What Machine should I buy to perform RV9 - EHQ=80 realtime encoding ?

[ookzDVD]

As the title said.

I plan to buy new machine,
with plain encoding (no filter),
what processor is enough to do the realtime RV9 EHQ=80 encoding ?

PS. With my current machine (Athlon +1700) the encoding time is
about 3.5-4x the movie duration :(

So I need +1700 x 4 = Athlon +6800 ?
Hmm.... When that machine will be available ? :(

[General Lee D. Mented]

Dual Opteron 246s? =)

[scharfis_brain]

which resolution do you try to encode?

maybe lowering that would help.

[Sirber]

Just buying a P4 will boost your speed by 25%, and DualChannel DDR is a must. RV9 is dependant of SSE2, and in high memory bandwith also :) After buying that, you could encode at EHQ=90... :D

[iwod]

The best you could get now is a Pentium 4 Extreme Edition 3.2 Ghz.

And with 2x 256 DDR 400 SDram ( i.e Dual Channel ) you should be done.

However karl said eailer that he want to do a ultra super douper complex algorthium for EHQ 100 to increase the quality a litte bit more but it would demand even more CPU power compare to EHQ 80.

Therefore i suggest you to wait until next intel Pentium Processor the prescot.............. May be Helix will even optimse their encoder more with SSE3...........

Or you could get a Athlon 64 cpu where it support SSE2.........

[Sirber]

I'm bored of AMD being behind for Multimedia CPU extentions. I'll get a P4 next time.

[karl_lillevold]

an Athlon 1700 is really only something like a 1.4 GHz, so if you get a P4 2.8 GHz, you get 2X * 1.25 (25% extra for SSE2) = 2.5 speedup. If your encodes in the past were 4X realtime, they will be 4/2.5 = 1.6X with a P4 2.8 GHz. Not quite there, but almost. By the way, is 4X including 2 pass or just for 1 pass?

RV9 incl EHQ has already been optimized for SSE3, but not yet released. This speedup should be approx 5% in addition to SSE2, more measurements later.

EDIT: I don't know the effect of Dual DDR 800 MHz memory, but I am guessing it will help quite a bit. Video encoding is very memory access intensive.

[Sharktooth]

Dual Opteron 246. It supports SSE2 and its faster than dual P4 Xeon.

[Sirber]

Opteron is quite expensive...

[Sharktooth]

Then go for an Athlon 64 FX51 or FX53(when available).
64bit codecs will be available very soon...
Memory bandwidth shouldnt be a problem since the FX has a dual channel memory controller.

[karl_lillevold]

re 64bit: we currently have no plans to optimize either encoder or decoder specifically for 64 bit processors, simply because all the functions that would benefit from parallelizations have already been optimized for MMX/SSE/SSE2 which for SSE2 offers up to 128bit wide operations. It will be very interesting to see how RV9 encodes and decodes on Athlon 64 though, now that AMD has included SSE2. Maybe we have to make some changes in places where shift operations assume 32bit, but it should be fairly minimal.

[deXtoRious]

Sirber

Does that mean that your AMD XP+ 2000 (mine too) is a piece of crap?
You know, you got me worried ;)

[Sirber]

With a P4, same speed, I could encode at least 35% faster :(

[slavickas]

Originally posted by karl_lillevold
re 64bit: we currently have no plans to optimize either encoder or decoder specifically for 64 bit processors, simply because all the functions that would benefit from parallelizations have already been optimized for MMX/SSE/SSE2 which for SSE2 offers up to 128bit wide operations. It will be very interesting to see how RV9 encodes and decodes on Athlon 64 though, now that AMD has included SSE2. Maybe we have to make some changes in places where shift operations assume 32bit, but it should be fairly minimal.

but amd64 has twice more xmm registers

btw Karl i think setting via registry patternAdaptivity=3 probably won't work (have tested only on 1 klip), via job file seems to work

[TheXung]

Your best bet is the aforementioned P4 3.2 system. Build yourself a waterchilled cooling system and overclock it. It could get you into the 3.6-4.0 Ghz range. Something else that would make a difference is to use a capturing or encoding program that is multithreaded.

I have heard the SSE2 implementation on the Athlon FX to be described as "stoddy", so take that for what it is worth.

