TCCD DDR600 2.5-3-3-7 twenty four seven: The ultimate settings

[kakaroto]

Kakaroto has done some more research for you because according to the poll I posted a short time ago in the TCCD DDR600+ guide, it seems DDR600 2.5-3-3-7 is still interesting among the people active in XS and off course the whole world of overclocking. Now 2GB is becoming mainstream for most gamers, TCCD still gives nice overclock adventures. In this guide I will discuss some of the optimization settings you can find in the BIOS. The guide will not be very huge like the previous one, but will only be focused on the BIOS settings to get your TCCD stable at DDR600 2.5-3-3-7. And stable is at least a SPI32M run and some hours of Prime.
Some of the important terms of DQS strobe timing, drive/data strength and DRAM refresh rate (tREF) will be discussed.

As you al know DDR600 2.5-3-3-7 is not an easy task. TCCD does not like tRP 3 at that high speed. DDR600 2.5-4-4-x and 2.5-4-3-x is more reasonable. Even though the performance difference between tRP 4 and 3 is very small, tRP at 3 looks much better. You never can tell if your memory can run DDR600 2.5-3-3-7, so you will have to test step by step and see how your memory reacts when adding some more juice.
What I already can tell you is that what you must have is a TCCD kit which have potential to reach the ultimate DDR600 2.5-3-3-7. I had enough situations where DDR600 2.5-3-3-7 was an impossible task to do, so this guide does not guarantee you a 100% success. If your memory already craps out at 280MHz no matter what voltage and settings, is it not worth to try.

For this guide I used a memory kit of G.Skill PC4400 1GB LE (275MHz 2.5-3-3-7) with TCCD 507 chips. This is a kit that will do at least DDR600 2.5-3-3-7, but needs to be tuned. The week code of the TCCD is quite important. The old 431/437/440 chips are better overclockers then the new 507/517/525. Sad to say, but there is one positive news and that is they love higher voltages more then the old revisions. I remember my old 431 only worked properly at 2.7v!

The system combination you use is very important too! A crappy mainboard won’t bring you to the satisfied results. So make sure you use a descent overclock mainboard. I used a:

- DFI NF4 Ultra-D (BIOS 623)
- Crappy Asus 450Watts PSU (24pins) (~3,21v, 4.91v and 11,64v under load)
- AMD Athlon 64 3000+ Venice 0517 with TTBT cooler
- Samsung 40GB 5400rpm

I will not use Memtest86++ because it sucks big time when tuning for a 24/7 setup, windows is what matters so programs as SuperPI32M for initial test are going to be used and for long term stability SP2004 (RAM Stress).

STEP 1

The memory is plugged in the orange banks for the most optimal results.

First an initial SuperPI32M @ 275MHz 2.5-3-3-7 2.62v run with optimized settings:

Tcl: 2.5
Trc: 7
Trfc: 14
Trcd: 3
Trrd: 2
Tras: 7
Trp: 3
Twr: 2
Twtr: 2
Trtw: 3
Tref: 133MHz 1.95us (4196)
Twcl: 1

DQS Skew mode: off
DQS Skew: 0

R/W Bypass: 16x
Bypass Max: 7x
Max. Async. Lat.: 8ns
Read Preamble: 5ns
Idle Cycle Limit: 0
Drive strength: 7
Data strength: 2

Vddr: 2.62v



I passed a full SPI32M run, duh its PC4400!

STEP 2

Let's increase the MHz a little bit to 290MHz with still the same settings:



This gives me an instant error when trying SuperPI32M...

STEP 3

What I do now is increasing the voltage to 2.72v (BIOS = 2.7v +0.03v). If you want a long happy TCCD life then 2.72~2.83v is the max. So always try your memory at these voltages and see what happens.

