My conclusion so far: low latency works best for me

[bigpow]

I did a few tests and so far low latency wins over extra bandwidth (even w/ bigger RAM).
Strange because I always felt that running 2GB at reduced latency was faster than 1GB LL.

Setup:
-Venice E3 3000+ 9x278 (HTT 4x) 1.55V
-MSI Neo4 SLI

RAM test combinations: (all at max; 2.85V)
1) Corsair XMS3200C2 2x512M 2-2-2-6 1T 133/200
2) Corsair XMS3200C2 2x512M 2.5-4-4-9 2T 166/200
2) Corsair XMS200C2/VS mixed 4x512M 2-3-3-6 2T 133/200

Everest Results:
#__Read___Write___Latency
1__5192___2163___44.6
2__5972___2062___53.2
3__5072___1523___51.2

SiSoft Sandra Results:
#__BW Int____BW Float___Cache&Mem___Speed Factor
1__5367_____5379______6412_________7.2
2__5414_____5395______6278_________7.6
3__4287_____4291______5785_________8.9

AquaMark3
1__68626
2__67871
3__66984

3DMark05
1__4735
2__4705
3__4664

PCMark05
1__4147
2__4109
3__4083

Well, any comments & questions are expected.

 

[fbrdphreak]

Try 3DMark01

 

[keidanny]

You are comparing 1T to 2T, which isn't exactly a comparison. A64s are shown to take a huge hit by using 2T rather than 1T. Can you test #2 with 1T instead of 2T?

 

[Shimmishim]

yeah. you're testing method is really flawed.

you need to compare 1T vs 2T at same speed and divider or much higher HTT vs. lower HTT with 1T vs. 2T.

or 1T 2.5-4-4-9 vs. 1T 2-2-2-6 as keidanny stated.

 

[bigpow]

The reason I did my test like that was to find out the best value for me.
- My mobo doesn't let 4x512M at any speed to do 1T.

I also did all the tests with my normal configuration, with AVG & ZA running on the background.
I could have tested with the same reduced timings, but I went with all the best scenarios possible, again for me.
- Maybe we can compare test results?

What I still don't understand:
- why it doesn't let me use 1T when I have 4x512M?
- will this mobo let me use 1T if I have 2x1GB?
- why only 100/133/166/200 dividers with my setup? Is it the mobo or the CPU DDR ctrl?
- why 2GB (although produce lower scores) feels faster during day-to-day use?

Perhaps some DFI users could confirm that they can indeed use 4x512M dual-channel at 1T?
And perhaps some Opteron users could confirm that the Opteron DDR ctrl allows more dividers such as 143/150/183/etc?

EDIT:
- My crappy & overpriced Corsair XMS can't do 1T at 166/200x278 (231MHz/DDR462)

 

[aLeoN]

-They were talking about #2 man, the 2x512 at 2T...unless you typed that wrong.

-A64's don't do 1T with 4 dimms.
-No clue for the rest.

 

[forumposter32]

Yeah, synthetic benchmarks tell you everything about real-world performance...duh!

 

[michal1980]

didn't we already do this before? but with real world test.

conclusion was with amd timings make a small difference (1-7%) and thats with T being 1 or 2

 

[keidanny]

I've read lots of things around many forums. I got confused after reading it all. Some say bandwidth is the way to go, others say timings.

 

[orangat]

For AMD64 systems, latency is more important.

 

[Mucker]

I always liked this report, scroll down for the graphs: http://www.xbitlabs.com/articles/memory/display/ocz-ddr600_4.html

Says quite a bit about timings/performance....

The best way to run a machine is MAX cpu frequency and MAX memory frequency (whatever the timing as long as it doen't exceed 3-4-4)