[karl_lillevold]

@slavickas: xmm regs; I did not know that about amd64. Also, I just checked patternAdaptivity via the registry. Works fine here. Make sure you spell it correctly:

Windows Registry Editor Version 5.00

[HKEY_LOCAL_MACHINE\SOFTWARE\RealNetworks\RV9]
"patternAdaptivity"=dword:00000003


@TheXung: multi-threaded encoders; yep, RV9 encoder and decoder are both fully multi-threaded. A true dual system will encode (and decode) about twice as fast as a single CPU. There is currently not much advantage to hyper-threading for the encoder, due to cache contention, but it will be, when Intel increases the cache size. Maybe Prescott already has this, I am not sure.

[ookzDVD]

Originally posted by karl_lillevold
an Athlon 1700 is really only something like a 1.4 GHz, so if you get a P4 2.8 GHz, you get 2X * 1.25 (25% extra for SSE2) = 2.5 speedup. If your encodes in the past were 4X realtime, they will be 4/2.5 = 1.6X with a P4 2.8 GHz. Not quite there, but almost. By the way, is 4X including 2 pass or just for 1 pass?


Thank yuo Karl for your information, I'll buy P4 2.8Ghz then :)
4x realtime is just for 1 pass :(

[Sirber]

strange...

with my 2000+ I need 2 days for a 4h movie (Rock et belles oreilles), so 1 day per pass, 6h / 1 movie h, so 6x. :devil:

[ookzDVD]

Originally posted by Sirber
strange...

with my 2000+ I need 2 days for a 4h movie (Rock et belles oreilles), so 1 day per pass, 6h / 1 movie h, so 6x. :devil:

Maybe you use too much filter :)

[ookzDVD]

Originally posted by scharfis_brain
which resolution do you try to encode?

maybe lowering that would help.

I prefer to encode 640x... for 1CD-rip.

[Sirber]

anamorphic for 1 CD sometimes improve the quality...

[slavickas]

Originally posted by karl_lillevold
@slavickas: xmm regs; I did not know that about amd64. Also, I just checked patternAdaptivity via the registry. Works fine here. Make sure you spell it correctly:

Windows Registry Editor Version 5.00

[HKEY_LOCAL_MACHINE\SOFTWARE\RealNetworks\RV9]
"patternAdaptivity"=dword:00000003




yup it worx, in http://forum.doom9.org/showthread.php?s=&threadid=60712 that thread your reg sample contained extra space in name.
any target date of new milestone or dll build with fixed scalingFactor ?

[karl_lillevold]

@slavickas: nice catch, i did not see that extra space at the end. I have corrected my post.

Re ScalingFactor fix: I don't know when the next Milestone will be ready. However, I added a slight improvement I would really like some feedback on before it becomes the new default, so I may post a DLL to test soon. This will also have the scalingFactor fix included.

[Sirber]

What's that fix about? 30% more quality? :D

[karl_lillevold]

To clarify, the scalingFactor fix is a minor fix only applicable when the scalingFactor param is used. I posted the details in this thread (http://forum.doom9.org/showthread.php?s=&postid=385462#post385462).

The improvement is something else, more details later. Nothing like 30%, probably not even visible most of the time, just another small step in the right direction.

[ToiletDuck]

I'm on a dual opteron 240 system running at 1500mhz right now with EHQ=100 and I am encoding at around 40-45% actual movie speed. I don't really understand the whole SSE optimization thing. Wasn't there an article on how you could just remove some flags or something similar and get a lot more speed out of the AMD chip?

[Hiro2k]

Originally posted by ToiletDuck
I'm on a dual opteron 240 system running at 1500mhz right now with EHQ=100 and I am encoding at around 40-45% actual movie speed. I don't really understand the whole SSE optimization thing. Wasn't there an article on how you could just remove some flags or something similar and get a lot more speed out of the AMD chip?

Which AMD chip?? There are many different cores and it makes a difference. Also I'm glad to hear that RV9 well suited for dual processors, one more reason to upgrade to dual Opterons! :D

If only vdub could use both procesors. :devil:

[ToiletDuck]

when using AutoDub it uses both... and as far as which processor i didn't know that opteron processors had different cores.

[McoreD]

AMD processors pretty much suck when it gets to Video Encoding. I can backup my statement with practical benchmarking tests done at TomsHardware Labs.

