Avisynth + VDUB = Multithread Support?!

[LB]

Uhh, yea this is somewhat weird. I run dual AMD 1800s and just found that while outputting to a huffy avi via an input avs file with the source being some vobs, I get multithread support!? I'm doing 30fps! in vdub with on-the-fly IVTC in the avs and both cpus are being utilized 95% accoding to WinXP... Can someone explain?

BTW, here is my avs file since filters do make a difference.

LoadPlugin("C:\Virtualdub\MPEG2DEC.dll")
LoadPlugin("C:\Virtualdub\ivtc.dll")
mpeg2source("F:\Sample.d2v")
ivtc(30,15,100)

[trbarry]

Yep, I believe that for some time Vdub has used a different thread for the input and output. So if the 2 are fairly balanced loads then you get free multiprocessor support.

- Tom

[LB]

Sweet man. I was using tmpg before and only averagin 15fps in vdub with 30min proccessing on a 20min clip in tmpg. Now, I go 2x as fast with no processing in tmpg. MUHAHAHA. And with cce moving at 3.2 I'm in heaven. ;p

[Acaila]

I believe you should be getting much better speeds than only 30fps with dual 1800's especially with RAID0 (since I can get over 60fps) even with the ivct filter in there. Are you sure you've got VDub on fast processing mode?

[LB]

Let's make sure we got the same setup.

1. You use ivtc
2. You are outputting to a huffy avi
3. Your source is a vob

With just that, I do 32fps. What about you?

[Acaila]

With those settings I get 26fps. Encoding to XviD though is about 5fps faster.

Disabling IVTC gives me over 50fps for a huffy and about 60fps for XviD (which is weird, because saving in uncompressed format is supposed to be faster than compressed, unless my HD is limiting me).

I never knew the IVTC plugin slowed everything down so much (good thing I never have to use it). Sorry for the false alarm.

May I ask why you are outputting from a VOB to Huffy? VOB's are already compressed with MPEG-2, so encoding that with a lossless codec will only give you a huge filesize, but no quality increase/decrease.

[LB]

heh, well we are talking apples and oranges then man. you're doing xvid, something totally different than what I'm doing. I'm making svcds, and for those you obviously want to convert to huffy avis since CCE will speed up three fold if its in a huffy avi format versus a mpg2 format. welp, cheers heh.

[TotalChaos]

It seems I've finaly found a thread that is almost exactly concerning the things I want to know. I was under the illusion that virtualdubmod was actualy multithreaded. when in fact it was as you describe above as realy being TWO diff single CPU apps. (avisynth and I assume virtualdubmod) It was mentioned that if the load was semibalanced then the utilization of two CPUs (or all of a single HT CPU) would be nearly 100%. This makes sense, assuming I am rationalizing correctly, because one program wouldnt be able to tranfer video data faster then the other could handle. To add to this headache of "how fast can I get my computer to create DivX/Xvid content and what kind of hardware is best suited for this task?" Hyperthreading CPU have made it so that one CPU appears as two to the OS. Through the hardware sites I like to visit I've found that HT CPUs are still SINGLE CPUs, and that that single CPU's resources (ALU, FPU, cache, ect) are shared among the two threads being proccessed at any one time. Which is faster? A HT P4 3.2Ghz 800 FSB or DUAL Zeons running at some slower speed of say 3.06GHz and slower FSB of 533MHz both with 1GB RAM. Avisynth having one WHOLE CPU to itself and virtualdubmod having one WHOLE CPU to itself, looks like a better choice, but who can tell without having this equipment. I'm realy just hoping somone on this forum reads this post and has enough info on these two programs to hypothesize an outcome of these two configs. The question you might be thinking now is, "what type of filters are running and all that." Who can tell this is all hypothetical perhaps some worst case scenario resizing, IVTC, noise filters, and for encoding GMC, Qpel, B frames. I know this sound hysterically ridiculous, maybe somone wouldn't mind humoring me with a post on what they think would be the better machine or for that matter what could be the best if these two set ups are not the best!

