rdo --bime --weightb --ref 3 --mixed-refs --direct auto --filter 0:0 --bitrate 12000 --stats "H264_12Mbps.log" --qcomp 0.75 --ipratio 1.10 --pbratio 1.33 --analyse "all" --8x8dct --me "hex" --subme 6 --no-fast-pskip --trellis 2 --threads 2 --progress -o ../output.mkv ../blue_sky.yuv 1920x1080
x264 [warning]: width or height not divisible by 16 (1920x1080), compression will suffer.
x264 [info]: using cpu capabilities MMX MMXEXT SSE SSE2
x264 [info]: slice I:15 Avg QP:21.73 size:285361 PSNR Mean Y:43.50 U:43.79 V:44.75 Avg:43.72 Global:43.27
x264 [info]: slice P:98 Avg QP:21.84 size: 94601 PSNR Mean Y:42.68 U:42.37 V:43.18 Avg:42.69 Global:41.99
x264 [info]: slice B:104 Avg QP:24.14 size: 20223 PSNR Mean Y:41.66 U:40.99 V:42.27 Avg:41.62 Global:40.92
x264 [info]: mb I I16..4: 24.7% 41.8% 33.5%
x264 [info]: mb P I16..4: 7.1% 3.9% 0.9% P16..4: 44.6% 20.5% 7.8% 0.7% 0.3% skip:14.2%
x264 [info]: mb B I16..4: 0.1% 0.1% 0.0% B16..8: 17.8% 0.5% 1.2% direct:11.6% skip:68.5%
x264 [info]: final ratefactor: 21.23
x264 [info]: 8x8 transform intra:38.0% inter:53.1%
x264 [info]: direct mvs spatial:97.1% temporal:2.9%
x264 [info]: ref P 92.2% 5.4% 2.4%
x264 [info]: ref B 98.5% 1.1% 0.4%
x264 [info]: SSIM Mean Y:0.9803406
x264 [info]: PSNR Mean Y:42.247 U:41.805 V:42.851 Avg:42.248 Global:41.515 kb/s:14428.12

encoded 217 frames, 1.93 fps, 14430.21 kb/s


./x264 --keyint 15 --min-keyint 1 --vbv-maxrate 29400 --vbv-bufsize 9500 --qpmin 10 --level 4.1 --bframe 2 --b-rdo --bime --weightb --ref 3 --mixed-refs --direct auto --filter 0:0 --bitrate 12000 --stats "H264_12Mbps.log" --qcomp 0.75 --ipratio 1.10 --pbratio 1.33 --analyse "all" --8x8dct --me "hex" --subme 6 --no-fast-pskip --trellis 2 --threads 4 --progress -o ../output.264 ../blue_sky.yuv 1920x1080
x264 [warning]: width or height not divisible by 16 (1920x1080), compression will suffer.
x264 [info]: using cpu capabilities MMX MMXEXT SSE SSE2
x264 [info]: slice I:15 Avg QP:21.73 size:287253 PSNR Mean Y:43.53 U:43.83 V:44.79 Avg:43.76 Global:43.35
x264 [info]: slice P:98 Avg QP:21.87 size: 94448 PSNR Mean Y:42.67 U:42.35 V:43.16 Avg:42.67 Global:41.97
x264 [info]: slice B:104 Avg QP:24.18 size: 20008 PSNR Mean Y:41.64 U:40.96 V:42.25 Avg:41.60 Global:40.91
x264 [info]: mb I I16..4: 24.4% 42.3% 33.3%
x264 [info]: mb P I16..4: 7.1% 4.0% 0.9% P16..4: 45.1% 20.5% 7.8% 0.7% 0.3% skip:13.8%
x264 [info]: mb B I16..4: 0.1% 0.1% 0.0% B16..8: 17.8% 0.5% 1.2% direct:11.3% skip:68.9%
x264 [info]: final ratefactor: 21.23
x264 [info]: 8x8 transform intra:38.4% inter:53.5%
x264 [info]: direct mvs spatial:96.2% temporal:3.8%
x264 [info]: ref P 91.9% 5.6% 2.5%
x264 [info]: ref B 98.4% 1.2% 0.5%
x264 [info]: SSIM Mean Y:0.9803155
x264 [info]: PSNR Mean Y:42.234 U:41.784 V:42.836 Avg:42.233 Global:41.503 kb/s:14419.88