Video Benchmarks for:
Mainconcept MPEG Encoder 1.4.1
Pinnacle Studio 9
XMPEG 5.0.3 / DivX 5.1.1 Pro
http://www.tomshardware.com/cpu/20040201/prescott-14.html
Windows Media Encoder 9
Windows Movie Maker 2.0
http://www.tomshardware.com/cpu/20040201/prescott-15.html

If you check those pages, you will see AMD processors get no where near 1st, 2nd or 3rd.

That's enough for me to go with Intel®. :)

[nFury8]

I can backup my statement with practical benchmarking tests done at TomsHardware Labs.
Its well known that AMD procs cant keep up with Intel's P4 when it comes to video encoding. But for pity's sake, there's a lot more reputable and credible sites to base one's decisions on, other than 'that' site. Of course these more respectable sites still conclude that P4 is faster in video encoding.

regards

[Sirber]

I agree that P4 is better for viseo encoding, but AMD is better in a student's budget :D

[slavickas]

Originally posted by Sirber
I agree that P4 is better for viseo encoding, but AMD is better in a student's budget :D

true, true :D nothing can't beat overclocked 1400 duron for price/performance :cool:

[ToiletDuck]

I don't know about that. with my dual opteron 240's running at 1500mhz I encoded xvid at 640xXXX with sharpening and deblocking options all turned on and I still get about 60-70fps. I think it comes down to the codecs are more optimized for intel chips. I can use adobe premier and edit video in realtime rendering with out matrox cards or anything. Should there be a 64bit optimized code for the AMD64 architecture I think the opteron's would clobber the intel hands down.

[ToiletDuck]

I don't know about that. with my dual opteron 240's running at 1500mhz I encoded xvid at 640xXXX with sharpening and deblocking options all turned on and I still get about 60-70fps. I think it comes down to the codecs are more optimized for intel chips. I can use adobe premier and edit video in realtime rendering with out matrox cards or anything. Should there be a 64bit optimized code for the AMD64 architecture I think the opteron's would clobber the intel hands down.

********Edit*******
http://www.extremetech.com/article2/0,3973,1402705,00.asp

The operon gives the p4 a good run for it's money here. Wins more than loses. And all this is 32bit.

In uniprocessor systems, Intel's Hyper-Threading technology often gives it an edge over an AMD64-based CPU in some benchmarks. However, once we toss in a second CPU, all bets are off. The dual Opteron system outperformed the Xeon 3.06GHz (and would likely outperform the 3.2GHz version) in benchmarks where Intel had won on single CPU performance.

[Big_Berny]

Well, I think it's a lot cheaper to get a DVD-burner, isn't it?
Why do you want a fast PC to encode realtime?

Big_Berny

[Hiro2k]

Because not all of us have the time to spend dedicating our computers to 1 single task. I know with windows 2000 you can set the priorities lower and do other things, but sometimes if your doing a test on a filter, then speed is an issue and you want to get through with things as soon as possible.

[ToiletDuck]

Well, I think it's a lot cheaper to get a DVD-burner, isn't it?

What are you doing here then lol.

[Big_Berny]

Originally posted by ToiletDuck
What are you doing here then lol.

At the moment I haven't even enough money to buy a dvd-burner! :D
I'll wait for the new doublelayer-generation... Until then I have to use one of these codecs.

Big_Berny

[MfA]

P4 would be a so much nicer processor if it didnt stall at the drop of a hat ... I would definetely get a Prescott if you want Intel, the SSE3 instruction which allows non aligned accesses, without the ridiculous stalls you get on a normal P4, will probably become quite important for video encoding apps (for the rest SSE3 is pretty much inconsequential).

