Interesting 1t vs 2t benchies, post yours

[rise4310]

http://forums.anandtech.com/messageview.as...6&enterthread=y

 

quote:

 

--------------------------------------------------------------------------------

The Importance of Command Rate

Socket 754 Single-Channel motherboards performed best with a memory Command Rate setting of 1T in BIOS, but that generally was a stable option with only one DIMM. 2 or more DIMMs normally required a 2T Command Rate setting for most stable performance. There was a performance increase at the 1T Command Rate setting, but the real performance increase was very small.

 

Socket 939 Dual-Channel motherboards were found to exhibit a very wide performance difference between a Command Rate setting of 1T and a setting of 2T. The impact on memory bandwidth is dramatic between these 2 settings. In SiSoft Sandra 2004 standard buffered Memory Benchmarks, a 1T command rate showed a Sandra bandwidth of 6000 Mb/sec, while a 2T rate with the same 2 DIMMs in Dual-Channel mode was only 4800 Mb/sec. This is a huge difference in memory bandwidth and the Command Rate setting definitely impacts performance test results on Socket 939 motherboards. All AnandTech benchmarks were run at a Command Rate setting of 1T. This includes all benchmarks that were run in the CPU tests, as all benchmarks were rerun in the CPU tests as soon as we had verified the performance impact of Command Rate settings.

--------------------------------------------------------------------------------

 

 

i've been following this thread over at my other favorite forum and finally ran a few benches of my own. using all the same stable timings run each and any set of application benches you want twice, changing only the CPC between 1t and 2t. if you have the time, you may want to run 3 or more passes of each and use an average to be more accurate.

 

i found the results interesting as i thought the performance hit would be 5-10% or more when in reality it looks to be between and 1-5% worst case. this means to compensate for having to drop to 2t on an oc @2500 you need only gain about 25mhz as opposed to 125mhz or better.

 

keep in mind we are talking benches about real world benefits, ie application benchmarks. synthetic benchmarks such as sandra and everest will show tremendous differences. my results are posted there but i will add them here. i haven't run any fps in actual games yet.

 

winnie @ 2475, 9x, 2.5-4-3-7 @275 mhz

superpi 8m

1t 6:10

2t 6:27

 

pcmark04

1t 5143

2t 4996

 

aquamark

1t 66,903

2t 65,616

 

3dmark01

1t 24,361

2t 23,212

 

3dmark03

1t 11,710

2t 11,592

 

far cry volcano 12x10 4x/4x very high

1t 89.71

2t 87.65

 

DVDshrink encoding no compression

1t 15.40

2t 15.40

 

dvdshrink encoding compression 10:6

1t 28:40

2t 29.42

 

as a side note, before switching to 2t in BIOS, i ran the benches using a64tweaker to change to 2t and the results across the board fell just about in the middle of the BIOS 1t and 2t benches. in other words, the difference wasn't as great as using BIOS to change it which may or may not surprise you or mean anything. i'm new to that program so i'm not really sure how accurate it is.

 

at any rate, hope you find it interesting and any input is appreciated. if you do post your benches, please be sure to make the change in BIOS and not in windows.

 

edit, i know all you have sigs with specs, lol, but if you could just list the core, speed and timings near your results it may be easier to read.

[red930]

Looks like all of your results are well within the margin of error except the SuperPI which is on the outside at 4.7%.

 

For a more valid series of tests you should run each of them three times back to back then change the CPC setting in BIOS for the next series of three each.

[rise4310]

true, all the 1t have been run too many times to count, lol, but they are what i would call my average. the 2t i only ran a couple twice, but they were so close i figured i'd post initial anyway. i don't see it changing much but i will tomorrow.

 

and yeah, the pi was 5% with all the others 1% or less. i think i'll time some dvd rips/burns tomorrow too.

[Angry_Games]

2 or more DIMMs normally required a 2T Command Rate setting for most stable performance. There was a performance increase at the 1T Command Rate setting, but the real performance increase was very small.

 

what kind of rig was this dude running????

 

i have NEVER EVER had to run 2T with 2x dimms (2x512 double-ranked DDR) on any of my 754 cpu's =/

 

makes me wonder what these guys are up to, and it reminds me to never really pay much attention to some websites since their info/results are....lets say....strangely different than mine and most of our customers here....

[rise4310]

first line in the link i posted:

 

"In just about every mem review Wesley makes qualitative statements like this:"

 

and then proceeded with the cut and paste i did. sorry for the confusion.

[Angry_Games]

again, you should really be careful about what you believe from some of these sites...some have been caught in blatant lies, some have been caught giving mfg's preferential treatment, some have just made things up to increase their reader base or to make it look like they were really smart...

 

im not saying anything about anyone in particular...but you really....dont take things as truth just because someone says it, no matter where you heard it said.

 

the fact that the statement like '2 or more DIMMs normally required a 2T Command Rate setting for most stable performance. There was a performance increase at the 1T Command Rate setting, but the real performance increase was very small.' was made and no one called him on it...makes me nervous to really believe anything that this person would write as it is so blatantly in error, and unchecked, unverified.

 

I've pushed at LEAST 15GB of memory through the DFI NF3 and the VIA K8T800/K8M800 and never once have I ever had to resort to 2T regardless of settings.

 

only when using 4x512 did i have to drop to 2T, and that is only the NF4 since the NF3 cannot use 4x anything.

[rise4310]

i think you're missing my point. this is an attempt at calling him on it. not so much how many sticks he runs how, but what difference does 1t vs 2t make. i should have just cut it at:

 

"Socket 939 Dual-Channel motherboards were found to exhibit a very wide performance difference between a Command Rate setting of 1T and a setting of 2T. The impact on memory bandwidth is dramatic between these 2 settings. ...."

 

and i agree with you, alot of info is thrown out there and accepted as gospel. the thread at Anandtech was started by a very knowledgeable member, not reviewer, who was questioning why 1t vs 2t is shown as such a huge performance hit when there is no real life experience showing that to be true.

[mucker]

That's a great post Rise, will be interesting to see what kind of picture is painted from real world test results.

 

m

[Sharp]

Hello,

 

Using SisoftSandra 2005 lite (Could the Anandtech person use version 2005 too?)

 

==2T==

4904mb/s

4897mb/s

 

==1T==

5375mb/s

5310mb/s

 

==Diff==

471mb/s = 9%

413mb/s = 8%

 

Could this be the effects of integrating the memory controller to the CPU?

It has become more sensitive to small changes.

 

I wonder, is the 754 percentage in SisoftSandra 2005 only 4%?

[red930]

Sandra and other synthetic benchmark programs show a greater difference between 1T and 2T than real applications.

 

This is the reason why the earlier test were made using application benchmarks. The spread is much closer using applications versus synthetics.

[rise4310]

Sandra and other synthetic benchmark programs show a greater difference between 1T and 2T than real applications.

 

This is the reason why the earlier test were made using application benchmarks. The spread is much closer using applications versus synthetics.

 

exactly exroadie, i'll edit my first post to relfect what you said better than i managed to. i'll edit it when i add the encoding/compressing bench i'm doing.

[lmmaudio]

In the past I've done some of those tests on my actual rig, just to "see" what CPC was all about as far as bandwidth/speed was concerned.

 

Never got to register the tests on my spreadsheet so I could post here too, since the loss of performance was so evident, at least on synthetic applications which were used. The only real application used was a HL2 benchmark, and the results were pretty similar 1T or 2T...

 

With Marks&Sandras the performance difference was also greater than on Pimod, as I well recall.