encoded 217 frames, 3.23 fps, 14423.97 kb/s


./x264 --keyint 15 --min-keyint 1 --vbv-maxrate 29400 --vbv-bufsize 9500 --qpmin 10 --level 4.1 --bframe 2 --b-rdo --bime --weightb --ref 3 --mixed-refs --direct auto --filter 0:0 --bitrate 12000 --stats "H264_12Mbps.log" --qcomp 0.75 --ipratio 1.10 --pbratio 1.33 --analyse "all" --8x8dct --me "hex" --subme 6 --no-fast-pskip --trellis 2 --threads 8 --progress -o ../output.mkv ../blue_sky.yuv 1920x1080
x264 [warning]: width or height not divisible by 16 (1920x1080), compression will suffer.
x264 [info]: using cpu capabilities MMX MMXEXT SSE SSE2
x264 [info]: slice I:15 Avg QP:21.73 size:287253 PSNR Mean Y:43.53 U:43.83 V:44.79 Avg:43.76 Global:43.35
x264 [info]: slice P:98 Avg QP:21.87 size: 94448 PSNR Mean Y:42.67 U:42.35 V:43.16 Avg:42.67 Global:41.97
x264 [info]: slice B:104 Avg QP:24.18 size: 20009 PSNR Mean Y:41.64 U:40.96 V:42.25 Avg:41.60 Global:40.91
x264 [info]: mb I I16..4: 24.4% 42.3% 33.3%
x264 [info]: mb P I16..4: 7.1% 4.0% 0.9% P16..4: 45.1% 20.5% 7.8% 0.7% 0.3% skip:13.8%
x264 [info]: mb B I16..4: 0.1% 0.1% 0.0% B16..8: 17.8% 0.5% 1.3% direct:11.3% skip:68.9%
x264 [info]: final ratefactor: 21.23
x264 [info]: 8x8 transform intra:38.4% inter:53.5%
x264 [info]: direct mvs spatial:96.2% temporal:3.8%
x264 [info]: ref P 91.9% 5.6% 2.5%
x264 [info]: ref B 98.4% 1.2% 0.5%
x264 [info]: SSIM Mean Y:0.9803156
x264 [info]: PSNR Mean Y:42.234 U:41.784 V:42.836 Avg:42.233 Global:41.503 kb/s:14419.89

encoded 217 frames, 3.25 fps, 14423.98 kb/s


./x264 --keyint 15 --min-keyint 1 --vbv-maxrate 29400 --vbv-bufsize 9500 --qpmin 10 --level 4.1 --bframe 2 --b-rdo --bime --weightb --ref 3 --mixed-refs --direct auto --filter 0:0 --bitrate 12000 --stats "H264_12Mbps.log" --qcomp 0.75 --ipratio 1.10 --pbratio 1.33 --analyse "all" --8x8dct --me "umh" --subme 7 --no-fast-pskip --trellis 2 --threads 4 --progress -o ../output.264 ../blue_sky.yuv 1920x1080
x264 [warning]: width or height not divisible by 16 (1920x1080), compression will suffer.
x264 [info]: using cpu capabilities MMX MMXEXT SSE SSE2
x264 [info]: slice I:15 Avg QP:21.67 size:283916 PSNR Mean Y:43.64 U:43.93 V:44.88 Avg:43.86 Global:43.45
x264 [info]: slice P:98 Avg QP:21.81 size: 94561 PSNR Mean Y:42.72 U:42.40 V:43.20 Avg:42.72 Global:42.02
x264 [info]: slice B:104 Avg QP:24.05 size: 20711 PSNR Mean Y:41.69 U:41.03 V:42.30 Avg:41.66 Global:40.96
x264 [info]: mb I I16..4: 26.9% 40.7% 32.4%
x264 [info]: mb P I16..4: 6.8% 3.9% 1.0% P16..4: 41.6% 20.7% 7.9% 0.8% 0.3% skip:17.1%
x264 [info]: mb B I16..4: 0.1% 0.1% 0.0% B16..8: 17.7% 0.5% 1.3% direct:11.9% skip:68.4%
x264 [info]: final ratefactor: 21.19
x264 [info]: 8x8 transform intra:37.6% inter:53.2%
x264 [info]: direct mvs spatial:98.1% temporal:1.9%
x264 [info]: ref P 90.7% 6.4% 2.9%
x264 [info]: ref B 98.3% 1.2% 0.5%
x264 [info]: SSIM Mean Y:0.9803996
x264 [info]: PSNR Mean Y:42.291 U:41.849 V:42.883 Avg:42.290 Global:41.556 kb/s:14451.28

encoded 217 frames, 2.55 fps, 14455.37 kb/s

[Doom9]

Well, the first QuadFX results are in: http://www.hothardware.com/viewarticle.aspx?page=15&articleid=911
AMD has their work cut out for them if they aim to recapture the performance crown in 2007.

[Sharktooth]

in many other sites reviews, even if intel quadcore is faster, the AMD 4x4 scaled better.
i wonder if hothardware really tested it or just put up a quick and inconclusive review based only on known facts...

[Inventive Software]

I'll wait for PC Format to gimme a good review, cause they're really thorough and compare things well. QuadFX I can see used in encoding and servers, but they're gonna have to sort the price to keep it mainstream and not just with the enthusiasts IMO.