[Sirber]

P4 is also generaly more quiet than an AMD machine :)

[nFury8]

P4 is also generaly more quiet than an AMD machine :)
You mean the non-Athlon64's? I believe A64's are much more quieter than P4's comparatively. But I like my procs a bit noisy cause that would mean it's doing a lot more work than a silent one. :D ;)

[Maverick]

[/QUOTE]
Originally posted by karl_lillevold an Athlon 1700 is really only something like a 1.4 GHz, so if you get a P4 2.8 GHz, you get 2X * 1.25 (25% extra for SSE2) = 2.5 speedup.
Wow. That's the most incredbile statement I've ever read. Now, disregarding the Athlon has a 12 stage pipeline and a Pentium 4 has a 20 stage pipeline, the cache mispredictions the Pentium 4 has so often, AND the fact it's tiny L1 instruction cache can barely cache a thing...
RealVideo's encoder must be doing some REALLY funky stuff. You can't say double the clock speed = double the MIPS/speed. The P4 3.2 Extreme does so well because of it's giant 512 L2 + 2048K L3 caches. Most video encoding I see has *Big* data sets to process, so cache helps enormously, to the standard un-extreme P4's detriment.


Originally posted by karl_lillevold
RV9 incl EHQ has already been optimized for SSE3, but not yet released. This speedup should be approx 5% in addition to SSE2, more measurements later.
If it's only 5%, then Prescott at equal clock speeds, unfortunately, it going to be slower than Northwood. Intel decided it would be 'REALLY COOL' if they made the Prescott 30 stages. Yay. Intel will be releasing a newly optimised C++ compiler soon, look out for it.

Originally posted by karl_lillevold EDIT: I don't know the effect of Dual DDR 800 MHz memory, but I am guessing it will help quite a bit. Video encoding is very memory access intensive.
Just curious, what hardware do you guys test your encoder on?

[Sirber]

always stuff that we don't have :(

[ToiletDuck]

However as cheap as DVD burners are getting it almost seems like in 2-3yrs that divx, xvid, and all other codecs will not be needed anymore. Maybe for live streaming media. Even then the new bandwiths will be raising so fast it is unreal. However I could be biting my toung in the future by not thinking of what else is coming out. The new "blue wave" or whatever technology will be coming out. Pretty soon we might have 40gb disk that we are wanting to put on a 4gb disk...... ehhhhh only time will tell i guess.

[Sirber]

Higher disk capacity mean more movies per disk :D Higher bandwith means more realtime streaming. XviD/DivX/VP6/RV10 won't die soon :)

[iwod]

It is quite funny looking back at my own reply from September. Where i still believe Intel is going to Make Prescott better than rumors.

However everything in the wild came true where it suck power like hell, i could even imagine buying so a expensive chip and to pay extra for its elecrtity Bills.

People tend to say Intel is better for encoding only because all comparsion has been done on application which are specificly optimize for intel CPU only.

And so far i haven't seen any article on encoding RV10 EHQ 100. ( Suddenly thought i could recommand this to Anandtech )

[ToiletDuck]

iwood i encode at 100hq and can do a 2hr movie in about 4hours total. That is ripping it, doing audio, and 2pass encode.

[nFury8]

Originally posted by ToiletDuck
iwood i encode at 100hq and can do a 2hr movie in about 4hours total. That is ripping it, doing audio, and 2pass encode.
Really? What's your exact rig setup? Encoding parameters? That rig screams, man. Awesome.

[Maverick]

Makes my 36hr WM9 encoding sessions (P4 2.6C) sound like marathons...

[nFury8]

I just searched a couple of posts back in this thread and realized this guy(ToiletDuck) is running a dual Opteron (240?). Jeez, where do these guys get all the money to buy life's expensive stuff? :D Duck, mind posting your whole setup like memory and mobo, as well as encoding parameters?

[ToiletDuck]

I have dual opteron 240's@1.5ghz. MSI K8T800 motherboard. 1gb 4x256pc3200ram. 3x120gb harddrives. 1x60gb 8mb cache harddrive. 1x30gb drive.

2x240's=$382
motherboard=$225
ram=$197
-------------------
$804

$804 at dell gets you-Dimension2400 with integrated audio and graphics and pentium 4 2.66ghz... gotta love building your own. The rest of my system is about 2 years old. I get parts here and there. Rest is ATI 9700pro, 4.1 surround sound, 21" trinitron monitor(Flat screen not panel) and soundblaster extigy soundcard.

For encoding I usually do 640xXXX after cropping. I use autoRV10 for my encodes. I was using xvid but have been having some lockup issues that I can't explain on the second pass of the codec.
Duck

[Soulhunter]

Originally posted by ToiletDuck
However as cheap as DVD burners are getting it almost seems like in 2-3yrs that divx, xvid, and all other codecs will not be needed anymore.
Why not H.264 -> DivX 6 n' XviD 2 or WMV10 to fix 120 min. HD content on a DVD5 !?!

