• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

Memory Timings....strange results achieved by upping the row active delay

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.
T

-=TriX-R4-KidS=

Member
​
Joined
Mar 25, 2003
Location
Indiana University, USA
I mentioned this in the "lets beat each other out thread" but I might get an answer here:

ive been tweaking my ram timings and have found some type of anomoly or something. i thought the lower timings the better....but

when i increase row active delayi get a boost in sandra bandwidth, pcmark memory test (the only apps ive tested this with)...

::::::::::::::::::::::::::::::::::::::::::

Timings* --------------- ram bandwidth--- PCMark2002
-------------------------------------------------------------------
4-3-3-CL3 -------------- 3282/3054 ------- 5598
6-3-3-CL3 -------------- 3283/3054 ------- didnt test w/ pcmark
9-3-3-CL3 -------------- 3316/3076 ------- 5666
10-3-3-CL3 ------------- 3316/3075 ------- 5657
11-3-3-CL3 ------------- 3322/3085 ------- 5669
13-3-3C-L3 ------------- 3316/3066 ------- 5646

[settings of 14/15 get worse than 13-3-3-CL3

*[Row Active Delay]-[RAS to CAS Delay]-[Row Precharge Delay]-[CAS Latency Time]

::::::::::::::::::::::::::::::::::::::::::::::::

overclock_DDR434.jpg

11-3-3-CL3 -- 2.9v - DC

strange....what causes this?

-TriX
 
Cool! Mine have the same behavior. I am glad to see I am not the only one. I have 2x512 TwinMos Winbond CH-5 running in dual channel on Abit NF7-S v2.0. At 220fsb, 8-3-2-2.0 scores higher than 7-3-2-2.0 which scores higher than 4-3-2-2.0. These were tested in Sandra. I get a nice boost in 3dMark2001SE as well. I haven't tried increasing the timing to find the optimal performance yet. I'll try it and let you know. Hopefully, someone can explain this.
 
good to hear im not the only one :/. for me it seems to be 11....any higher and scores start dropping and lower and they decrease as well.

quite strange. maybe it has something to do with dual channel...???
 
Holy cow! You are right. My optimal performance setting is at 11-3-2-2.0. Here were my test results in Sandra:

8-3-2-2.0 -------- 3351/3132
9-3-2-2.0 -------- 3353/3137
10-3-2.2.0 ------- 3355/3140
11-3-2-2.0 ------- 3357/3142
12-3-2-2.0 ------- 3356/3136

I also gained higher scores in 3dMark2001SE. Granted, the gains were very small, but I'll take what I can get. I can get quite a bit higher scores with 11-2-2-2.0 but it wasn't stable.

I think you maybe onto something. Perhaps it does have something to do with dual channel. I assume you have an nForce2 board like me.

Can anyone here with GraniteBay/Canterwood/Springdale boards confirm this (in dual channel)? My other PC is GraniteBay but I am too lazy to swap memory around.
 
chung_chang said:
Holy cow! You are right. My optimal performance setting is at 11-3-2-2.0. Here were my test results in Sandra:

8-3-2-2.0 -------- 3351/3132
9-3-2-2.0 -------- 3353/3137
10-3-2.2.0 ------- 3355/3140
11-3-2-2.0 ------- 3357/3142
12-3-2-2.0 ------- 3356/3136

I also gained higher scores in 3dMark2001SE. Granted, the gains were very small, but I'll take what I can get. I can get quite a bit higher scores with 11-2-2-2.0 but it wasn't stable.

I think you maybe onto something. Perhaps it does have something to do with dual channel. I assume you have an nForce2 board like me.

Can anyone here with GraniteBay/Canterwood/Springdale boards confirm this (in dual channel)? My other PC is GraniteBay but I am too lazy to swap memory around.
Click to expand...

yeah ive got an abit nf7-s rev2

i lowered my ram speed back to its rock solid stable oc... DDR 415 (207.61x2) and got some new results...i just tested my tightest attainable timings at this DDR266 to DDR415 oc w/ a 6 row active delay against my previous 11 settings (still kept the 3-3-CL3)

:::::::::::::::::::::::::::::::::::::::::::::::::::::

Timings -------- Sandra RAM Bandwidth --- PCMark02 memory
----------------------------------------------------------------------------
11-3-3-CL3 ---- 3147/2912 ------------------- 5383
6-3-3-CL3 ------ 3162/2939 ------------------- 5381

*10 results were recorded and then averaged

::::::::::::::::::::::::::::::::::::::::::::::::::::::::::::

so w/ a lower RAD...i get a better bandwidth score (as it should be..and about the same pcmark ram score. .....maybe the higher speed your RAM is at...you should try the higher RAD setting....i dont know :confused:

i also tested the CL, RAS to CAS delay and row precharge delay settings and w/ these the lower....the better performance.

If i can get some free time in the next few days ill switch to single channel and see what kind of results I get.