[Doom9]

@Sharktooth & all the other AMD fanboys out there: I urge you now to stick with the numbers.. I like AMD, too, and I had high hopes for an AMD AM2 chip to replace my socket 939 X2.. but intel's chip simply was faster. I haven't bought any intel chip since the first half of 2000, and I've had a pro AMD bias ever since my first PC (a 286, 12 Mhz). Still, there's such a thing as being graceful in defeat. Intel was taking a beating for years (the fact that many people didn't get that clock frequence didn't equal performance did take a long time to get through.. despite AMD having a great run with the whole Athlon line since its inception), now the tables have turned, until at least the introduction of a real quad core on a single die, scheduled for summer 2007.
Until then, ven theoretical advantages mean nothing when they don't translate into real world gains (e.g. scaling (http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2879&p=7)).
In the end, what do you care more about: that your architecture may be technically better and have some theoretical advantage, or that your rig can churn out a couple more frames per second while encoding? I may not like intel, but I'm a practical guy.. I don't care about synthetic benchmarks, not about games (their performance is in a much larger percentage influenced by GFX cards anyway), but about FPS when encoding video. Oh, and the whole thing needs to fit into a barebone (good luck trying that with two sockets ;)
And just to rule out any intel bias, consider the numbers of encoders that have not undergone serious intel optimizations (like often seen with commercial software). Even the XviD and Lame numbers currently look favorable for intel. Heck, my E6700 lacks almost nothing behind 4 AMD cores running 333Mhz faster (http://www.xbitlabs.com/articles/cpu/display/amd-quad-fx_9.html)

[Sharktooth]

Oh, i didnt say AMD is faster. I was criticizing a part of the hothardware review where they said Intel scales better (which is false).
There's another thread with linked reviews (http://forum.doom9.org/showthread.php?t=118850) and almost all of them show the contrary. So in a heavy SMP configuration (let's say something like 16 cores) AMD is still faster.

http://www.hwupgrade.it/articoli/cpu/1611/megatask_1_perc.png

http://www.hwupgrade.it/articoli/cpu/1611/megatask_2_perc.png

http://www.hwupgrade.it/articoli/cpu/1611/megatask_3_perc.png

The graphs are showing the scaling efficiency (not the pure performance) in heavy multitasking operations (megatasking).

For desktops and small servers Intel CPUs are undoublty faster at this time.
However the AMD true quad core is not so far:
http://www.amd.com/us-en/Corporate/VirtualPressRoom/0,,51_104_543~114496,00.html
http://www.amd.com/quadcoredemo

[Doom9]

And they were doing what exactly during those tests? And what is used as a reference? And is the site decidedly non biased in how it sets up the tests? What is used as a reference? Those graphs make no sense to me.. in a scaling test you're supposed to take a reference, then show how many percent performance improves by adding cores. You can set this up using either the dual core, or a single core as reference.
Suppose a single core was used here (I assume that because because there's a dual core chip in all graphs and it's not set to 0%).. then how much sense does a test make where adding a second core reduces performance by 30% And how realistic is a multitasking scenario where using the most efficient dual core architecture on the market today only adds a 4% performance gain? I mean, something is really fishy here.

Now look at Anand's statistics: they take the dual core chip of manufacturer a running at frequency x, and compare it with the same chip but with double the number of cores.. If you know what the test involves, that's how you can reasonably compare scaling. I'd do things the same if I were into hardware benchmarking (and you might recall me posting some x2 numbers when I first got the chip, comparing on how x264 fares using one and two threads.. it's a simplified version of this as I didn't physically switch off a core.. but it goes towards what a reasonable comparison should be).

With the graphs and no information surrounding I have no option but to conclude that they are incorrect and misleading. They don't live up to any scrutiny whatsoever.

And speaking of heavy multitasking scenarios (which are essentially server scenarios), other benchmarks that show AMD's platform to almost continually lose out, still conclude that in scenarios you rarely find on the desktop, AMD's solution scales better (http://www.xbitlabs.com/articles/cpu/display/amd-quad-fx_12.html). But if it takes a considerable number of CPU intensive processes, it's not realistic for any desktop or workstation.. you don't have 6 instances of x264 running simultaneously. At best, it makes sense to run as many instances as you have cores. And, being a dual socket solution, when we're talking servers you can compare it to a dual dualcore-xeon solution (and then intel has the dual quadcore xeon too ;) ). QuadFX is trying to push server technology (which is well proven - we all know it works and that it's efficient) into desktops... which is expensive, bulky (try putting that into a barebone), and uses a load of power (P4 dualcore anyone?).

AMD's mainstream Quadcore solution is half a year away, but if I got it right, it will require a new platform. Not a problem for somebody like me who swaps out PCs every year, but not everybody is like that.. and if they can plug in an intel Quadcore into the existing mainboard... well you can finish that sentence for me.