Bye

[iwod]

Well......... MY GOD!...... i would consider anybody have a ATthlon 64 to be rich already and you got a Dual Opetron!!!!!!!!!!

Anywayi heard that D0 Core of Prescott in Socket T ( aka LGA 775 ) is suppose to improve things and add the speed up to 3.6 GHz.

So i may have to be a bit more paitent with Intel.

Considering Even with the topest Computer we still can't do Real Time Rv10 encode..........( And yet there is H.2x4 ) i am now convine why we need things like a 10 GHz 64 Bit CPU.... :D

[ToiletDuck]

Not rich lol. Build your own and you would be suprised. I work for dell right now while in college and can tell you that every computer sold has at least 30% margin or more between the price to build and price to cost. All the time I sell $1400 systems that have $1100 in margin. The total cost of my system is about $1400. That isn't rich. That is just looking around and taking advantage of opportunities when they arise. My 21" monitor that sells for $450 at Dell(mine actually is the same exact dell) I bought on ebay for $150. Want some tips to build a great system for cheap. Look on ebay at 2am. Tons of people post things and don't set the time at which the auction ends. At 2am no one is awake and no one is bidding and I snatch up some great deals! Also I buy things that I can keep for a long time. Such as the monitor and sound card. Those things stay. So whenever I want a new system I just get a new motherboard, ram, and chips and BOOM. thats it. Not to hurt anyone's feelings but if you are sitting on a dell, gateway, any computer you didnt build yourself, then you could have gotten way more for your money just by building it yourself. It is also very interesting and a great learning experience to do this. If you have any questions about how to build ultimate computers for cheap go around www.overclockers.com. Great people there.

I've also heard that divx and such use live streaming data so cache doesn't matter. However I find this odd. Anyway if any of you are wondering what you could get for a pricerange PM me a quote. I'd be more than happy to see what kind of system you could get for your cash. I'll even compare it to what you could get for that amount if you ordered from me at Dell or called Gateway.

Happy encoding!
Duck

[lilhobo]

most like the dual opterons are for the porno rips right??? all the porno vids are in freakin Xvids and wmv9's

awesome freakin quality too if i may say so ;)

[nFury8]

Originally posted by lilhobo
most like the dual opterons are for the porno rips right??? all the porno vids are in freakin Xvids and wmv9's
:confused: :confused: :confused:

[Maverick]

Originally posted by ToiletDuck
I've also heard that divx and such use live streaming data so cache doesn't matter. However I find this odd.

Divx, WM9, (don't know about that hobo RM :)), most encoders use *big* data sets up to 10 frames long. That means memory bandwidth, and to save memory bandwidth... CACHE.
That's why my EE just burns through WM9.
At least that's what I've gathered from my research over the years...

[trooper11]

well as far as waiting for the new prescott socket, who knows. There have been so many rumours, ill just wait and see, it may not be as glorious as they say. whne i can see the ahtlon socket 939 out as well as the socket T then I can compare. As far as encoding goes, its very true intel chips outpace since they are mhz hungry type apps, granted there are many that are optimized for intel chips anyway. SSE3 will be implimented in the new socket 939 for amd so that advatange will be gone. Maybe, if we are lucky, Id like to see more amd optimized codecs for testing, such as 64bit capable. Eitehr way, itll still be several months before we see much of anything.

[karl_lillevold]

Like I wrote in an early post in this thread, RV10 encoding gains around 5% from the new SSE3 instructions on Prescott. However, what is really nice is that finally, due to increase cache sizes, hyper-threading actually makes a difference for encoding (it was always good for the decoder). I measured an additional 20% speedup from running the encoder in dual threaded mode, on a single CPU Prescott system with HT enabled, provided by Intel. So a total of 25% is not bad. Unfortunately, I still have not seen or been able to try RV10 encoding on AMD64, but I am curious how it performs.

[Sharktooth]

i have a dual 248 opteron. dont know how it performs on RV9-10 coz i dont like realmedia stuff...
however i can easily compare other codecs performance on single and dual processor mode.
the average gain on a beta build of windows xp for 64 bit extended processors (divx standard settings) is about 80% (from single to dual) and a freaking 220% comparing the dual 248 with my old athlon 2600+ @ 3200+.
as soon as i have some spare time i will post some logs and stuff...

edit: i've just acquired from some internet sources that the new athlon 64 models (newcastle core) should have 1Ghz FSB (200x5) and SSE3 support...

