Study in CPU speed versus Memory bandwidth with my P4 2.4b C1...

[Duvie]

Ok....I went for the highest speed I can get with a 4:5 ratio versus the highest overall mhz with a 1:1 ratio....

VCORE
2.4@3.20 (4:5) 445 = 1.58v
2.4@3.24 (4:5) 450 = 1.61v
2.4@3.29 (1:1) 366 = 1.65v
2.4@3.33 (1:1) 370 = 1.70v

SANDRA: MEM
2.4@3.20 (4:5) 445 = 3418/3416
2.4@3.24 (4:5) 450 = 3438/3435 ***best***
2.4@3.29 (1:1) 366 = 2714/2710 (2,6,3,3)
2.4@3.33 (1:1) 370 = 2740/2735

3DMARK2K1SE
2.4@3.20 (4:5) 445 = 8692
2.4@3.24 (4:5) 450 = 8719
2.4@3.29 (1:1) 366 = 8698
2.4@3.33 (1:1) 370 = 8725 ***best***

BESWEET Mp3 Encoder (min/sec)
2.4@3.20 (4:5) 445 = 4min/10sec
2.4@3.24 (4:5) 450 = 4min/7sec
2.4@3.29 (1:1) 366 = 4min/3sec
2.4@3.33 (1:1) 370 = 3min/59sec ***best***

GKNOT 0.27 DIVX 5.02pro (min/sec)
2.4@3.20 (4:5) 445 = 4min/23sec
2.4@3.24 (4:5) 450 = 4min/19sec***best***
2.4@3.30 (1:1) 366 = 4min/33sec (cas 2,3,3,6)
2.4@3.30 (1:1) 366 = 4min/21sec (cas 2,2,2,5) *ran stable in this test but 3dmark would crash unless I had 2,5,3,3*
2.4@3.33 (1:1) 370 = 4min/29sec (cas 2,3,3,6)
2.4@3.33 (1:1) 370 = 4min/19sec (cas 2,2,2,5) *ran stable in this test but 3dmark would crash unless I had 2,5,3,3*

DVD2AVI (Lord of the Rings II) forced film with all tracks (min/sec)
2.4@3.24 (4:5) 450 = 3min/53sec
2.4@3.24 (1:1) 360 = 3min/55sec (2,3,3,5) **due to 2,2,2,5 not being stable
2.4@3.33 (1:1) 370 = 3min/52sec (2,2,2,5)

ReMPEG v1.52 (Lord of the Rings II VOB_5 file is reduced to 80% bitrate: same output in size and ac3)***it is being re-encoded to lower bitrate to help it fit a certain size...its benefit is getting a dual layer dvd to fit on one dvd-rw***
2.4@3.24 (4:5) 450 = 14min/18sec audio/video resynch = 1min/24sec .....file size is 217MB (orig. 262MB)(2.5,3,3,7)
2.4@3.24 (1:1) 360 = 14min/21sec audio/video resynch = 1min/48sec .....file size is 217MB (2,3,3,5)
2.4@3.33 (1:1) 370 = 14min/04sec audio/video resynch = 1min/38sec .....file size is 217MB (2,3,3,5)

Just for fun comparison
GKNOT 0.21 DIVX 3.11 (Gone in 60seconds...1hr58min...2pass, same settings)(hr/min)
1.6@2.40[i845d] (3:4) 400(cas 2.5) = 2 hr/39min
1.6@2.66[i845d] (3:4) 442(cas 2.5) = 2 hr/19min
1.6@2.73[i845d] (1:1) 342(cas 2) = 2 hr/19min
2.4@3.20[i845pe] (4:5) 445(cas 2.5) = 1 hr/58min
2.4@3.24[i845pe] (4:5) 445(cas 2.5) = 1 hr/52min



may try for 1-2 more fsb above 450ddr, but I don't think I am going to get it....lowest cas and memory timings now and had to adjust to that to get 5more mhz from 3.2ghz...

 

[Duvie]

bump for ppl to take a look at cpu versus mem bandwidth....common question that comes up....

Thugs please enter some of your gaming data when you tested this...I will change title so better help ppl find this...

 

[THUGSROOK]

id love to help out ~ but i gave all that up for granite bay

Duvie's results are extreamly consistant with what ive seen before.
in his case his cpu is a lot better then his memory.
if his cpu wasnt quite so good, memory would win more of these tests.

each system will have a different "comprimise" depending on parts.

 

[human2k]

nice results.

 

[Duvie]

PPL run some test of your own...I want to see some other encoding numbers from different programs, gaming, and any office apps...

Run chip with 400+ ddr if you can then up the cpu 80-100mhz and drop ratio to 1:1 and run again...I want to see if many see numbrs I have seen with my 1.6a and now my 2.4b/// think my numbers are far more scientific then a guy who wrote an article at www.overclocker.com...



Article to read....

let me know your thoughts about this.....I think the author shows no proof and my numbers and some of thugsrook's numebrs of the past contradict his findings...

In his case we should all run lowe 300mhz ddr as long as we can get raw cpu speed way up cause that is all that really provides performance..Nevermind gaming apps show huge love for added bandwidth...BUt you know it is written on the net so it must be gospel!!!

 

[oldfart]

I'll run a TMPGEnc and Pinnacle Studio 8 AVI->MPEG2 encode.

 

[Duvie]

Thanks I appreciate it....