I tend to think the success of the Athlon product line, and the ensuing inroads into the market (heck, even Dell sells AMD bases systems today), AMD has in many ways become a little like intel.. innovation has gone down (AM2? Not really much of an improvement), prices up (remember the days where buying AMD meant a cheaper system?), and just like Intel quickly cobbled together two P4 cores to be the first with a dualcore chipo (heck they did it again for quadcore, but the much better architecture means much less of the drawbacks the previous generation of "two chips on one die" had), QuadFX is much more of a "me too" product. Not only does it fail to beat intel, it also reverses much of the progress made when more efficient and less power hungry CPU architectures were introduced.

That said, I'm looking forward to real quad core chips coming in 2007 and another round in the eternal battle and hope that sooner or later the performance crown will be passed again.. it makes the whole game a lot more interesting and it is good for consumers if both players are forced to compete at a high level.

[Sharktooth]

QuadFX is a platform. It will be compatible with the next gen AMD processors such as the quad cores FX processors.
It's for enthusiasts only and proposes an upgrade path for the future CPUs without having to renew all the system, but only the processor(s).
However the review where i took the graphs is quite unbiased.
It's from hwupdate but it's in italian language.
Heres the link: http://www.hwupgrade.it/articoli/cpu/1611/quad-fx-la-ricetta-amd-per-i-quattro-core_index.html
The megatasking tests were made running multiple applications and taking the time to complete all operations vs the time to complete running those applications alone. The gain or loose is reported on the graph. There is no reference coz it's not necessary.
So basically it is an multiprocessor architecture scaling efficiency test.
In the first graph were tested those applications: Mainconcept H.264 Encoder, DVD Shrink 3.2, Sony Vegas+DVD 7.0b and Povray 3.7
All apps are multithreaded apps... and so on.
I have no time to translate it...

[Doom9]

In the first graph were tested those applications: Mainconcept H.264 Encoder, DVD Shrink 3.2, Sony Vegas+DVD 7.0b and Povray 3.7And you'll have those running in parallel how many times in your life? Sorry, just as I suspected those graphs are pretty darned pointless. If you're going to benchmark a scenario that you won't encounter on a desktop, better run a server benchmark, at least that is a scenario that is often ocurred in the real world (lots of servers out there that are essentially desktop PCs). These three tests (the one most applicable to this place btw showing both solutions head to head) have more threads than cores so we have the applications all trying to get a chunk of the resources. Wouldn't a scenario like that be as much of a test of the scheduler in Windows as of the CPU itself and hence graphs of the CPU usage would have to be presented as well? With more threads fighting for the attention of the cores, balancing those to max out the use of each core becomes a key task.. I wouldn't be surprised if the average CPU utilization in these unrealistic "megatasking" tests were higher on the AMD platform than on the intel platform. That alone is not an indicator of efficiency though.. you'd then have to descend into an analysis why one platform is used more efficiently than the other.. is it really hypertransport versus fsb or are there other forces at play? If you want to claim a more efficient architecture, do it in a fashion that stands up to scientific scrutiny or leave it be.


And even in your handpicked site.. if you look at realistic scenarios, intel (http://www.hwupgrade.it/articoli/cpu/1611/quad-fx-la-ricetta-amd-per-i-quattro-core_8.html) wins hands down. Even the graphs from the same page you got the ones from you posted shows another light.
I've had more than enough maths during my 5 year master studies to know if you pick the proper scenario you can pretty much show anything.. your graphs are a perfect example of that. you take an unrealistic scenario, then look at relative numbers and finally you can make your point about efficiency. Little does the reader realize that these numbers are essentially meaningless and that if you look at actual executing time, even with 333MHz less, the chip that doesn't look so good in those graphs still performs better (just think about how it would perform with another 333mhz under the hood ;)
And reading the conclusions at hardware upgrade - I can't help but think this is an AMD fansite. All the talk about what may be next year, all the effort that is put into painting a solution that is currently inferior and unlikely to make it to the mainstream into something better... I'm sorry, are we evaluating what's available now or speculating about what may be? This sadly reminds me of THG in their best "intel rules" days.

[Sharktooth]

well, in no way i doubted intel has superior performance right now...
but affirming it has superior scaling is false.
you cant say it's a benchmark for the windows scheduler coz both platform run the same OS. The AMD platform was simply more efficient (but not faster!) than the Intel one.
That means having massive multitasking/multithreading on AMD platforms with a large number of processors could lead to more performance even if the AMD CPUs are less performing (taken 1 vs 1) than Intel ones.
It is a confirmation of the better "base" architecture (direct interconnect, hyper-transport, integrated memory controller etc) AMD has and the addition of 128bit SSE and the other cool stuff in K8L will surely make AMD CPU faster than the actual intel tech.
This is not speculation but logic facts (better architecture with the same load of features = better performance).

