intel T8300 vs quad 9300 speed test

[jdobbs]

In most cases that will actually slow it down. That's the reason DVD-RB defaults to NOSMP when multiple encoder instances is selected.

[jdobbs]

mediacoder works as it should be at 4x speedAs has been said earlier... that means absolutely nothing. Making that comparison is like telling me pancakes are different than doghouses.

[fjhdavid]

Jdobbs

when I said HC_NOSMP change nothing, it was in term of fps speed, at least for my computer. I am sure it changes a lot of things in the way HC is working. But the solutions is not with this parameter for my configuration.

I thought that the mediacoder example show that my quadcore can go 4x faster with 4 cores, but you are right, let's talk about dvd-rb


Groucho 2004

I have 4GB of RAM. But, I did a test under windows XP with only two cores (I wrote "Encode_Processes=2" in the rebuider.ini) and I saw "multiple encoder process (2)). Then the memory size used by DVD-RB was cut by half and no more a problem. And, as I said before, I have the following results:

one core : 8fps
two cores: 2x 4fps = 8ps (on my laptop with a core2duo I have around (2x 7.7fps =15.4 fps)

so the speed problem occurs with 4 cores AND 2 cores.


Jdobbs and Groucho

Could the problem be with the chipset driver as I use a nforce 630i? or with the wireless kayboard driver which is maybe installed from the beginning?

Jdobbs what is your chipset?

because I re-installed everything from the beginning, only with a genuine windows XP SP3 and dvd-rb....

[jdobbs]

If you are running a program that has SMP enabled (which you have in your example), it is NOT GOING TO RUN 4X FASTER WITH A QUAD CORE. The single instance is already using all four cores. That's also the point with the example of mediacoder. It only runs 4x faster because (apparently) it is not multithreaded and only uses one core when it you run a single instance.

The chipset isn't the issue.

I think I'm going to have to bow out of this discussion. I feel like I'm on the receiving end of a one-way conversation.

[fjhdavid]

just to be sure:

in order to enable or disable SMP under XP, I have just to write:

HC_NOSMP=0 or 1 in the "options" section of rebuilder.ini, save the file and start dvd_RB?

[SpazzHH]

Or just make sure you remove the existing line. DVD-RB reverts to off by default. Also, I would definitely disable your anti-virus software that you have running on your machine when you're running DVD-RB. It can sure play havoc with encoding times.

[fjhdavid]

Jdobbs and others, I finally understood what you explained to me! You were completly right as DVD-RB is always multithreaded...

I made different test with my two computers and I think the speed difference between them could be in the memory bandwith (as Sandra Sisoft told me).

Anyway, If you have to choose a computer and you are using avisynth 2.57 and very intensive filters like FFT3D, the following results with HC encoder could help you (both computers run vista and have 4GB RAM):

- with "multiple encoder =1" my 2.1GHz core2duo 45nm is on par with my quad 9300 at 2.5ghz (6.5fps vs 6.6fps)

- with "multiple encoder =2" my 2.1GHz core2duo 45nm is better than my quad 9300 at 2.5ghz (2x5.5fps vs 2x4.5fps), surprising but that's it.

- with "multiple encoder =4" my quad 9300 at 2.5ghz gave me 2x2.2fps, I hoped a little bit more but that's it.

Conclusion: with my specific computers (for example my quad PC uses a single memory channel with a "limited" 4.4GB/s bandwith), the best option is to choose a laptop with a core2duo...and if you want to stay with a quadcore then there is no significant difference between 2 and 4 processes.

it should be intresting to have a "test script" and that users send their results as it seems that the processor speed is not the only key of performance....and that more cores (if you are not looking to others part of the PC) doesn't give always more speed...

[Groucho2004]

Jdobbs and others, I finally understood what you explained to me! You were completly right as DVD-RB is always multithreaded...
Not true. DVD-RB is single threaded and just (sorry about the "just", jdobbs, you know what I mean) a frontend for the decoder, encoder, etc.

I made different test with my two computers and I think the speed difference between them could be in the memory bandwith (as Sandra Sisoft told me).
The influence of memory speed is much smaller than many people think, particularly wth the Core 2 or Quad architecture.

Conclusion: with my specific computers (for example my quad PC uses a single memory channel with a "limited" 4.4GB/s bandwith), the best option is to choose a laptop with a core2duo...and if you want to stay with a quadcore then there is no significant difference between 2 and 4 processes.
Your conclusion is probably flawed because I still think that something is not configured correctly and that you don't fully understand how DVD-RB and the encoders work together, i.e. multiple encoder processes/multithreaded encoder.

[fjhdavid]

Groucho2004

you are right, DVD-RB is a front end and it is HC encoder which is multithreaded. SMP is active if "multiple process=1" is choosen in DVD-RB, and SMP is inactive if "multiple process>1" is choosen in DVD-RB. But Jdobbs gave us the possibility to additionally tweak multithread/multiprocess of HC encoder with the help of the "HC_NOSMP" parameters.

So I was wrong from the beginning in the way HC encoder "dispatch" among the different cores.

Anyway, I tested all the possibilities (I know) with "HC_SMP" and "multiple process" parameters under Vista and XP. And the results I gave are what I have.

As I have no others software installed on my computer, maybe I can play with others parameters than the two others I mentioned but I don't know them

[Groucho2004]

@fjhdavid

Just one thing that strikes me as odd:

On Nov 16th you posted this:
my results are with dvd-RB pro 1.28.2:

for a PAL movie 704x576 with HC encoder 0.23 "best profile" without "lossless" option, I have with a quadcore Q9300 at 2.5Ghz:

with 1 core: 140fps for 1st and 2nd pass
with 4 cores: 40pfs for each core and each pass

At that time your quad seems to have worked just fine...

[fjhdavid]

the problem is that with my laptop I have 140fps with one core and 2x100fps with two cores.

I said I made a mistake in my way to understand how HC encoder "dispatch" over the cores (in the example I was expecting 4x140fps which is ridiculous when you know how HC encoder works).

But what is surprising from the beginning is that (remember that my laptop is 2.1Ghz and my quadcore is 2.5Ghz)

when I use "multiple process=1" I have similar results between my laptop and my Quadcore (I hopped a little bit more from the quad due to its frequency advantage)

but when I use "multiple process=2", the laptop start to become better....by about 20% !!!

and when I use "multiple process=3 or 4" the quadcore doesn't show any better result than with "multiple process=2"....as if somewhere there was a bottleneck....

that is my point from the beginning