[APF_Gandalf]

Originally posted by TotalChaos
Who can tell this is all hypothetical perhaps some worst case scenario resizing, IVTC, noise filters, and for encoding GMC, Qpel, B frames.
you forgot VHQ=4 in Xvid ;p
btw, I just bought a dual xeon 2.4, FSB 533 to encode, I'll soon make some tests with "heavy" (for old original japanese Kurenai Sanshiro episode) and "light" filtering (other NTSC animes, DVD sources)

[TotalChaos]

Hats off for you. I have been looking for some results from a dual CPU set up verses single HT CPU for quite a bit. Also it would be interesting to see how HT, being on and off during your ZEON test runs, affect encode times. Would there be more then 2 threads running to take advantage of the virtual quad CPUs? **anxiously awaiting the results**

[APF_Gandalf]

ok, finally, the comp is complete and quite ready to use.
2 xeon 2.4@3.0 Ghz, I still miss the serial ATA HD that I'll get this week end and it'll be ok. On the few tests I made with an ATA 100 HD, it seems that HT and the second CPU are used when you filter with avisynth so the use of the first CPU is always near 100% and the usage of the second depends of the filtering.
some tests with scripts, Xvid/ffvfw/Quenc/CCE settings and encode times really soon.

edit: typo

[TotalChaos]

Havn't forgot about this thread, I'm still listening! I'm not sure the HDD will be much of a bottleneck if any at all though.
*Egerly awaits further results*

[LB]

Didn't know this thread was still alive ;p I've long since upgraded to dual XEON's OC'd to 3ghz each. I've noticed this issue more in the area of MPEG2 encoding with CCE.

There are folks with a 3.2ghz P4 who get about 3.0 in CCE, whereas my dual 3ghz XEONS gets me only 2.9-3.1 in CCE. So what gives? Well, I tried installing XP with only 1 cpu enabled, and ran CCE again and scored about 2.8. So, why isn't my dually setup faster? Well, the answer lies in the application driving it, and here is where the answer applies in all realms, like XVID too, not just MPEG2 encoding.

A program can only use as many threads as it's designed too. I don't know anything really about programming using threads, but the general concept is that your cpu will only use as much as the program can send it. With my single CPU, I was at 100% usage and got a score of 2.8. With both CPUs, I was at 60% usage on each and scored 3.1. So this tells us that CCE is not programmed as efficently as it could be for dual cpu support. So what about single cpu hyperthreading?

Well, it depends. Assuming you have EITHER: (1) a single P4 w/HT; OR (2) a dully xeon, you will usually find that Virtualdub + Avisynth will easily max out any single cpu rig, so enabling hyperthreading not a real factor because the cpu will already be maxed out. As for a dually rig, each program will auto set itself to a respective cpu, so there is nothing to worry about here.

So what's the moral? The moral is, with CCE, you won't get a giant speed increase in going from a single cpu setup to a dually setup because of CCE's poor multithread programming. But, you will free up a lot of extra cpu usage so that you can do a lot of background work while cce encodes. As for VirtualDub + Avisynth, you will definitely notice a speed increase, usually on the order of 100% in going from single to dually. Hope that helps.

[DSP8000]

You can try MainConcept Encoder if you want.
Supports multi-processor encoding.
And yeah, CCE it's not designed for dual system encodings.

DSP8000

[TotalChaos]

Are there any differences when running one of your Zeons with HT enabled vs running both Zeons without HT enabled? Seems that would be the perfect comparison to see wether HT is realy that much an improvment over non HT CPUs. Also it would answer wether going for a faster P4 or slower Dual Zeons. I use gordian knot, I guess that means I use avi synth + virtualdubmod only. Perhaps a test with these programs using some heavy filtering!!!! I'm ready for faster encoding and will soon be creating another system. This info is extremely helpful for making a decision.

[LB]

Originally posted by TotalChaos
Are there any differences when running one of your Zeons with HT enabled vs running both Zeons without HT enabled? Seems that would be the perfect comparison to see wether HT is realy that much an improvment over non HT CPUs. Also it would answer wether going for a faster P4 or slower Dual Zeons. I use gordian knot, I guess that means I use avi synth + virtualdubmod only. Perhaps a test with these programs using some heavy filtering!!!! I'm ready for faster encoding and will soon be creating another system. This info is extremely helpful for making a decision.