[Doom9]

There you go again with "what is to be". The PC biz is moving so fast that it makes little sense to speculate of what may be... you need a box now, buy now. You need one in half a year, check the situation then. You don't think intels R&D is going to rest for half a year to allow AMD to play catchup, do you? Intel is already taping out their 45 nm Penryn chips (no word yet whether the quadcore variety will put all four cores on the same die though)
"Your" benchmarks are handpicked by an AMD fansite with the aim of affirming that AMD's architecture is superior - regardless of any common sense that would tell you the scenario is completely out there. And the whole conclusion... you won't find the same effort to sweettalk a flat out loss at any non biased site. I guess I should search for come AM2 benchmarks to see how they managed to turn the disappointment into something to be excited about.
If you run a server on a QuadFX box, it is likely to perform better than an intel Quad box (taking the QX6700 chip.).. So great, this forum would run better on a QuadFX box. But, then it's the choice between using consumer hardware for a server, or actual server hardware (with all the obvious benefits (reliability, dual of pretty much everything to take over if a component fails) and drawbacks (expensive beyond good reason). Compare a QuadFX platform to a regular dual slot Operon platform. What is the difference? QuadFX is a "consumerized" version.. it uses consumer RAM, a consumer chipset and high end consumer chips.. but the architecture itself is a server architecture. So if you're looking for a consumer server, it's your best choice .. if you're looking for an encoding box, you can get more FPS for your buck - even more FPS for less of your bucks. The answer to the original question has to be "get an intel quadcore".
You can ask the question again in late summer 2007 when the situation is likely to have changed.

[Sharktooth]

Well, you fail to see the potential of QuadFX. I see it as a competitor (even if actual AMD CPUs are not completely up to the task) for kentsfield but it's also an upgrade path for the future.
As i said it is a platform and it ensures you will be able to "upgrade" your system with the next generation CPUs, up to 2 quad-cores on the same MB for a total of 8 cores for consumer PCs and workstations.
Ok, you may argue no one will need that "beast"... but IMHO since the trend is to rise the number of cores (intel is prepearing a 80 cores CPU for 2010 - TeraScale (http://www.pcper.com/article.php?aid=302)) instead of the clockspeed, it will be no waste of money...

However if you think hwupgrade.it is a fansite then look here:
http://enthusiast.hardocp.com/article.html?art=MTIzMyw5LCxoZW50aHVzaWFzdA==
http://enthusiast.hardocp.com/article.html?art=MTIzMywxMCwsaGVudGh1c2lhc3Q=
As you can see the efficiency of the QuadFX is much better in multitasking operations and it's up to the competition with kentsfield.

[dragongodz]

i was just going to post the hardocp review but sharktooth got in first.

instead i will add this one
http://www.firingsquad.com/hardware/amd_athlon_64_fx_74_4x4/

my personal thoughts on this is that its not really meant to beat Intel but to reduce Intels bragging rights to how much faster they currently are. its already known that AMDs true quad core cpus wont be ready until roughly the middle of next year by which time of course Intel should have its true quad core cpu out. i think that will be a more interesting time because both should then be trying to develop them further, rather than these 2 quad solutions.

[Doom9]

Well, the only hardcop benchmark where AMD beats intel is once again the most unrealistic. TMPG, WME and Winrar? How often are you going to encounter that in the real world? I've had a dual core system since the X2 first came out and for most of the time, my cores pretty much idle around.. the only time they ever get really busy is when I'm encoding video but there's generally not anything else very CPU consuming running at the same time. The most extreme I've ever did was play a game while encoding video (and then encoding just takes a real hit as you don't want to run the encoding process at anything but low priority to ensure there are always resources for the game).

We have to be realistic, even with two or four cores, most of the time you're not going to max out the CPU. And us humans (especially guys) basically being unable to properly juggle more than one complex task simultaneously, we do the same in computing.

but IMHO since the trend is to rise the number of cores (intel is prepearing a 80 cores CPU for 2010 - TeraScale) instead of the clockspeed, it will be no waste of money...dual socket has been around since the turn of the millenium.. it never made so much as a dent for non server applications. I dare venture it's not going to in the future either. If you need an 8 or 16 way system, most likely it's going to be some kind of a server, where size, noise and power consumtion and price aren't the most pressing concerns and where you run applications that are made to scale (databases, webservers, etc.). Consumer apps, at least currently and for the forseeable future, don't fall into that category. For a consumer, the one socket solution will remain the viable one. The past 6 years have proven that quite well (while most servers came with two sockets during that time already.. it's just a whole different market): it takes x cores on a single die to make it attractive for more than a few of diehard users. I'm convinced 8 cores will only ever be a mainstream thing once they all go into the same socket.

