I tested Single vs. Dual channel and 6 3 3 3 vs. 6 3 3 2, please post opinions

[c627627]

I have three 256MB PNY PC2700 (Bravo chips) RAM sticks

I used 3DMark05 to tested them like so to find that


1.)
1 stick in slots 1&3 (for dual channel, with slot 2 empty)
vs.
1 stick in slots 1&2 (supposedly single channel even though Cpu-Z still reported dual(?), with slot 3 empty)

shows less than 3% difference in 3DMark05 score.


2.) All three slots filled for 768 MB of RAM instead of 512MB of RAM show a 9% score decrease (!)

3.) I tested two sticks in slots 1&3 (dual channel) at

6 3 3 2
6 3 3 2.5
6 3 3 3

only to find less than 1% difference in 3DMark05 scores.

I used 3DMark because I though I would see how these things affect playing games.


What do you think about the differences as mentioned above in terms of how they would affect things if I used different testing methods?

What is your answer when asked: should I go for extra xx FSB or tighten the CAS Latency from
6 3 3 3
to
6 3 3 2.5 or 6 3 3 2

_________________________
AMD Athlon XP 2400+ mobile 35 watt AXMD 2400 FJQ4C
IQYHA 0402 SPMW
[192] FSB x 12.5 = 2400 MHz @ 1.85 Vcore with memory frequency at 100% [192 FSB] @ 2.90 Vdimm
Epox 8RDA+ v1.1
Thermalright SLK-947U with variable speed 80x25mm Thermaltake Smart Fan 2
AIW 9600XT 128MB; Antec SX-835II case ; 380W Antec TruePower TP380
Four 80x25mm Case Fans + 1 Variable speed 80x25mm Fan YS-Tech FD1281259B

 

[pershoot]

if your fsb is 192 max that u can achieve, tighten your timings as tight as u can make them, and still keep stable. if u can push higher then that, stay on cas 2.5.

 

[Joe Camel]

here are some stats i got when testing my VX in an Abit AV8 with a 3500+ "winnie" (copy/paste from other thread)
note that this is ALSO 512 MB vs 1024 MB...

got 1 stick ("good")to:
255x10 (cpu:1.6V) 1:1 222,10,12,14 @3.4V
the other stick ("bad")to:
250x10 (cpu:1.6V) 1:1 222,10,12,14 @3.4V

Benches with ONLY "bad" stick @250:

sandra MEM.......Sandra CPU........01...........03.........Aqua......S.Pie (1M)
3854/3854.........11,531............26,698.....14,179.. ...76,520.....35 sec

BOTH sticks (DUAL) @250:
sandra MEM.......Sandra CPU........01...........03.........Aqua......S.Pie (1M)
7002/6917.........11,532............27,361.....14,205.. ...78,081.....33 sec

 

[fabulouscoops]

3DMark05 is mostly a video card benchmark. A more appropriate bench might be PCMark04. I have not seen a big difference in CAS latency 2, 2.5 or 3 either so I'd say up your fsb as much as possible while at 3, then drop the latency if it will take it. If not, use the higher fsb with the higher latency. It is a balancing act to get the tweaks just right.

 

[c627627]

Hm. What is the reason we are so much after 'tight' timings in the first place if the improvements do not seem to be noticebale?

I have not seen a big difference in CAS latency 2, 2.5 or 3 either so I'd say up your fsb as much as possible while at 3, then drop the latency if it will take it. If not, use the higher fsb with the higher latency.

 

[Gautam]

3DMark05 is a GPU-dependant benchmark more than anything else; all it really tests is your video card. You could try 3dmak01 instead, or Hexus PIFast would be a good idea too. Those should probably show about a 4% gain when running dual channel. But frankly, the improvements between dual channel, tight timings, etc, is not going to be noticeable these days, because everything is just too fast as it is. People want tight timings because, well, it's just better, and as enthusiasts plenty of us will take any gains we can get.

 

[c627627]

...But frankly, the improvements between dual channel, tight timings, etc, is not going to be noticeable these days, because everything is just too fast as it is...

Thank you Gautam. Just one final question then: when asked if it's better to squeeze the highest FSB you can with auto timings, even if they are 8 4 4 3 and your sticks can do 6 2 2 2 with slightly lower FSB, go for the highest FSB every time?