XP with CS3 is faster than vista with CS4?

resolve technical issues related to use of Neat Video
Post Reply
sonics
Posts: 11
Joined: Mon Feb 16, 2009 7:01 pm

XP with CS3 is faster than vista with CS4?

Post by sonics » Tue Mar 17, 2009 6:42 pm

I own an E6400 @ 2.6Ghz

With windows XP 32 bit with premiere CS3.20 I rendered a 10 second 1440x1080 clip with temp filter 1 and adaptive 1

I did the same with windows vista 64 bit with premiere CS4.00

result:

XP 32 -> CS3 : 4 minutes 16 seconds
Vista 64 -> CS4 : 5 minutes 46 seconds


How can this be? :s

I expected it to be the same at least.

NVTeam
Posts: 2260
Joined: Thu Sep 01, 2005 4:12 pm
Contact:

Post by NVTeam » Tue Mar 17, 2009 10:53 pm

Vista is slower than XP, a lot of people have been discussing that all over the Internet..

Vlad
Image Image Neat Video team
noise reduction for video and photos

ECB
Posts: 11
Joined: Thu Aug 21, 2008 6:14 pm

Post by ECB » Wed Mar 18, 2009 2:32 pm

I suggest you rerun the same test but use CS3 on Vista X64. There are some performance issues with PP CS4.

EB

jpsdr
Posts: 196
Joined: Mon Aug 11, 2008 7:33 am

Post by jpsdr » Thu Mar 19, 2009 8:32 am

Hello.

Maybe simply try XP 64 and not Vista 64 ?

sonics
Posts: 11
Joined: Mon Feb 16, 2009 7:01 pm

Post by sonics » Sat Mar 21, 2009 6:09 pm

The update from cs4.00 to cs4.01 did some good.

I did all tests now:

1440x1080 clip 15 sec neat video rendering

XP32 + CS3.20 5 min 50
xp32 + CS4.01 6 min 05

vista64 + CS3.20 5 min 56
vista64 + CS4.01 6 min 29

1440 x 1080 H264 rendering showed similar results (1 min 40 clip no filters)

XP32 + CS3.20 8 min 50
xp32 + CS4.01 10 min 15

vista64 + CS3.20 9 min 05
vista64 + CS4.01 10 min 50

some performance loss when using CS4 and also with vista64

sonics
Posts: 11
Joined: Mon Feb 16, 2009 7:01 pm

Post by sonics » Sun Mar 22, 2009 9:36 am

Today I upgraded to a Q9400 and put it on 3.2 Ghz
In theory it should be 120% faster than my E6400 2.66 Ghz

It showed some interesting results:

Neat video rendering were around 72% faster

H264 encoding (no filters) were around 130% faster

somehow when rendering the work area in CS3 / 4 neatvideo doesn't benefit the CPU boost as much as H264 encoding?

NVTeam
Posts: 2260
Joined: Thu Sep 01, 2005 4:12 pm
Contact:

Post by NVTeam » Sun Mar 22, 2009 10:30 am

Perhaps the encoding really needs some of the new processor instructions available in Q9400 but not available in E6400, while NV doesn't need/use them. The most important parameters for NV to run faster are processor speed and memory bandwidth.
Image Image Neat Video team
noise reduction for video and photos

sonics
Posts: 11
Joined: Mon Feb 16, 2009 7:01 pm

Post by sonics » Sun Mar 22, 2009 10:27 pm

NVTeam wrote:Perhaps the encoding really needs some of the new processor instructions available in Q9400 but not available in E6400, while NV doesn't need/use them. The most important parameters for NV to run faster are processor speed and memory bandwidth.
Well I had a 2.66Ghz dual core, now a 3.2 Ghz quadcore, that is 120% more processor speed. My memory bandwith went from 333Mhz to 400Mhz.