[foxyshadis]

Actually, I remember when you could get PII-based Xeon dual-sockets... so it's definitely beyond the millenium mark. ;) (About 98 or so.)

HardOCP is an ultimate amd fansite, and their review was a bit of a farce before shoring it up a bit (see the forum drama surrounding it), but not all of them are. Doom9's still right, marketing a high-end workstation chipset to gamers who don't know what to do with 4 cores and high-end raid is a little silly. (Intel's quad is in the same boat.) It's a rebranding that lets them shift the opteron chipset to the value end of the market, so they can make a little more off it when the new K8L chipset comes out. It's not quite as useless to encoders, but it's still just too expensive to compete with anything else in the consumer space, and almost requires 2003 server or vista for top performance (besides the insane multitasking), whereas alternatives don't.

4x4 may be an upgrade path, but it would be an insanely expensive one. The only thing that will make it worthwhile is the ATI coprocessor, which is years down the road and shouldn't affect your decision to buy today.

[woah!]

i have the QX6700 and to get a quadcore system,i just bought the chip and stuck it in a single socket mobo.

that amd setup is crazy when you need to fill all the ram slots (4gb) , then 2 power hungry cpu's bolted to a dual slot mobo.

you need 500W of power just for that, and if you throw in a 8800 gcard you will need 700W!

amd should just ditch this version of the quadcore and wait for the K8L version on 65nm so it isnt such a power hog.

i was waiting for the 4x4 until i saw all the extra's needed.

oh but this qx6700 is fast tho...

[woah!]

thought i would show some results of this quadcore aswell with a encode from 1920x1080i to 720x304 with a 2:35:1 format :




H:\>"H:\602.x264.exe" --pass 1 --bitrate 1975 --stats ".stats"
--cqmfile "H:\M4G_High_Detail_AVC.cfg" --bframes 3 --filter -1:-1 --subme 1 --me
range 12 --analyse none --me dia --threads 3 --colormatrix bt709 --progress --no
-psnr --output NUL "H:\2.35.1.avs"
avis [info]: 720x304 @ 23.98 fps (133508 frames)
x264 [info]: using cpu capabilities MMX MMXEXT SSE SSE2
x264 [info]: slice I:1215 Avg QP:11.74 size: 32706ta 0:00:00
x264 [info]: slice P:50604 Avg QP:12.80 size: 15509
x264 [info]: slice B:81689 Avg QP:14.20 size: 6633
x264 [info]: mb I I16..4: 21.3% 0.0% 78.7%
x264 [info]: mb P I16..4: 24.9% 0.0% 0.0% P16..4: 68.1% 0.0% 0.0% 0.0% 0
.0% skip: 7.0%
x264 [info]: mb B I16..4: 2.1% 0.0% 0.0% B16..8: 38.1% 0.0% 0.0% direct:
37.6% skip:22.2%
x264 [info]: final ratefactor: 13.89
x264 [info]: SSIM Mean Y:0.9892260
x264 [info]: kb/s:1963.0

encoded 133508 frames, 37.01 fps, 1965.10 kb/s

H:\>"H:\602.x264.exe" --pass 2 --bitrate 1975 --stats ".stats"
--cqmfile "H:\M4G_High_Detail_AVC.cfg" --ref 5 --mixed-refs --no-fast-pskip --me
range 12 --bframes 3 --bime --weightb --direct auto --filter -1,-1 --subme 6 --b
-rdo --analyse all --me umh --threads 3 --colormatrix bt709 --thread-input --pro
gress --no-psnr --output "H:\VIDEO.MP4" "H:\2.35.1.avs"
avis [info]: 720x304 @ 23.98 fps (133508 frames)
x264 [info]: using cpu capabilities MMX MMXEXT SSE SSE2
mp4 [info]: initial delay 1001 (scale 24000)
x264 [info]: slice I:1215 Avg QP:11.84 size: 32026ta 0:00:00
x264 [info]: slice P:50604 Avg QP:12.33 size: 15813
x264 [info]: slice B:81689 Avg QP:13.98 size: 6556
x264 [info]: mb I I16..4: 39.0% 0.0% 61.0%
x264 [info]: mb P I16..4: 11.1% 0.0% 18.0% P16..4: 26.8% 20.8% 12.4% 1.8% 2
.1% skip: 7.1%
x264 [info]: mb B I16..4: 1.0% 0.0% 2.1% B16..8: 40.9% 7.1% 16.3% direct:
5.2% skip:27.5%
x264 [info]: direct mvs spatial:0.0% temporal:100.0%
x264 [info]: ref P 71.0% 13.2% 8.1% 4.1% 3.6%
x264 [info]: ref B 83.2% 8.8% 3.9% 2.1% 1.9%
x264 [info]: SSIM Mean Y:0.9918634
x264 [info]: kb/s:1975.0

encoded 133508 frames, 27.24 fps, 1977.08 kb/s

2min clip of resulting encode is here:

http://w11.easy-share.com/700528.html

[Doom9]

--threads 3? wouldn't you rather go for --threads 4 seeing that x264 doesn't scale linearly (meaning it doesn't max out two cores, and thus is likely not to max out the 4 cores).. so you'd have one heavy duty thread for each core and one that can use up remaining cycles. I think it would at least be worth a try to see if it improves CPU usage.

