p4 sucks at CCE?

[Demi9OD]

Just doing some experimenting with a p4 setup I bought recently. Seems like CCE 2.5 has NO p4 optimizations and Athlon XP chips do a much better job of encoding.

Encode test was done using Me Myself and Irene R1. DVD2AVI ripped in YUV mode on both comps, same version of avisynth used (whatever comes with Gordian Knot .26 update)

Athlon 1800xp - 1.47
P4 2.4B - 1.36

Now I know clock for clock the Athlon XPs are faster, but come on, a 1533mhz chip outpacing a 2400mhz?

Anyone else experiance the same kind of speeds or is my p4 setup screwed up some how.

Peace,
Demi

[auenf]

im guessing you are running DDR on the athlon, but what memory are you running on the P4?

CCE requires 3dnow Pro or higher or SSE (not sure about SSE2), but is still FPU intense, which is where the P4 is mostly lacking.

other thing is that 2.5 is rather old, and P4 speed is probably better with the newer vers, i think 2.5 came out very early in the life of the P4, but i cant remember back that far.

Enf...

[Demi9OD]

Memory is the same on both systems, pc2100 DDR, running at 133mhz (266 DDR) on both boards.

Indeed CCE 2.5 is an old program, I believe when it was released the top of the line cpu was a p3 700 or so. I remember I couldn't run it on my celeron 300 running at 504 (god I loved those chips) because it lacked SSE. Just pretty incredible how slow the p4 is when software is not optimized for it.

Demi... :D

[jrg]

clock for clock, a PIII would beat a P4 if you ran them at the same speed. Intel lengthened the pipeline in P4 so it runs faster, but does less work per clock. As long as software is optimized, it performs well and Intel does a good job of that.

[mmgrover]

I have a P4 with 2.2g processor.

Using Avisynth2, mpeg2dec2 and CCE 2.5 I get about 1.65

With 2.66 I get about 1.8, and video looks better with 2.66


mike

[Arky]

Originally posted by mmgrover
I have a P4 with 2.2g processor.

Using Avisynth2, mpeg2dec2 and CCE 2.5 I get about 1.65

With 2.66 I get about 1.8, and video looks better with 2.66


mike


1.8?! :eek:

What framesize are you encoding? Surely not full D1?

Although admittedly I am so lazy that I still use VFAPI instead of AVIsynth (thus performing inefficient and unecessary YUV > RGB conversion), I only get 1.09 top speed with D1 framesizes and 2.66.

I am running a P4 (533mhz FSB) 2.26ghz with a gig of RAM, so I am surprised that you are getting almost double the performance.

While I realise that using the AVIsynth/MPEG2dec option would bring noticeable speed improvements, would it really account for such a dramatic difference, or are you also encoding smaller framesizes (e.g. SVCD)?

Cheers.


Arky ;o)

[Matthew]

Dual 2.2 Ghz xeons, I get say 1.8 in a fairly good instance with 2.64 (latest build, which supports avs). With 2.66 I get like 2.2 for same source. These are rough approximations from memory. But with CCE 2.66 the speed doesn't appear to drop below 2.

avs script is only 2 lines, so no resize or anything like that (and no filters in CCE). Plus I only use DVD source.

avisynth only has single processor support, otherwise the speeds would be better =)

[ndphan]

I converted movie DVD "Ronin" NTSC to SVCD using exact default setting of DVD2SVCD software with Simle Resize, CCE 2.50, no noise filter,480x480, multi pass=3, with 2 PC, one is AMD XP 2000+,512MB PC2100RAM, 7200RPM HD
the other is DELL P IV 2.4 GHZ 533 MHZ bus, 512 MB PC2700 DDR RAM, same hard drive
the CCE encoder speed was about same for both PC which is ~2.01-2.15, total time about 4-5 hrs.