Re-read my above post again. Granted, it's kind of confusing at first pass, but the answer is in there. But, I'll restate it again anyway. :D
HT will help in (1) and only (1) situation. The situation has (2) requirements:

a. Current CPU usage is BELOW 100%; AND
b. The program has multithread support OR you're running multiple programs

All hyperthreading does is allow XP to see TWO virtual processors, instead of your ONE physical processor (note: it will see 4 if you have: 2 cpus + ht enabled + xp pro), and when xp sees TWO virtual processors, it will "split" the load between them. How? Two ways. ONE: if the program has multithread support, like CCE (but CCE's is poor), then the single program will use both virtual processors. OR, TWO: if your program does not have multithread support, but you run multiple programs, XP will balance the load. A perfect example of the second situation is with virtualdub + avisynth. They both do NOT have multithread support, but they are two seperate programs. So with hyperthreading enabled, XP will run avisynth on virtual processor #1, and virtualdub on virtual processor #2.

Now the key is to understand BOTH requirements (a+b) stated above. If the load on your CPU when encoding is already 100%, then HT will NOT help you in encoding, but it also won't hurt you because for other applications, applications which don't use 100% of the cpu, XP will balance the load. In the case of Vdub+Avisynth, you're generally already maxed at 100%, so HT will not help you. With CCE, you're already maxed at 100%, so HT won't help you.

Hope that did a better job of explaining.

[TotalChaos]

I think you missunderstood my question. My question wasn't, "will Hyper Threading out perform an identical non hyper Threaded CPU machine while running CCE OR the avisynth + virtualdubmod combo." It also wasn't, "will Hyper Threading increase the performance of a DUAL ZEON machine while running CCE OR the avisynth + virtualdubmod combo." It's true HT makes one CPU appear as two CPUS to the OS. There is more then just that. HT, according to the articles I read, also makes the CPU more efficient. It keeps more of the registers occupied then running single threads at one time. It sounds to me that even though this doesn't create two complete CPUs, it does makes one CPU more powerful if one multithreaded program or two single threaded programs working on the same project is/are being used. Another diff between a HT CPU and two physical CPUs are the resources a program has at it's disposal. With a HT CPU, core components (FPU, ALU, cache) of that SINGLE CPU are being shared among each thread running on that CPU. With multiple CPUs you have CPU-count times the resources available for program threads to use.
My Questions: Assume the programs run are AVIsynth + virtualdubmod combo for all testing!!
1. Does a DUAL ZEON machine encode faster using only one ZEON HT enabled (IE having only one ZEON in the system) or two ZEONs HT disabled on both? The Idea is to find out if HT can give me DUAL CPU potential for encoding.
2. Money aside, Would it be in the best interests of a buyer to purchase a P4 3.4GHz HT CPU and supporting components OR dual 2.8GHz/3.06GHz ZEONs and supporting components? I ask because I assume HT has an advantage over non HT in our above encoding tests.

I just want to know if springing the extra bucks for an actual dual cpu machine is going to increase my encoding speed over the 3.0GHzOCed3.25Ghz HT P4 I am using now. Will two CPUs cruch faster then one with avisynth + virdualdubmod even if that one CPU runs at a higher clock speed and has HT?

[LB]

Originally posted by TotalChaos I think you missunderstood my question. My question wasn't, "will Hyper Threading out perform an identical non hyper Threaded CPU machine while running CCE OR the avisynth + virtualdubmod combo." It also wasn't, "will Hyper Threading increase the performance of a DUAL ZEON machine while running CCE OR the avisynth + virtualdubmod combo." It's true HT makes one CPU appear as two CPUS to the OS. There is more then just that. HT, according to the articles I read, also makes the CPU more efficient. It keeps more of the registers occupied then running single threads at one time. It sounds to me that even though this doesn't create two complete CPUs, it does makes one CPU more powerful if one multithreaded program or two single threaded programs working on the same project is/are being used. Another diff between a HT CPU and two physical CPUs are the resources a program has at it's disposal. With a HT CPU, core components (FPU, ALU, cache) of that SINGLE CPU are being shared among each thread running on that CPU. With multiple CPUs you have CPU-count times the resources available for program threads to use.
EDIT: I didn't misunderstand. You were merely confusing. Here is your original post: "Are there any differences when running one of your Zeons with HT enabled vs running both Zeons without HT enabled?" You used the word "enabled". I was under the impression since you used the word "enabled" that you knew XEONS have HT in them, apparantly not though. The rest of the post futher drives in this point :p