You could also do a nice scaling test by editing your boot.ini and appending /numproc=2 .. that gets you an E6700 essentially.

[dragongodz]

amd should just ditch this version of the quadcore and wait for the K8L
huh ? is somebody making you buy it ? infact is it effecting you in any way, shape or form ? no ? then why such a ridiculous statement ?
to anybody that bothers to put 5 seconds thought in to it they would know AMD only really released this to reduce Intels bragging rights about how much faster their current cpus are.
fact is the average joe doesnt have the cash for these high end cpus so it is really a non-issue. its all about PR to raise the profile of both companies, to hopefully generate sales of their more mainstream priced cpus.

and if you throw in a 8800 gcard you will need 700W!
and if you have the money to spend on all this new expensive hardware then you would either already have a large psu or be getting one anyway. reading places like nvnews forum you can easily find people with 700w-1000w psus before they bought an 8800.

thought i would show some results of this quadcore aswell
which is rather meaningless without being able to compare the same source + settings etc on other cpus.

[woah!]

well you would think threads 4 would be better than 3 huh, but i found 3 runs faster than 4 for some weird reason? with the thread-input aswell tho techically i am running 4 threads.

that encode ran at about 85-90% cpu power.


i can do a scaling test for sure :)

[woah!]

huh ? is somebody making you buy it ? infact is it effecting you in any way, shape or form ? no ? then why such a ridiculous statement ?
to anybody that bothers to put 5 seconds thought in to it they would know AMD only really released this to reduce Intels bragging rights about how much faster their current cpus are.
fact is the average joe doesnt have the cash for these high end cpus so it is really a non-issue. its all about PR to raise the profile of both companies, to hopefully generate sales of their more mainstream priced cpus.


and if you have the money to spend on all this new expensive hardware then you would either already have a large psu or be getting one anyway. reading places like nvnews forum you can easily find people with 700w-1000w psus before they bought an 8800.


which is rather meaningless without being able to compare the same source + settings etc on other cpus.


got out the wrong side of the bed or something today ...

you think its ok to just burn a Kilowatt of juice for a high end system then? i have had nothing but amd systems for the last 6yrs. but i think amd should have just waited until K8L.


as for a compare results your right i will make up a package if you want to compare thats not a problem.

[woah!]

heres a quick scaling screenshot from 1 core to 4 cores:

seems the 2nd pass scales very well but the first pass is not so, but at 4 cores it was sitting around 55% cpu power on the first pass?

anyways make of what you will..

1 CORE:
http://images.dr3vil.com/files2/default/1core.jpg

2 CORE:
http://images.dr3vil.com/files2/default/2cores.jpg

4 CORE:
http://images.dr3vil.com/files2/default/4cores.jpg

[dragongodz]

got out the wrong side of the bed or something today
hmm did i rant ? did i call you names or accuse you of things ? seems your the touchy one reading things that are not there.

you think its ok to just burn a Kilowatt of juice for a high end system then?
show me exactly where i said or even suggested that. i gave the reason for why AMD released this even though its not going to be taken up by anyone except extremests. either you understand what PR is or you dont.

but i think amd should have just waited until K8L.
and if you bother to go further back you will see i already said that that will be the real interesting time and not these "quad" solutions. however why should AMD give Intel such a big lead time for lots of PR about how fast their fastest cpus are compared to anything AMD have out ? that would not be very good business sense.

as for a compare results your right i will make up a package if you want to compare thats not a problem.
yes that would be much more informative. we already have a few websites, with tests, posted in this thread but more is always interesting. especially when the comparisons are done with software used by those on this forum.

[Inventive Software]

'Scusey for being pedantic, but you specified "--threads 3", in all iterations. Any particular reason? I know you said it was faster, but I don't believe you. ;)

And can you provide screenies with "--threads 1" for 1 core, "--threads 2" for 2 cores, and "--threads 4" for 4 cores to prove me wrong? :D

[Sharktooth]

im not even sure x264 accetps threads option values different than 1, 2 or 4...
however higher CPU usage doesnt mean faster encoding at all...

[Doom9]

Any particular reason you're always using --threads 3 even when the number of cores is lower? I seem to recall some numbers showing a slight degradation if more threads than cores are being used. And why is there a -thread-input only in second pass? It's actually implicit for --threads X where X > 1.

[Sharktooth]

yes. having threads greater than the cores (or virtual cores) will slightly degrade performance due to overhead.
However in some particular cases (im not saying it's the x264 case) it may still give a performance boost.