 

[oldfart]

Ran a couple of tests. I would have to run them a 2nd time to double check for accuracy, but here they are:

I used a 4:22 720x720 AVI file for each test.

Pinnacle Studio 8.5
Render and convert file to DVD format and save to HD as a file.
P4 2.4B @ 3.06 | 170 FSB | DDR425 2.5-3-3-7 11:18
P4 2.4B @ 3.15 | 175 FSB | DDR350 2-2-2-5 11:05

TMPGEnc 2.57
Compress to an SVCD compliant MPEG2 480 x 480 using 2 pass VBR encode, high quality
P4 2.4B @ 3.06 | 170 FSB | DDR425 2.5-3-3-7 15:04
P4 2.4B @ 3.15 | 175 FSB | DDR350 2-2-2-5 14:37

When running the tests 1:1 ratio, the CPU speed is faster as is the FSB. The ram timings are also better. The combination of the three outweigh the gain from the increase in mem speed. This is one example. There are many factors. Each system is different. If you are at/near the wall on your CPU speed do to Vcore/temp/cpu limit, you can get some performance with the extra mem BW. I'm not done tweaking my system up yet, so I'm not sure in my case. I cranked up the Vcore to run the test @ 3.15. I'm not yet sure how much Vcore I would need for that speed or higher.

This is why DC DDR is nice. The mem ratio thing goes out the window. Bring on Canterwood/Springdale!

 

[Duvie]

I think I need to rerun it with the memory at cas 2,2,2,5...i only ran it at 2,3,3,5...I guess I will try turbo settings and rerun some of the test...

With my 1.6 I did run the test at 2,2,2,6 versus 2.5,3,3,6 and the higher speed of 100mhz just made up for the 100mhz of bandwidth...The nice thing was the lower speed only needed 1.63v while the higher speed needed 1.71v to run....

 

[oldfart]

The nice thing was the lower speed only needed 1.63v while the higher speed needed 1.71v to run....

Exactly. Again, it depends on your setup. By increasing mem speed, you can gain some extra performance without cranking up the Vcore to get that last bit of CPU speed. I like to run a low Vcore if at all possible.

 

[Duvie]

OK I ran 3.3ghz and 3.33ghz 3 times each with +/- 1 sec difference....averaged it out and 3.33ghz with cas 2,2,2,5 numbers imporved the score of the 3.33ghz but it still just tied the 90mhz slower cpu....advantage not much except I can do 3.24ghz at 1.58v and the 3.33ghz needs 1.7v.....3.3ghz was slower then 3.24ghz with faster timings and just above 3.20ghz and 445mhz ddr.....

I am going to test some other programs I can dig up to test this bandwidth thing somemore....

I may do a dvd2avi for a diff. program....I also can use info edit to mod vob files for dvd burning....I have no real games other then flight sim and I don't know what that will tell me

 

[OddTSi]

While you guys are doing bandwidth/ram related benchmarks I was wondering if someone can do one that I've been wanting to see for a while (or link me to it if it has already been done). Make no change in the system other than RAM timings (cpu, fsb, and mem frequencies all the same). I want to see how much of a difference it makes going from default RAM timings to more aggressive ones.

 

[oldfart]

advantage not much except I can do 3.24ghz at 1.58v and the 3.33ghz needs 1.7v

Depends on how you look at things. I'd run 3.24 for the lower Vcore and ~ the same performance. I like cool/quiet/stable. 1.7 is more than I like to run.

 

[THUGSROOK]

OddTSi ~ here

 

[OddTSi]

Originally posted by: THUGSROOK
OddTSi ~ here

Thanks.

 

[THUGSROOK]

Originally posted by: oldfart

advantage not much except I can do 3.24ghz at 1.58v and the 3.33ghz needs 1.7v

Depends on how you look at things. I'd run 3.24 for the lower Vcore and ~ the same performance. I like cool/quiet/stable. 1.7 is more than I like to run.

 

[Duvie]

added some new bench tests I will be finishing here shortly....

DVD2AVI test....creating a frameserver project of a DVD prior to Divx encoding....

ReMPEG v1.52...nice program that can resize VOB files by reducing bitrates based on percentage to help fit size requirements...IE like getting it to fit on 1 dvd-rw...I had a movie that was 4.7gb but the dvd-rw for a video dvd would not let be burn past 4.5gb...so I took 4.7/4.5 and came up with a percentage...I then reduce the vob files by that percentage to get it down to 4.5gb. That reduction had no noticeable visual effect. Larger reductions likely will.

The rempeg program is a good encoding program so I am interested if it likes the bandwidth or is a cpu intensive app....My hunch is the dvd2avi is but we will soon see...

 

[Duvie]

OK...finished..ran dvd2avi twice ach and will rerun the rempeg 1 more time as well..I saw no fluctuation in dvd2avi.....


The dvd2avi is a wash as 90mhz more cpu or 2.7% more clock speed amounted to 0.0% gain versus the slower cpu with 80mhz more memory speed....

The rempeg the video encoding amounted to a 1.6% increase with faster cpu but the audio/video resynch was slower so total was 3.33ghz cpu was the same time again...

Edit: I had to redo the scores for 3.33ghz as the cas 2,2,2,5 was found to be unstable in other apps....I want to compare timings that can be attainable on my board and achieve the same level of stability I enjoy now...