Originally posted by TotalChaos 1. Does a DUAL ZEON machine encode faster using only one ZEON HT enabled (IE having only one ZEON in the system) or two ZEONs HT disabled on both? The Idea is to find out if HT can give me DUAL CPU potential for encoding.
Again, I answered this above. I'll give you a example now:

Avisynth + VDUB

DUAL XEON RIG: 3.0 GHZ + HT ENABLED | Speed = 2x
DUAL XEON RIG: 3.0 GHZ + HT DISABLED | Speed = 2x
SINGLE P4 RIG: 3.4 GHZ + HT ENABLED | Speed = 1.2x
SINGLE P4 RIG: 3.4 GHZ + HT DISABLED | Speed = 1.2x

EDIT: And just another note. You're being "confusing" here again because when you say: "Does a DUAL ZEON machine encode faster using only one ZEON HT enabled (IE having only one ZEON in the system)" it looks like you understand that XEONS have HT built into them. But at the end of the sentance where you say "IE" it, bleh, I can't even explain how illogical that sentence is. You start by saying a "DUAL ZEON" then "ONE ZEON HT ENABLED" then "ONE ZEON IN THE MACHINE." If that isn't the most confusing sentence I've ever read in my life, then I'd like to see what is. :p

EDIT EDIT: Ahh, one last thing I thought that might be confusing. When I write "enabled" or "disabled" it does NOT mean the CPU does not support HT instructions. Rather, it means in the BIOS I "enabled" or "disabled" HT instructions.

Originally posted by TotalChaos 2. Money aside, Would it be in the best interests of a buyer to purchase a P4 3.4GHz HT CPU and supporting components OR dual 2.8GHz/3.06GHz ZEONs and supporting components? I ask because I assume HT has an advantage over non HT in our above encoding tests.

XEON chips HAVE HT in them.
P4 chips HAVE HT in them.

And again, the above example should further explain that a DUAL XEON system is twice is fast as a single p4 system.

Originally posted by TotalChaos I just want to know if springing the extra bucks for an actual dual cpu machine is going to increase my encoding speed over the 3.0GHzOCed3.25Ghz HT P4 I am using now. Will two CPUs cruch faster then one with avisynth + virdualdubmod even if that one CPU runs at a higher clock speed and has HT?

Again, I feel like I've already answered the question three times now. I keep saying the same thing. HT or NO HT, it doesn't matter when it comes to encoding. If you want the power of TWO cpus, then you have to BUY two cpus. HT does nothing in terms of encoding.

[TotalChaos]

What makes you think that HT will ONLY help if your CPU usage is below 100%. If TWO computation intensive threads are competing for the same computing time on a non HT CPU there will no doubt be a 100% load presented to that CPU. Enabling HT allows the OS to schedual more time for threads. I've tested this myself on my P4 running AVIsynth + virtualdumbmod. With HT enabled the total CPU usage is from 80% to 95% with brief peaks at 100%. With HT disabled the CPU usage remains consistantly at 100%. This tells me your statment, "Virtualdub + Avisynth will easily max out ANY CPU" is incorrect. I also ran a few tests with HT enabled and disabled. I used the same vob files and ran the exact same scripts on them. The outcome: WITH HYPERTHEADING ENABLED I RECIEVED A LOWER ENCODING TIME, and had some CPU cycles to spare. You can prove this using only one of your ZEONs as I am WELL aware ZEONs are equiped with HT.

I understand that even a HT enabled CPU is going to pale when compared to two physical CPUs of the same clock speed and bus speed. The problem when comparing ZEONs and P4 are many fold. One, you have a large diff in bus speeds. (533MHz ZEON - 800MHz P4) Two, top model P4 have higher clock rates then top model ZEONs. (3.4GHz P4 - 3.2GHz ZEON) Considering the cost at this point, a DUAL ZEON more then doubles the price (nearly triple), but does it offer double the encoding power then that of a HT P4. If the DUAL ZEON does offer double the proccessing power (for encoding with virtualdubmod + avisynth) one could purchase a system with two 2Ghz ZEONs and still out pace the 3.4GHz P4.

