disappointed in Phenom II x4 upgrade

[GunsMadeAmericaFree]

I've been playing around with several different cpu's recently. I started out with my trusty old Athlon II x2 2.8 GHz processor. Then I tried a Phenom II x2 Black Edition 3.2 GHz. Now I've switched over to a Phenom II x4 2.8 GHz.

I'm using Nero Recode to encode a DVD that I ripped to the hard drive. Same hard drive, same file, same exact encoding to the same size file. The only difference between the tests was what processor was doing the encoding.

Using my old Athlon II x2 2.8, I was getting through approx 6.xx% of the first pass encoding after one full minute. When I installed the Phenom II x2 3.2, that went up just a smidgen, to almost 7% completed after a minute. I overclocked the Phenom II x2 3.2 on up to 4 GHz, and then it completed about 9.6% of the first pass encoding after a full minute.

I figured I would probably get a better boost from a quad core processor running at default settings, since it had 4 cores instead of 2 - something more along the lines of an 80% increase. While the clock settings are the same (2.8 GHz), there are twice as many cores working on the encoding, right?

Well, I ended up with roughly the same exact % finished as the 2.8 GHz Athlon II x2 cpu. Where is all the benefit from having twice as many cores?

I've double checked with cpu-z, and it definitely is showing 4 cores active. I've also double checked with Nero, and Nero 9's version of Recode (updated to version 4.4.40.2) is supposed to use additional cores. (has ever since version 2)

So what is going wrong? Why am I pretty much getting the same exact results as a dual core processor running at the same clock speed?


Thanks!

 

[sm625]

I dont know about nero specifically, but I have used several encoders and none of them use more than one and a half cores during the first pass. Knowing what goes on during the first pass, this kind of makes sense. The second pass is what really pegs the cpu.

I've double checked with cpu-z, and it definitely is showing 4 cores active.

They may all be active, but that is just scheduler games. I bet your average cpu usage on the quad core during the first pass is about 38%. On the dual core it is around 75%.

On a side note, it does irritate me that they do not split the job into two parts and run two first passes simultaneously and then combine the results. That is the only way a quad core is going to help on the first pass. When we get to octo cores they will need to split the job into 4-6 parts. But then what if the first part is all high action scenes and the second part is mostly still backgrounds? Then you'd lose overall quality per file size. I really dont think there is anything they can do to speed up the first pass without totally rewriting the codec from scratch, which I am expecting to happen anyway. I think codecs are seriously lagging behind today's hardware.

 

[nanaki333]

nero recode probably doesn't support more than 2 cores. which is why you would be getting the exact same results as the dual core at the same speed. does task manager show all 4 cores lit up 100%, or is it 50% and all cores are just fluctuating up and down?

 

[sangyup81]

I'm not sure where these benchmarks were but the 3rd and 4th core don't help anywhere near as much as the 2nd core. The gains diminish as you get more and more cores.

Also your Athlon II X2 has 1MB L2 Cache per core while a Phenom II X4 has 512K L2 per core and tries to make it up with the 6MB of L3. Not all applications can take advantage of the L3 Cache

 

[KingFatty]

I think something like recoding may tax your memory system (CPU cache, and the separate SDRAM). i remember seeing an article here on Anandtech where boosting the memory settings noticeably affecting the performance of video encoding (e.g., nero).

So check if your motherboard reverted to a slower memory setting automatically after you switched chips, maybe you need to adjust the memory settings. For example, perhaps you have DDR3-1600 memory, but your computer reverted teh settings down to DDR3-1333 and you have to manually set them to the higher value.

 

[Arkaign]

Good lord, use Handbrake

 

[formulav8]

Good lord, use Handbrake

:thumbsup: Handbrake definitely Pegs all 4 of my Q9450 cores using h.264 encoding.

 

[richierich1212]

yep use handbrake for sure.

 

[apoppin]

I've been playing around with several different cpu's recently. I started out with my trusty old Athlon II x2 2.8 GHz processor. Then I tried a Phenom II x2 Black Edition 3.2 GHz. Now I've switched over to a Phenom II x4 2.8 GHz.

There's your problem. Your Ph IIs are all at stock clocks. Overclock your CPU and your will finally see a real tangible performance increase.

And pick SW that uses more than 2 cores when you benchmark to see any difference between dual and quad.

 

[soccerballtux]

it's tough finding encoders/software that can use > 2 cores.

 

[tweakboy]

Pull her up :0

 

[SonicIce]

tweakboy y r u so weird lol?

 

[Stuka87]

Nero only uses 2 cores. Use handbrake.

 

[greenhawk]

it does irritate me that they do not split the job into two parts and run two first passes simultaneously and then combine the results. That is the only way a quad core is going to help on the first pass. When we get to octo cores they will need to split the job into 4-6 parts.

only issue I can see in the short term for that approach is that you end up at the same situation as currently when encoding several files at the same time. Disk I/O limitations.

