Incredible speed boost with 2.59 and hyperthreaded dual proc

[ninja.rogue]

Dear friends,
I have successfully installed a Supermicro P4DCE+, which mounts two Intel Xeon 2.4 and 1.024 MB of Rdram.
The speed of this monster simply blazes any Athlon mp-based PC.
This is for true the best PC for video editing:
Hyperthreading (4 processors enabled into XP)
SSE-2 software extensions

If you use any SSE-2 enabled software and-or SMP enabled software you will find an enormous speed boost.
In the case your software is also hyperthreading-enabled, such as TMPGENC, you will notice that earth is not under your feet anymore...

Just to make a simple comparison:
P4 1.9 512MB RDRAM
Encoding DV AVI PAL 720x576 with Tmpgenc Plus 2.59, highest quality, 2-pass vbr, 9000 max, 6000 avg, 2000 min, no filters enabled.
Speed: 15x1

Using Dual Xeon as above: Speed 4,5x1

This is for real!
I do not want to flame another Athlon vs Xeon war, but this are my results.
I am willing to provide any further detail.
Cheers

Ninja.rogue

[SatStorm]

The price of this baby, is SciFi for most home users you know....
I have to work hard 6 months just to collect the money it costs...

For pro use, yes, it is today's TOP solution.

[ninja.rogue]

Hi Satstorm,
I just wanted to add some technical details.
When it comes to price/performance problems, there obviously are differences among users, depending on whether or not they are having jobs on video.
If my post sounded offensive to you, I beg pardon.
Cheers

Ninja.rogue

[SatStorm]

I know that I have a reputation of the easy offensive type person, but I didn't realise that I scare users!

No, your post didn't sounded offensive in any way! I liked your report, and I wish you happy encoding!

It seems I have to change attitude....

[Dragonsf]

What do you mean with 15:1? 15 hours conversion for 1 hour footage?
What is your result format ? SVCD mpeg-2 or DVD mpeg-2?
My Athlon makes this in 3:1, that's why I'm confused!

[BJ_M]

yes -- i get the same results on a dual ibm m pro xeon and metioned it in a post before , tmpgenc now blazes and so does main concept which are both hyperthread enabled. the time factor that cce had in speed was eroded mostly and the MC is faster in many cases (though not a direct settings comparison) ..

[ninja.rogue]

@ Dragonsf:
The output file is DVD compliant MPEG-2 with bitrate as follows: min 2000, avg. 6000, max 9000.
The input file is a single 10' 30" DV PAL AVI.
And, yes, the encoding time is as I have posted.
I will soon post onto 2cpu my results for Tmpgenc test.
See proper thread there.
Cheers

[Dragonsf]

I tried to do the same (10 min DV PAL avi to 6000 DVD single pass) and my program took 30 min to convert. How do you comapre that to your 45 min, your system did take?

[ninja.rogue]

Sorry Dragonsf,
I forgot to say that I used 2-pass vbr with Tmpgenc Plus!
So it should be almost twice as long (almost due to the fact that you can enable a disk cache for second pass - this I have already done).
I also used Highest quality setting for motion estimate.

When I use W2Kpro sp2, I can get as low as 4-1 or 5-1 depending on motion complexity.
Still having 2-pass and highest quality.

[Craig Tucker]

An hour of video also takes my system approximately 15 hours to convert using TMPGEnc for PAL DVD with 2 pass VBR and 2000,6000,9000kbps, and highest quality for motion search precision. As soon as prices come down a bit I will be upgrading to an Athlon XP2800+, so I can utilise the 333mhz FSB and match it with my 333mhz ram. Will be interesting to see how much of a speed increase this will yield.

[DJRumpy]

No no no..Your all thinking TMPGenc. He's talking CCE. In CCE, we go the other way, where 1X is equal to playback speed. He's encoding at 15X the playback speed of the movie.

If your moving is two hours long, then you would divide that by 15 to get the total time it takes to encode. The numbers are definately believable. I get simular, albeit slower numbers with a single PIV 2.4 Ghz. I have to be encoding in SVCD, or VCD resolutions to get numbers like those, but you can encode an entire movie in mere minutes.

[ninja.rogue]

What I meant was that original file was 11' 30"; encoding time varies from 50' to 58'.
As 50/11,5 makes 4,3 and 58/11,5 makes 5, I indicated about 4,5-5:1 encoding time - i.e. 5 times the length of the movie you are encoding.
Again, please see that all settings have been set to maximum quality, and original and destination file have maximum DVD resolution (pal).
Cheers to all
Paolo

[Craig Tucker]

No no no..Your all thinking TMPGenc. He's talking CCE. In CCE, we go the other way, where 1X is equal to playback speed. He's encoding at 15X the playback speed of the movie.

If your moving is two hours long, then you would divide that by 15 to get the total time it takes to encode. The numbers are definately believable. I get simular, albeit slower numbers with a single PIV 2.4 Ghz. I have to be encoding in SVCD, or VCD resolutions to get numbers like those, but you can encode an entire movie in mere minutes.