[ToiletDuck]

and i was just starting to love this system :)

[iwod]

Well...... ok may be you are not rich. Just that i am Poor. :)

I don't 100% agree that Encoder are Mhz type app statement. The only reason why todays encoder are faster on INtel is simply becoz they are ALL optimize for intel chip. ( Or say they don't have a AMD chip in their test lap. )

Either intel have lost track of what they are doing or they have something in hand. Prescott doesn't look too bright with all these Heat and power problem. While AMD Athlon 64 got a long way to go before they really need to move on to 0.09 um.

@ Karl. Yea i remember you said HT does not speed anything up in encoding when it was Northwood time. So has the prescott with 1 M cache and HT improved the speed by 20%, is that what you mean?

Or are you referring to P4EE?

( Because If Prescott HT improve Encoding by 20% then it would be a totally different perpective.... )

[Sharktooth]

The prescot advantage in encoding is not due to HT.
Its SSE3 and the larger cache that matters.
However if you search on the web there are news about the newcastle core for AMD 64 cpus talking about 1Ghz FSB and SSE3 support.

[bill_baroud]

But don't forget the prescott's cache latency which is almost twice as northwood's cache !
Atm, prescott are like the first p4 willamette, i.e they are designed to run at 4ghz and more, they can't compete with actual northwood at the same frequency... and they are going soon to change the socket (478 to LGA775) so prescott aren't definitly an option.

And for the reason why p4 perform always so well in encoding app is 1) because a lot of those app are heavily optimized for p4 and 2) p4 are designed for that task, i.e computing (integer) "data stream" with no unpredictible branching, which is typically a video encoding :)

[nFury8]

originally posted by iwod
I don't 100% agree that Encoder are Mhz type app statement. The only reason why todays encoder are faster on INtel is simply becoz they are ALL optimize for intel chip. ( Or say they don't have a AMD chip in their test lap. )
originally posted by bill_baroud
And for the reason why p4 perform always so well in encoding app is 1) because a lot of those app are heavily optimized for p4 and 2) p4 are designed for that task, i.e computing (integer) "data stream" with no unpredictible branching, which is typically a video encoding :)
To my mind, one of the compelling reasons why AMD's Athlon has endeared itself to the enthusiast's heart is the fact that right out of the box it is such a powerful and capable architecture without having to rely so much on software optimization and special instruction sets, etc etc. Makes me wonder how the competition would fare if the Athlon has only received as much attention and its own optimizations from those major developers. It's a bit silly, that for all the strong performance achievements it has shown since it first came out, so many are still unable to give it the credit that it rightfully deserves. The general mindset is still "AMD is just a cheap second-rate catch-up player". It is a mentality that I witness everyday in my office, in some friends and of course some of the tech forums.

[Wolfman]

Another point about CPU's etc is that I/most people dont compare 2.6 athlon with 2.6intel and then say, ahh intel fastest, NO. People say I have £65 to spend on a cpu and Mobo, what is the best I can get for that?? inevitably if your on a budget AMD is cheaper. price per VEB Amd is normally cheaper.

For realtime encoding in rv9 EHQ=110?? Beowulf cluster 64 sun sparcs running Linux .. 10 teraflops computing power.(as no budget given!):rolleyes:

[ToiletDuck]

bah. I imagine that dual opteron 248's should be able to. I'm not too far off with my 240's. They are overclocked to 1500mhz but a 2.2-2.5ghz pair of opterons should do the job nicely.

[Sharktooth]

Originally posted by ToiletDuck
bah. I imagine that dual opteron 248's should be able to. I'm not too far off with my 240's. They are overclocked to 1500mhz but a 2.2-2.5ghz pair of opterons should do the job nicely.
my dual 248 eats everything the universe throw at it:)

[Soulhunter]

Originally posted by Sharktooth
my dual 248 eats everything the universe throw at it:) Thats nothing against thus japanese Earth Simulator (http://www.es.jamstec.go.jp/esc/eng/ES/performance.html)... :p


Bye