[Romario]

http://www.cdrinfo.com/Sections/Reviews/Print.aspx?ArticleId=19231

Look closely, because they used x264 via MeGUI for this test.

[Sharktooth]

They even used my x264 r602 build... :)
Nice stuff...

[Blue_MiSfit]

Amazing... Usually when they benchmark video, hardware sites use Xmpeg or AutoGK (if you're lucky). How behind the times :)

A proper x264 benchmark is refreshing. Maybe if they added MT AviSynth things would speed up a bit more!

The XviD benchmark could have used the newer builds - they support more threads right?


just realized they are doing HD encoding. Wow.


~MiSfit

[Romario]

Well, xmpeg.net said that Xmpeg 6.0 should be released within several weeks from now.

www.xmpeg.net


BTW, where can I find MT AviSynth?

[Blue_MiSfit]

MT AviSynth is in the Avisynth Discussion forum - its near the top right now.
http://forum.doom9.org/showthread.php?t=94996
Fun stuff - replace your AviSynth.dll and install MT.dll in plugins folder... then add SetMTMode(x,y) as first line of all scripts, where x can be between 2 and 5, 2 being the fastest, 5 being the most compatible, and y being the number of threads, 0 being auto.

Good stuff. Really helps if you're feeding something like CCE or have a filter heavy script that is bottlenecking the encoder. Keep in mind a lot of cool plugins don't work, like MVTools, and fft3dgpu. fft3dfilter does work, however, and may be faster using both cores than fft3dgpu as pure GPU. Depends.

~MiSfit

[morph166955]

so im curious if any of this has changed in the 2-3 months since you guys last reviewed this. I'm currently considering upgrading my freebsd workhorse/server to a dual xeon E5320 1.86ghz quad-core setup primarily to do x264 (but also xvid) encoding. I currently have a prescott 3.2HT doing my compiling and frankly im getting a little disgruntled with 4-5fps i get now. The big question for me is "is it worth it?" The other consideration here is that I'm going to be selling my current system to afford this one and in terms of how much i can get for it, its pretty obvious that doing this now is going to get me more money then in a few months from now.

Any suggestions?

[morph166955]

anyone? im looking to buy this thing with in the next 7-10 days so if anyones got any suggestions or info that has changed since this post was done a few months ago that would be amazing. either that or if someone could confirm that this post is still current and good with its info that would be good too.

[jeffy]

so im curious if any of this has changed in the 2-3 months since you guys last reviewed this. I'm currently considering upgrading my freebsd workhorse/server to a dual xeon E5320 1.86ghz quad-core setup primarily to do x264 (but also xvid) encoding. I currently have a prescott 3.2HT doing my compiling and frankly im getting a little disgruntled with 4-5fps i get now. The big question for me is "is it worth it?" The other consideration here is that I'm going to be selling my current system to afford this one and in terms of how much i can get for it, its pretty obvious that doing this now is going to get me more money then in a few months from now.

Any suggestions?
Dual Xeon E5320 as of now 2x $495 at NewEgg=$990
But according to TechReport.com, there are Q3/July price cuts bringing this Dual Xeon setup to 2x $256=$512
or you could get 2x E5345 2.33 GHz for $910 at that time, possibly.

Not mentioning X3220 2.40 GHz for $266...
http://www.dailytech.com/Intel+Readies+New+Xeons+and+Price+Cuts/article6493.htm

[foxyshadis]

Is it even possible to have enough memory bandwidth on an Intel platform to utilize 8 cores effectively, even for something like video processing/encoding? 8mb cache doesn't go far for HD. You probably need some killer memory modules for that.

As someone once said, it's never the right time to upgrade, wait six months and prices will always drop. ;) You'll actually probably get nearly as much out of your current system now as in six months, though; bleeding edge prices drop much faster than old systems, so concentrate on how much you need it rather than how much you'll get from what you have now.

[morph166955]

well the other reason for selling the server is because im planning on working it into a deal I have going right now that wont be around in a few months. Any idea what kind of a speed boost i'd see if i waited for the E5345 vs the E5320 (even a guess is good)?

@foxy...can you go more in depth about the intel/memory bandwidth thing you mentioned?

[jeffy]

Can't post E5345 vs the E5320, but can point you to 5160 vs. 5355 (both DUAL = 8 cores, unfortunately used only 4)
Windows Media Encoder x64 Edition
http://techreport.com/reviews/2006q4/xeon-vs-opteron/index.x?pg=9

Not that much of a difference :confused:

But this is E5345 vs E5320, again WMV & Lame MT
http://www.gamepc.com/labs/view_content.asp?id=x5355&page=10
Better :)

Memory bandwidth scaling
http://www.gamepc.com/labs/view_content.asp?id=x5355&page=5