That's why I wondered why rendertime only decreased 72%. I did another test with a 15 sec 1440x1080 clip:

1 core enabled: 07 minutes 11
2 cores enabled: 04 minutes 09
3 cores enabled: 03 minutes 35
4 cores enabled: 03 minutes 24

It seems neatvideo (in combo with premiere?) does not realy benefit using core 3 and 4.

My E6400 @ 2.66Ghz did this in 5 minutes 50
My Q9400 with 2 cores enabled in 4 minutes 09, is 40% faster.
20% of it due to more clock speed and the rest new instructions?

ah well maybey i go too much offtopic here for this forum.

NVTeam
Posts: 2260
Joined: Thu Sep 01, 2005 4:12 pm
Contact:

Post by NVTeam » Sun Mar 22, 2009 10:56 pm

My interpetation of the figures for 1, 2, 3 and 4 cores - the memory bandwidth of that specific computer is the main bottleneck in the computation process when more cores are enabled. With 1 or 2 cores, the bottleneck is the processor speed, with 4 cores - memory bandwidth. The change from 333Mhz to 400Mhz (20% increase) is clearly not significant enough to satisfy the needs of 2 times more cores doing two times more of memory intensive work.

So, Q9400 does work faster than E6400, but its potential is not fully utilized because of the limited memory bandwidth.

BTW, I think Q9450 would better (than Q9400) cope with the bandwidth problem thanks to its two times larger L2 cache.

Vlad
Last edited by NVTeam on Mon Mar 23, 2009 10:23 am, edited 1 time in total.
Image Image Neat Video team
noise reduction for video and photos

sonics
Posts: 11
Joined: Mon Feb 16, 2009 7:01 pm

Post by sonics » Mon Mar 23, 2009 8:55 am

NVTeam wrote:
BTW, I think Q9450 would better (than Q9400) cope with the bandwidth problem thanks to its two times larger L2 cache.

Vlad
How fast does your Q9450 render a HDV clip? Is it a big difference or like 10-15% compared to the Q9400?

NVTeam
Posts: 2260
Joined: Thu Sep 01, 2005 4:12 pm
Contact:

Post by NVTeam » Mon Mar 23, 2009 12:10 pm

Q9450 3.3GHz, XP32, Pr CS4.0.1, 1440x1080 25p, 15 sec, filtered by NV (rad 1), encoded into h264: 4 min 55 sec

Q9450 3.3GHz, XP32, VirtualDub, 1440x1080 25p, 15 sec, filtered by NV (rad 1), uncompressed output: 2 min 45 sec

Vlad
Image Image Neat Video team
noise reduction for video and photos

sonics
Posts: 11
Joined: Mon Feb 16, 2009 7:01 pm

Post by sonics » Mon Mar 23, 2009 6:15 pm

NVTeam wrote:Q9450 3.3GHz, XP32, Pr CS4.0.1, 1440x1080 25p, 15 sec, filtered by NV (rad 1), encoded into h264: 4 min 55 sec

Vlad
VBR 1 pass?

NVTeam
Posts: 2260
Joined: Thu Sep 01, 2005 4:12 pm
Contact:

Post by NVTeam » Mon Mar 23, 2009 6:57 pm

Most likely (my colleague who did the test is offline now).

Vlad
Image Image Neat Video team
noise reduction for video and photos

sonics
Posts: 11
Joined: Mon Feb 16, 2009 7:01 pm

Post by sonics » Mon Mar 23, 2009 10:24 pm

NVTeam wrote:Q9450 3.3GHz, XP32, Pr CS4.0.1, 1440x1080 25p, 15 sec, filtered by NV (rad 1), encoded into h264: 4 min 55 sec

Vlad
with vbr 1-pass I did the same one in 5 min 48

That is 18% faster

since your q9450 is more overclocked than mine (running 3% faster + 3% more bandwith) it would mean that the 6M more L2 means 12% faster rendering.

btw cbr rendering was only 6 seconds faster.

good to know :) thanks for the efforts

Post Reply