-TriX
 
yeah and according to your results....the difference seems to be smaller when compared to running 2-2-2 instead of 3-3-3. mine's generic pc2100 (SPD: 6-3-3-CL3) so im already happy w/ my current results ive been able to achieve w/ these little beauts :D

-TriX
 
Regardless, the gains seem minimal. So I'd bump the FSB/Mem frequency to highest stable, with the most relaxed timings. And then tighten the timings, until it becomes unstable.
 
JohnnyTheRed said:
Regardless, the gains seem minimal. So I'd bump the FSB/Mem frequency to highest stable, with the most relaxed timings. And then tighten the timings, until it becomes unstable.
Click to expand...

thats what i planned on dong....but im just wondering why it does this..
 
guys this has been a reported/known annomoly.ive seen it once some where else i cant remember.
never been answered ive seen tho.but its very intersting and id bet only an engineer will figure out why.

im bored so ill run a few tests also.
 
the testing is back.

all run on sandra with a nice bonus in the end,which ill save for last.
220fsb
2.0-2-2-3 3383/3176
2.0-2-2-5 3396/3191
2.0-2-2-8 3403/3194
2.0-2-2-10 3410/3201
2.0-2-2-11 3414/3209
2.0-2-2-12 3409/3204
2.0-2-2-13 3402/3170
2.0-2-2-15 3398/3169

-=TriX-R4-KidS= statements are true and very helpfull.
i too my settings of 220fsb 2.0-2-2-11 and ran 3dmark2k1
http://service.futuremark.com/compare?2k1=6545886
this is the same exact settings and i beat my old score by 200 points.
some may say thats not much improvement but ill add i gained also 10 or so mhz on my gpu i havent even tried out yet by offing my shim and as3 on the gpu.so it will go even higher.
i want 19000 in 3dmark2k1 and 6000 in 3dmark2k3
this puts me one step closer.

ty trix!
 
How do you get SDRAM Active Precharge Delay value greater than 8T ? Are you talking AMD motherboard system?

Mine is Intel P4 2.4B with Asus P4PE, and SDRAM Active Precharge Delay value are: 5T, 6T, 7T, 8T
 
asw7576 said:
How do you get SDRAM Active Precharge Delay value greater than 8T ? Are you talking AMD motherboard system?

Mine is Intel P4 2.4B with Asus P4PE, and SDRAM Active Precharge Delay value are: 5T, 6T, 7T, 8T
Click to expand...

yeah nforce2 boards sorry :(

you could test your 8t vs 5T/6T and see if you get any improvements.

-TriX
 
JohnnyTheRed said:
Regardless, the gains seem minimal. So I'd bump the FSB/Mem frequency to highest stable, with the most relaxed timings. And then tighten the timings, until it becomes unstable.
Click to expand...


I agree, WHO CARES? Its not like any of us can tell a difference in 20 points difference in a benchmark in day to day operation? Does the PC boot any faster at these new optimized settings that are only 20 points difference one way or the other in a memory benchmark? Or does that Excel spreadsheet now open with blazing quickness with these new optimized settings? :rolleyes:

Some people are a little TOO obsessed with benchmarks. :eek:
 
Blkout said:



I agree, WHO CARES? Its not like any of us can tell a difference in 20 points difference in a benchmark in day to day operation? Does the PC boot any faster at these new optimized settings that are only 20 points difference one way or the other in a memory benchmark? Or does that Excel spreadsheet now open with blazing quickness with these new optimized settings? :rolleyes:

Some people are a little TOO obsessed with benchmarks. :eek:
Click to expand...
some people dont have screaming systems and arent obsessed with every last ounce of squeezed out speed.

did i jion martha stewerts quilting forum and miss it?

i said i gained 200points.and for one setting change that is a big differance.it equals 10-15 extra mhz on my gpu in comparison.

ever play a game at over 9800pro speeds on a $230 9700np with all eye candy on high and high res?
im not obssesed with benchmarks.the benchmarks are a part of my whole oc obsession. and i wouldnt advise knocking ocing around here.:D
 
deathstar13 said:
some people dont have screaming systems and arent obsessed with every last ounce of squeezed out speed.

did i jion martha stewerts quilting forum and miss it?

i said i gained 200points.and for one setting change that is a big differance.it equals 10-15 extra mhz on my gpu in comparison.

ever play a game at over 9800pro speeds on a $230 9700np with all eye candy on high and high res?
im not obssesed with benchmarks.the benchmarks are a part of my whole oc obsession. and i wouldnt advise knocking ocing around here.:D
Click to expand...

:clap: :clap: :clap:

well said
 
aside from benchmarks do you find the higher RAD setting to stabilize the machine? I'm having stability issues with Kingston HyperX 512/3200.. well, I think it may be the RAM but not sure yet..
 
Hey TriX.

{HPG}Misanthrope here ; )

Interesting stuff. If I have any time to play around with my system I'll see what happens on a single channel P4 board. From what I've seen in the past though my timing changes don't really affect my benchmarks significantly. Only increasing my memory bus really seems to have much of an effect.
 
You must log in or register to reply here.
Back