To finaly ask this question in a manner we all can understand, Would the higher clock speed, higher bus speed, and the HT feature of the P4 allow it to keep pace with a DUAL ZEON??????

[LB]

Well if you say you have done tests and recieved 100% load with HT disabled, and then 80-95%, then results don't lie. I didn't experience that myself though, but I can say my fps speed doubled in vdub.

As for BUS speed, I believe it's overrated when comparing 533 v. 800. Sure, there is a performance increace, but it's no more than 10%.

The answer to your coherent :p final question, depends. I don't believe they make 2ghz @533. I think 2.4ghz is the lowest, could be wrong, but that's what I think. So why does the answer depend? Well, if you're in the market for (and I have to correct you here, it's XEON, not ZEON) a dual XEON system, I'd wait until June. Come june, the 3.4 ghz (and possibly 3.6) XEON @ 800mhz bus will be released, and it will eat the p4 for lunch. So you WILL recieve 2x the performance of a comparable single p4.

What about now? Well if you do some reading on the best dual cpu site in the world, 2cpu.com, you'll find that if you get an IWILL DP533 board, you can slap in 2 x 2.6ghz chips @533, and OC them to an air-cooled 3.6ghz each. That is the only board dual board which will allow you to do this.

Finally, since you're considering a new pc. You might want to hold off on a dual xeon, and go dual opteron. Reason I say this is because AMD's recent press release last week said come Q1 next year, or thereabouts, they will release new DUAL CORE chips, which have TWO physical cpus inside each chip. Best part is, that they are backward compatible with existing opteron boards. So, if you buy a single opteron board now, next year just upgrade to a dual core chip, and you now have a dually board. ALSO, and this is the cool part, the same goes for a existing dually boards. If you buy a dual opteron board now, come next year you can upgrade to two dual core chips, and you will now have 4 cpus. This dual core thing is nothing like HT. Instead of two virtual processors, you get two physical processors on a single chip (die). ;)

My recommedations? Well, I want to upgrade myself this summer, and would rather not wait for dual core chips. I'll probably end up getting dual xeon 3.6ghz @ 800mhz, and try to overclock them to 4ghz each. But if you can't wait even until summer, then do a search on 2cpu.com for "DP533 3.6" or something like that, and get the DP533 with some 2.6 chips. (Reason I didn't do that is because the DP533 doesn't have a 64bit PCI slot, so my raid card wouldn't work in it. The DPI533 does have a 64bit slot, but it doesn't have the OC capatilities like the DP533. :( )

[TotalChaos]

LMAO, your competely right, XEON haha.... I knew somthing looked wrong with that. Most of my posts have been written when I should have been sleeping and my main concentration was on my point not so much spell/grammer.
Thanx for the link. I've been looking for some 'serious' SMP info for quite awhile. I have read about both Intel and AMD using multi-core CPUs within the next few years. I didn't read about them coming so soon though.... how could I have missed such an event?? aaaahhh!!!
One Small argument; I've read on this forum, on tech sites, just about everywhere that when it comes to encoding Intel's P4/XEON are faster when compared to similar rated AMD AthlonXP/AthlonMP/Athlon64 2800+ - FX53. I can only assume the opteron, being nearly identical to the AthlonFX, will not present any faster encoding speed then an AthlonFX. I've often wondered myself wether going with the opteron and their highly scalable on board memory architecture. Even having dual cores CPUs (4 CPUs) wouldn't help out in our case as we only have 2 threads. (1-AVIsynth + 1-VirtualDubMod = 2) What are your thoughts on these matters?

[Cyberia]

FYI: The only reason the Pentium4's encode faster is because they perform MMX/SSE/SSE2 at 3+ Ghz and the Athlon's run at 2+ Ghz.

[TotalChaos]

Look, we should all agree that Intel and AMD make very powerful CPUs. It is also well known that each CPU has it pros and cons. I own 14 computers 3 of which are AMD. All three are dependable and perform their jobs better then adequate. Where the matter of encoding video is concerned it is well known the Intel P4 has the upper hand over AMD's AthlonXP/Athlon64/Athlon64FX. My statement wasn't in any way a putdown to AMD. If I missunderstood your statement, "The only reason the Pentium4's encode faster is because they perform MMX/SSE/SSE2 at 3+ Ghz and the Athlon's run at 2+ Ghz." as a comeback to my earlier post about opterons then disregard this post.