I don't know much about AVisyn or VFAPI, I just install DVD2SVCD and let it run. I think main factor for encoding speed is you CPU, even when I tried 128 MB RAM, only 1 20GB hard drive on WIN XP, I still got 2.0 on CCE speed, of course you couldn't anything else, the available RAM was about 3-4 MB. :) Upgrade you CPU then have fun. Pentium IV is not suck, depend on what're doing, It run software to search for prime number faster than same rating AMD CPU (www.mersenne.org), but AMD beat Intel on RC5 encryption software (www.distributed.net). Enjoy & proud what PC you have, but it best to have the best of both worlds. :)

[alturismo]

wuhuu

an XP2000 with 512 DDR and speed about ~2.0

i must doing something wrong here ;)

i have an XP1800 also with 512mb DDR CL2 timing Infineon

i come with full 720 x 576 at about ~2600 0-9500 to speed ~ 1.1

with resolution like 480 x 576 ~2600 0-9500 up to ~1.3

running on XP pro SP1 (without SP1 about the same)

enoding from IBM 7200rpm HD to another one like that

my avs file is a simple 2 line script, and no filters activated
in cce.

so what really interests me, where coukd be that huge difference now

XP 2000 to 1800 shouldnt double the speed i think

so could it be the boatd (chipset) or are there any drivers wich
"breaks down" cce so much

or is this an unsolved secret ;) wich nobody really knows

i will make some tests on an friends px (p4 1.8), lets see what
his mashine is able too

[ndphan]

I used AMD XP1500+ before and compared to my cousin's PC Intel PIV 1.7 GHZ, the CCE speed came out about same ~1.5-1.8, but when I encode from capture TV, VHS the speed fell to 1.1~1.2, due to Hybrid film detected, IVTC activated, slow down the whole process. It fast when encoded from DVD source. Hope this help.

[Phlex]

I think I need to recheck my system. I have a AMD XP 1900+, not overclocked, 512 DDR2100 CL2.5 (Crucial), 3 7200 rmp drives. I'm running XP Pro. I'm only getting .240 with CCE when converting from dvd to svcd. However, I do use several filters in avisynth.

I use avisynth 2.06 to frameserve to CCE 2.5. Filters I use are Mam, convolution3d, tomsmocomp, and the internal lanczosresize to resize to 480x480. Needless to say my last encoding lasted 40 some hours for a 4 pass vbr. I tried avisynth with only tomsmocomp and lanczos to resize and the best result was only .85 with CCE.

Phlex

[Demi9OD]

Phlex: Try running it with no filters at all, you should get around 1.7.

[Phlex]

@Demi9OD I commented out all the filters, I was only using the internal crop and lanczos to resize. I'm getting about 1.03 in cce.

Just for fun I overclocked my cpu to 2000+ around 1.7 GHz and overclocked my memory as well to cas2 and decreased the timing. I also increased the voltage another +.25 volts from the 2.5v default.
I loaded up my test.avs file and I was getting 2.4 from cce. After 4 minutes I got the BSOD, muhaha.. :D I always oc'ed my cpu in the past but the 3 7200 WD drives are causing too much heat in the case. I did have HD cooler fans for all the drives at one time but the noise was too much. Maybe next time I upgrade my system I'll try water cooling so I don't have to deal with the noisy fans.

Phlex

[Demi9OD]

Speaking from personal experiance water cooling is a major pain in the arse. Try running the cpu at default speed but with the most agressive memory settings, that will prb help a bit. Or unlock your athlon, drop you multiplier, and run a super high bus :). My unlocked 1800xp is running at 9.5*180, and it's getting around 2.1 in CCE now.

[Phlex]

I was thinking of buying a P4 2.4 but I think I get more bang for my dollar from a XP 2400+, or wait for the 2800+. My MB is not a great board for oc'ing. It's a Soyo Av7 Dragon Plus with a Via 266a chipset. I always have problems oc'ing the bus beyond 140 Mhz. Actually only CCE crashes on a regular basis at 140 or beyond. I think I'll unlock my cpu, I've seen the unlocking kit for about $10.

I'm impressed you were able to oc your bus at 180, I'll be happy if I can oc it to 160, I'll probably need DDR2700+. You'll definately have tweeked your system well! ;) It's nice to know a tweeked AMD 1800 system can achieve 2.1 in CCE.

Phlex

[Demi9OD]

You wont be able to go that high even with better memory. It's the chipset thats holding you back, I had the same problem with my 266a board. Soon as I put a MSI KT3 Ultra 2 in there (kt333) i could do the higher bus.

Peace

[Arky]

Forgive me if this has been mentioned in the past (I've never seen it mentioned, but I can't read every single Doom9 post...), but might I suggest an altrenative solution for those of you who have dual-processor (aka SMP) rigs, and who are using AVIsynth?

Why not choose a suitable juncture in your source VOBs (or DV files - sometimes I feel I am alone in encoding non-ripped material!), and split your encoding job into two. That way, you could run two instances of CCE-SP concurrently - each having it's own thread on independent processors. It's not a perfect solution, and I don't know if AVIsynth will run 2 instances concurrently, but has anyone attempted this? Perhaps windows itself will not allow 2 frameservers concurrently - I don't know. Not having an SMP rig, I can't make any test-runs to see if the idea works.