Tcl: 2.5
Trc: 7
Trfc: 14
Trcd: 3
Trrd: 2
Tras: 7
Trp: 3
Twr: 2
Twtr: 2
Trtw: 3
Tref: 133MHz 1.95us (4196)
Twcl: 1

DQS Skew mode: off
DQS Skew: 0

R/W Bypass: 16x
Bypass Max: 7x
Max. Async. Lat.: 8ns
Read Preamble: 5ns
Idle Cycle Limit: 0
Drive strength: 7
Data strength: 2

Vddr: 2.72v



As you can see, SuperPI32M is passed after upping the Vddr to 2.72v.

If your memory is not stable try this:

+ DRAM Refresh rate to 200MHz 15.6us (3120)
+ Max. Asyn. Lat to 9ns
+ Idle Cycle Limit to 128~256
+ Drive strength 8
+ Data strength 2

STEP 4

If Step 3 is successfully passed, you can try setting your HTT to 300MHz 2.5-3-3-7. In the screenshot below I set the memory speed to 300MHz with ClockGen. I leaved the other settings the same. Unfortunately it failed.



There are two things you can do now; tweak the settings or increase the Vddr.


+ The optimization settings in STEP 3


Let's try the second option first

STEP 5

I increased the voltage to 2.83V in BIOS that is ~2.8v in windows. I leaved every other setting the same.

Tcl: 2.5
Trc: 7
Trfc: 14
Trcd: 3
Trrd: 2
Tras: 7
Trp: 3
Twr: 2
Twtr: 2
Trtw: 3
Tref: 133MHz 1.95us (4196)
Twcl: 1

DQS Skew mode: off
DQS Skew: 0

R/W Bypass: 16x
Bypass Max: 7x
Max. Async. Lat.: 8ns
Read Preamble: 5ns
Idle Cycle Limit: 0
Drive strength: 7
Data strength: 2

Vddr: 2.80v



Unfortunately it fails again, bloody hell I give up! As you can see it's a little more stable but not enough. If your memory is stable after increasing to 2.8v, you can leave it at 2.8v or decrease it a little bit so you can tweak the settings.

But there is more to do!

STEP 6 - If STEP 5 is not stable

I tried some optimization with the settings, I took some looser timings. Still I leaved the DQS strobe settings and drive/data strength untouched. I also increased the voltage with 0.03v.

Tcl: 2.5
Trc: 7
Trfc: 14
Trcd: 3
Trrd: 2
Tras: 7
Trp: 3
Twr: 2
Twtr: 2
Trtw: 3
Tref: 200MHz 15.6us (3120)
Twcl: 1

DQS Skew mode: off
DQS Skew: 0

R/W Bypass: 16x
Bypass Max: 7x
Max. Async. Lat.: 9ns
Read Preamble: 5ns
Idle Cycle Limit: 256
Drive strength: 8
Data strength: 3

Vddr: 2.83v



Now SPI32M passed!! Switching banks really helped here. So much difference as you can see.


+ Switch the strongest memory module to the bank close to the memory controller


What I find out is that if you increased the drive strength and voltage the max. async. Lat. cannot be so high. If 9ns does not work so well, try 8ns.

Setting the Idle Cycle Limit to 256 always helps to get more stability. If you have memory is more unstable after setting to 256, leave it at 0 ~ 32.

If your memory is still unstable try this:


+ Set DQS strobe timing to enable with value=128~255 in the BIOS. This will give you more stability but very little. Sometimes this only increases instability on some systems.
+ Try 2.86v and lower the Drive strength to 7 and data strength to 2.
+ Lower the Max. Async. Lat. ~8ns
+ If Idle Cycle limit at 256 does not work properly, set back to 16


STEP 7 - Final testing

Now to prove these ultimate settings work a nice SP2004 Large FTT one hour shot. Some important things about stressing with SP2004; your memory will get hot and your CPU will be 100%. I can recommend a fan or something else that generates some airflow. My SP2004 Large FTT crapped out after some time without a fan. In my previous test I didn't used a fan.
Now with an 80mm fan aimed to the memory:



With a better PSU I could achieve some more, but this PSU from Asus is really the bottom of all. I hoped these settings really helped on other people. I appreciate all comments, tips, results. Enjoy!

Next guides will be DDR2 Conroe platform and AM2!