[LB]

Originally posted by TotalChaos
One Small argument; I've read on this forum, on tech sites, just about everywhere that when it comes to encoding Intel's P4/XEON are faster when compared to similar rated AMD AthlonXP/AthlonMP/Athlon64 2800+ - FX53. I can only assume the opteron, being nearly identical to the AthlonFX, will not present any faster encoding speed then an AthlonFX. I've often wondered myself wether going with the opteron and their highly scalable on board memory architecture. Even having dual cores CPUs (4 CPUs) wouldn't help out in our case as we only have 2 threads. (1-AVIsynth + 1-VirtualDubMod = 2) What are your thoughts on these matters?

Yes, but if you look at the FX-53? scores in encoding, which is essentially a opteron, there isn't a drastic difference between the P4 EE and the opteron. It's important to note that a opteron beats out a "normal" P4 when it comes to encoding, but not a P4 EE, and a 3.2Ghz P4 EE single chip costs $1,000; which is quite a lot more expensive than a opteron. Also, in the future if one did have a 2cpu system, with each cpu having two cores, even though only two out of the four cores would be utilized, you'd still be free to burn to a dvdr and chat on irc, and download 10+ files, while playing unreal tournament 2004 at 100+ fps at 1600x1200 :p Or, you could simply open TWO virtualdub sessions and thus have all four cores maxed out. :p

http://www.anandtech.com/cpu/showdoc.html?i=2002&p=8
Shows the FX-53 tied w/the p4, but not the EE

http://www.tomshardware.com/cpu/20040318/athlon-fx53-19.html
Shows the FX-53 beating even the P4 EE

So there you have it.

[LB]

Just a small update. I haven't visited 2cpu in ages, and decided to drop by last night. I found out something quite interesting which is worth noting in this thread. The new M0 stepping xeons are available. What does that mean? Well M0 stepping means they are rated at 800mhz bus, but are underclocked to 533mhz, plus a hell of a lot cheaper than a officla 800. People have bought the new dual xeon DH800 board by iwill which with a small mod, allows these chips to be run at 800mhz and above.

The current STABLE AIR overclocks are:

2 x XEON 2.4ghz @533mhz = base chip
2 x XEON 3.5ghz @970mhz = OC'd

Looks pretty tasty. Especially when you consider each chip is only $200 :p :p :p

[TotalChaos]

Talk about a tuff decision!!! Hmmmm... would you rank an Opteron 248 setup as equal/better/worse then two of these stepping XEONs?? The opterons to have the scaling favor, like two dedicated CPU-RAM busses instead of sharing the same 800MHz buss.

[LB]

TWO 3.5GHZ Xeons @1,000 MHZ bus would eat two 248's for lunch. I also checked and I'm pretty sure the 248 != FX-53, rather the 248 = FX-51. Apparantly the opteron equivalent of the FX-53 aint out yet.

Plus, MB + 2 x 2.4ghz M0 xeons = $750

Xeons come with a badass stock fan, so no need to buy one. So basically you get all you need for $750, which is about $1500 less than a much slower opteron solution. :D I've decided this is EXACTLY what I'm getting. I have a DP533 board and 2 x 2.4's C stepping I believe; going to dump them and go with that Iwill board and two 2.4 M0 chips. Just be careful when you order the chips that they VERIFY you are getting M0 ones, because there are ALSO many non-m0 stepping 2.4's out there, and if you get the wrong one, then they will be worthless (only m0 stepping 533mhz chips are true 800mhz).

[TotalChaos]

Can you point me in the right direction? Where to buy and how to verify these MO stepping XEONs. And where to find the Mobo you decribe. Finaly a link to the info about the "small mod" you mention for that mobo.

[LB]

1. how to verify you get M0 stepping 2.4ghz chips
Remember, the chips are 2.4ghz @533, so we just need to verify they are m0 stepping. Call a place you find on pricewatch, ask them to verify physically they are m0 stepping while you are on the phone with them, and if they decline, hang up and try someone else. I'd stick with reputable companies like newegg and such because they may be more willing to do this. But obviously just don't order 2.4's and try to "get lucky" :p

2. what board?
http://www.newegg.com/app/viewProductDesc.asp?description=13-129-131&depa=0

3. thread on mod?
Remember, the mod is only necessary if you want to EXCEED 800mhz. If you are content with a simple 533->800 overclock then you can simply do this in the bios (I believe).
http://forums.2cpu.com/showthread.php?s=&threadid=50793

that's the thread, but you gotta wade through it yourself as I have yet to do the mod. lot of folks have, and they are more than willing to help, but you're gonna need to read through the thread yourself to get an idea of what's going on.