Once enough people have a SSD then the problem goes away again and multiple cores can work on difference sections, but until then, I do not expect going from 2 to 8 cores to show a massive improvement.

 

[KingFatty]

only issue I can see in the short term for that approach is that you end up at the same situation as currently when encoding several files at the same time. Disk I/O limitations.

Once enough people have a SSD then the problem goes away again and multiple cores can work on difference sections, but until then, I do not expect going from 2 to 8 cores to show a massive improvement.

I'd have to disagree, I believe that encoding video is massively processor limited such that you cannot hit the disk I/O limitations because your CPU and/or GPU cannot produce enough output to approach the I/O limitations.

Then again, maybe you are running a massive 50-core CPU beast that spits out huge I/O, but I don't think that's the situation here. Moving from dual core to quad core, and quad core to hex core, should produce noticeable differences, while being comfortably under any limitations of the I/O systems. But maybe I'm missing something, perhaps you are talking about some other type of video encoding that doesn't use much CPU/GPU when encoding?

 

[BD231]

Handbreak is sloppy compared to Media Coder x64.

 

[sm625]

Handbreak

Last time I tried to use handbrake it ended the same way as most other encoders... with the audio being out of sync and me not wanting to waste my time trying every single setting to no avail. Like another poster mentioned, running multiple intances (ie multiple jobs) is one way to load your cpu more for first passes.

 

[soccerballtux]

tweakboy y r u so weird lol?

? seems notrmal to me

 

[GunsMadeAmericaFree]

Hmm. On the first encoding at least, the graph of task manager cpu usage seems to imply that all 4 cores are being used. However, they are all being used at only about 50% of capacity.

86% utilization on second pass. Hmm, and I thought that a 1 minute sampling would be enough to compare the processors. I guess I should have encoded the full movie and compared that between the processors - I had no idea the encoding of first and second passes would differ like that.

This sort of implies that anyone still encoding single pass with more than 2 cores would not see any benefit. Is that true?

 

[TakeNoPrisoners]

Hmm. On the first encoding at least, the graph of task manager cpu usage seems to imply that all 4 cores are being used. However, they are all being used at only about 50% of capacity.

86% utilization on second pass. Hmm, and I thought that a 1 minute sampling would be enough to compare the processors. I guess I should have encoded the full movie and compared that between the processors - I had no idea the encoding of first and second passes would differ like that.

This sort of implies that anyone still encoding single pass with more than 2 cores would not see any benefit. Is that true?

86% utilization would mean something else is working in the background that isn't your program. Would be my guess.

 

[Arkaign]

Last time I tried to use handbrake it ended the same way as most other encoders... with the audio being out of sync and me not wanting to waste my time trying every single setting to no avail. Like another poster mentioned, running multiple intances (ie multiple jobs) is one way to load your cpu more for first passes.

I'm curious about this. I've never had audio sync issues, and I've mainly used Gordian Knot, Handbrake, and Nero back in the day when I had a lowly dual-core.

And you say not just Handbrake, but most other encoders as well?

Once in a while I see something that someone transcoded that is out of sync. It always puzzles me as to how that happened.

 

[sm625]

I get audio problems all the time. As well as crashes during encoding, and artifacting, etc. I have transcoded well over 2000 hours of video, mostly dvd vob files to avi, but also high quality 8000 kb/s stuff like from a camera to lower youtube 800 kb/s quality. I will only use the simplest programs nowadays (like movie maker) because the more complex ones always crash or fail.

 

[Idontcare]

I get audio problems all the time. As well as crashes during encoding, and artifacting, etc. I have transcoded well over 2000 hours of video, mostly dvd vob files to avi, but also high quality 8000 kb/s stuff like from a camera to lower youtube 800 kb/s quality. I will only use the simplest programs nowadays (like movie maker) because the more complex ones always crash or fail.

That really sounds like faulty hardware somewhere in the rig. You shouldn't be having such a consistently awful string of bad luck.

Have you tried TMPGEnc?

 

[tweakboy]

tweakboy y r u so weird lol?

SonicSupreme your right, I dont know why I made that last comment,

 

[rbk123]

I agree with Arkaign. I've done a ton of encoding via Handbrake and it has only crashed on me when I use it to stress test my overclock (it's excellent for that). Outside of my OC stability testing, it has never failed me *except* when I have the Handbrake settings/offsets wrong (this is my fault, not Handbrake's). Now that is easy to do since there are many settings but once familiar with them, it's a non-issue.

When I say fail - it results in a completely unplayable file. Not out-of-sync or choppy, simply unusable. When I correct the settings it works 100% of the time.