I'm very interested to hear feedback on this, because I am still hankering for an SMP dually (probably AMD), when Clawhammer finally materialises, and matures enough for SMP boards to appear, from Tyan and MSI etc.

Incidentally, I am NOT suggesting that the resultant two files should be "pasted" together - this would invariably give rise to GOP inconsistencies. I know, it's not great, but I'm disucussing this purely as a *hypothesis* on how to use an SMP rig to full effect, while frameserving to CCE-SP, so no flaming please! :rolleyes:


Arky ;o)

[Matthew]

Arky yes running 2 simultaneous encoding jobs is possible (I've done it before when fiddling round with samples), although it's a bit of an irritation in both setting it up and when one ends up with 2 mpvs for the one film (as you mentioned). However if one has 2 separate DVDs to backup, then running the 2 projects simultaneously would be fine.

Also there's an advantage in having avisynth/CCE using only one processor - one can perform other normal tasks at the same time.

As I get an RT of 2 or above anyway with one processor, I'm happy to just leave the computer like that (performing a 4 pass VBR), whether I am using the computer for other tasks or not.

BTW I believe another less feasible solution is to convert the DVD to a huffyuv avi and feed that directly into CCE. The "less feasible" part derives from the need to have a 20 terabyte HD :p

[Arky]

lol - yeah, Huffy really is a *TOTAL* space-HOG!!! :scared: :D

Still a great codec though...


Arky ;o)

[asifanwar]

I used to have a dual AthlonXP 1600 on a Tyan S2460 mobo (AMD760 chipset) with 512Mb PC2100 Crucial DDR and RAID0 etc.
d2v to AVIsynth to CCE2.5 = 1.85 realtime

Now I have a 2.4B Pentium4 overclocked to 2.61GHz (which is a notably lower CPU score on SiSandra) but I now get a speed (on CCE 2.66) of 2.52x realtime. The big difference that I have not yet mentioned is I have gone away from DDR2100 and have RAMBUS1066 which is overclocked to RAMBUS1160. Opening up BANDWIDTH to your CPU is a key factor.

To illustrate my point: I just made an IWill MPX2 mobo with dual AthlonXP2000s (unlocked to allow SMP). I overclocked these two to Athlon2200s and my SiSandra CPU score hit the stratosphere! (CPU score 10900!). My encode speed (compared to the dual 1600) went from 1.85 to 2.1 cos I had the same DDR2100 sitting in there (with the same stinky AMD760 chipset)

I am very happy with the new system but will definitely go for a dual AMD in the summer when the Claws come out and the chipsets dont throttle me!

Check here for my original discussion on RAMBUS vs DDR:
http://forum.doom9.org/showthread.php?s=&threadid=34848

BTW: my avs script simply framserves, with nothing else.

[Demi9OD]

Nice post asifan, I can confirm that an increased memory bandwidth does make a difference in encoding speeds. When I bumped my unlocked athlon bus from 133 to 180, there was a 20% increase in speed despite the actualy clock speed being the same.

On another note, I got an Intel 2.4B and Intel 845gbv mainboard for a great deal as an Intel IPD (150 bucks), unfortunately it doesn't support even 333ddr, only 266. I'll probably move the 2.4 over to a Granite Bay board (dual channel DDR) which tests have shown is equivilent to 1066 rambus, as well as overclock the FSB some, see if I can break the 3.0 barrier by December :)

[auenf]

the other thing i noticed when testing my new cpu (XP1800+) last night, was that encoding from a MS DV (export from prem6), i was getting 0.5 (lot of static, not good), but encoding from a 'cleaner' mjpeg raw capture, i got 1.3 easily.

so the decompressor in CCE (DV is a built-in decompressor, rest uses vfw codecs) makes a difference too, altho those using avisynth/vfapi should know they are basically supplying CCE with a RAW stream ;)

Enf...

[TRILIGHT]

I know this is a rather old thread. I stumbled across it while searching for something else. Just thought I would share some info in case someone did a search for this particular info:

Speed with no filters: 2.5x
Speed running decomb: 1.5x

Hardware: P4 2.4Ghz o/c 2.7Ghz, 512MB PC1066 Rambus, Asus P4T533-C systemboard