If you read the first post ninja.rogue states that he is using TMPGEnc. And like I say 15x not 1/15 is a realistic time for encoding DVD with those settings (he was using a P4 1.9GHz, I am using an XP1900, and it took us the same time to encode), so he has gone from 15x to 5x, a 3x increase in speed.

Not sure if the source file would make much of a difference in encoding time, but my source was not DV it was 720x576 PAL YUY2 with huffy. This had a bitrate of about 10.5MB/s rather than the 3.6MB/s you would have with DV.

[DJRumpy]

I stand corrected. It was the 'blazing' that threw me off. I don't consider 15 hours for 1 hour of footage 'blazing'. I'd rather go to the dentist than wait 15 hours (or 30 for a full movie) to encode.

CCE is a must. Once you try it, you'll never go back. I can encode an 80 minute VCD in 8 minutes (10X). Hence my misunderstanding. At first I missed the easy click interface, but AVISynth more than made up for those.

[BJ_M]

in my tests - CCE didnt make very good vcd quality ... plus, though very very good for mpeg2, it is an expensive program which if you had to buy it , you wouldnt be using.

8 min for a 80 min vcd on a p4 2.4 ? sounds a bit to fast ..

what the first poster of this thread was pointing out quite rightly is that the newest version of tmpgenc (and as i pointed out also Main Concept) are very fast on hyperthreaded systems as many of us are using now. and that the speed waas now about the same as CCE ...

why do these threads keep turning into "CCE the ONLY way to go" or something to that effect .. sure CCE is VERY VERY good , but at 49$ and all that it can do - tmpgenc is the bargin of this century (so far - and add a few freeware apps in there) ...
maybe i'm a little miffed because we have to purchase a lic for every piece of software we use and it costs 100's of thousands a year in lic and support fee's for the the use of some of the same apps which are often mentioned in here, used by "joe 'puter" at home who doesnt have a clue on how to use them. sorry - rant of the week ..

[raven4]

FWIW, I've got a supermicro running 2x2.2 XEON's and just finished building an ASUS P4PE running a single 3.06 Ghz P4. The P4PE beats the pants off of that XEON machine by so much it's embarassing. IN tests for rendering speed, hyperthreading doesn't really do much to help out. The biggest gain with HT is about 20% *IF* the software supports SMP. Duallie machines are silky smooth, but, are they worth the extra cost? Not in my opinion. Upgrading is a LOT cheaper with a single CPU...and we all know how often we want to upgrade our machine, que no?

[ninja.rogue]

@BJ_M: you made my point!
This does not want to be an attempt to fire another CCE-TMPGENC war, but just to say that - once a while - Intel technology was well used...

[DJRumpy]

I actually get good VCD quality with it, but I had to change the matrix setting to do it. The 'Standard' setting leaves something to be desired.

On the whole CCE/TMPGenc thing, I doubt there are any many, if any individuals here who payed for it. Granted, it's a nice piece of software, but overpriced. It lacks what I consider to be very basic features, like pulldown and decent audio support. It's fast, but still very manual. If they priced it appropriately, they've probably make a small fortune, as people would be willing to pay a fair price for it. Look what happend to microsoft. They used to charge 'normal' prices for their OS. Now that they have a lock on the market, it's 200 dollars US for the Pro version. Pirating is rampant on the XP Pro version.

4 hours to encode one hour of video with a dual PIV system is still slow. I used to get 1.2x encoding speed on my old single 1.5Ghz PIV (note, that's 1.2 times playbackspeed or 50 minutes to encode 1 hour, not 1.2 hours to encode 1 hour ). On the new processor, it approaches 3x playback speed, or about 20 minutes to encode 1 hour of 720x480 video. 8 minutes to encode 352x240.

How did I get so far off topic?

[BJ_M]

the newest version of CCE has pulldown - it also has a something which makes it a lot more usefull to me - namely that it can now accept targa frames (which i use)

[DJRumpy]

What's the version BJ? Also, is it stable? Pulldown would be very useful. I'm on 2.64 (or .62 I think). It's stable enough, but gets a bit flakey sometimes when frameserving via AVISynth in XP.

[BJ_M]

2.66.01.07

it also works fine frame serving - no problems there (now - once again) ..

[BJ_M]

btw - cce was not multithreaded (truly completly multithreaded) until version 2.64 ..

[DJRumpy]

The interface on the new version is a bit neater, and easier to navigate. The pulldown feature will be something I'll have to test. It should be a timesaver.

Ran a few more tests. Encode times at 720x480 ran at 1.4X, or about 42 minutes to encode 1 hour of video.

VCD 352x240 ran at 8X, but the only VCD material I had around was 8 Crazy Nights, which is a cam capture, letterboxed with a 4:3 DAR (mimics 16:9), so this was hardly stressing the encoder. The actual image area of this thing is tiny. I'll have to create a full screen 4:3 VCD, and try another run.