[geoffwa]

In my experience the two main limiting factors for encoding are memory bandwidth and what SIMD instructions your CPU supports.

Using these two criteria an Operton would probably be the best choice for an encoding box.

I'm not sure how well a multi-cpu Operton box would scale, given the NUMA of it all.

[LB]

Originally posted by geoffwa
In my experience the two main limiting factors for encoding are memory bandwidth and what SIMD instructions your CPU supports.

Using these two criteria an Operton would probably be the best choice for an encoding box.

I'm not sure how well a multi-cpu Operton box would scale, given the NUMA of it all.

That is assuming the equal of a 3.5GHZ Xeon @ 1000mhz is released. At the moment, the Opteron 248 is the dual version of the FX-51. The FX-53 can tie the P4 3.4 EE, so arguably the yet to be realeased opteron would tie a dual xeon 3.5 @1000, or at least come close to it. But why bother (unless you want to wait for dual core) if the total price for this xeon rig is ~$1400 less than the unreleased opteron?

[TotalChaos]

Talk about a mind reader! Do you, LB, have a connection at tomshardware guide???? Check this http://www.tomshardware.com/motherboard/20040514/index.html out if you havn't already. Guess not much that a single CPU has can compare with two physical CPUs. Most of the encoding benchmarks show the dual Xeons on top.

[TotalChaos]

I found this http://techreport.com/reviews/2004q2/opteron-x50/index.x?pg=1 while looking on 2cpu.com. Are these benchmarks acurate in your opinion? Seems like the opteron was givin more credit then deserved. Other sites like tomshardware OCed the opteron and it didn't seem to perform this well.

[LB]

Originally posted by TotalChaos
I found this http://techreport.com/reviews/2004q2/opteron-x50/index.x?pg=1 while looking on 2cpu.com. Are these benchmarks acurate in your opinion? Seems like the opteron was givin more credit then deserved. Other sites like tomshardware OCed the opteron and it didn't seem to perform this well.

Yeah, I think they are accurate benches. Here are the benches in question: HERE (http://techreport.com/reviews/2004q2/opteron-x50/index.x?pg=14)
And you can see the 250's brother (the FX-53) HERE (http://techreport.com/reviews/2004q1/athlon64-fx53/index.x?pg=5) and the benches match up fairly well.
Finally, you can verify through Tom's HERE (http://www6.tomshardware.com/cpu/20040322/prescott-09.html)

So, yes, I'd gladly get an opteron over a XEON any day, but I also don't feel like paying $1,400 more for it. Like I already said, you can get 2 x 2.4 M0 chips + a board for a mere $700, and overclock them at a stable 3.6ghz @ 1000fsb, and doing so will bring them ahead of the 3.4EE's most likely (although they do suffer from only 512k, instead of the EE's 2mb).

But still, if you check out MP3 encoding HERE (http://techreport.com/reviews/2004q1/athlon64-fx53/index.x?pg=7), which closely resembles DIVX encoding, you can see the P4 EE easily overtakes the FX-53, and the results should be similar on any Pentium benchmark for encoding, including XEONs. So the moral of the story is that yes, Opteron 250's are no joke, but they will hit you hard in the wallet at about $1,000 each, compared to $200 each for XEON 2.4's. Also, opteron boards aren't cheap. I almost have enough for the 2.4's and should have them up in running in a few weeks, so I can post a few benches then :D

[LB]

Originally posted by TotalChaos
Talk about a mind reader! Do you, LB, have a connection at tomshardware guide???? Check this http://www.tomshardware.com/motherboard/20040514/index.html out if you havn't already. Guess not much that a single CPU has can compare with two physical CPUs. Most of the encoding benchmarks show the dual Xeons on top.

Yep, when it comes to encoding, nothing currently beats a XEON, except maybe two XEONS :D