DDR600 Guide to Stability NF4 TCCD

Intro: With everyone jumping on board the A64/NF4 bandwagon lately i have been getting a lot of PM's in reagards to the memory settings i run for DDR600. Well what i intend to do with this post is show those running TCCD memory what settings to build from to obtain DDR600 speeds. This is by no means a guide for everyone or every CPU, these are just the settings i use for that elusive 300HTT 1:1.

One thing to note right from the begining here is that just because you cannot reach 300HTT does not mean you have a bad set of memory. One common problem with A64/NF4 is the CPU's on-die memory controller crapping out before 300HTT. There is no way to predict how your controller will act at high HTT, so bare this in mind as you attempt to hit high speeds with your own CPU/memory combo.

Hardware: What i will be using as my test bed is as follows.

DFI NF4 SLI-D BIOS: 4/14-3
AMD 3500+ Venice .90nm
2x512 PC-4400 G.Skill LE TCCD 0437/0440
BBATI x800 XL
OCZ Powerstream 600w
1x74GB Western Digital Raptor

I will not use any added cooling, and the CPU is cooled by a Zalman CNPS7000B-Cu.

Method: There are many ways to test the stability of your memory. What i find to be the best, and fastest way is to use Memtest86. This can be downloaded and put onto a bootable floppy as well as a CD. With the latest BIOS from DFI this option is built right in. I personally use test 5 for my stability testing. Typically speaking if there is going to be an error in test 5, it will occur within the first few passes. This is nice because it makes fine tuning very easy, and you can always go back later down the road and run a 24hour test if you so desire. I chose to not add any added cooling for my modules, but you should be warned that this test will produce a large ammount of heat and may require active cooling. With the addition of active cooling to your memory you also can add stability at the same time. Just because you're memtest stable does not mean you're Windows stable. Be sure that once you've found the max HTT the memory can handle, go and test it in Windows. I always reccommend you test with a lower multi to take the CPU out of the equation.

Testing: To run test 5, do the following:

On the first screen, hit "C" to enter the configuration.



Once you enter the configuration, you will want to hit "1" to enter the Test Selection menu.



You then want to enter "3" to actually enter in the test you want to run.



At this point hit in "5" or whatever test you want to run and hit enter. This will take you back to the first menu and you just hit "0" to run the tests.

Settings: These are what i know to be 100% stable at 300HTT 1:1. They are not the best settings available, but ones i know my own memory is capable of. I will add some suggestions of what to try if you run into problems, but remember not all modules are created the same. I have found these to offer me the best performance and stability. I could run these settings 24/7 and even tighten a few up for benchmarks.

First here are the basics. My CPU is good through 2.9GHz on air, but we will only be testing it at 2700MHz for these tests. You can always tinker with the voltages to find what's best for you, but this offers my overall stability.







If those settings are not stable, you can change the following:

- Max. Async. to 9ns
- Lower the refresh interval (Tref)
- Trrd to 3
- Twr to 3
- Trwt to 3
- Idle Cycle at AUTO
- Enable DQS and set to 255
- Data Drive Strength to 2
- Drive Strength to 8
- Ras to Cas to 4 (largest performance hit)

Note:

-It is important you have LDT at 3x for HTT over 250.
-Playing around with tref can add stability and also can boost performance.
-I suggest you leave Dyn. Counter and Bank int. enabled with TCCD

Results: Ok, here is what we find with the above settings.

At 2.6v we get errors in only the 3rd pass.



At 2.8v we get even more errors in the 3rd pass again. As you can see, more voltage can actually hurt the performance of TCCD. My Results are similar on both my 2 sets with high voltage. I gain no stability over 2.7v



And now the sweet spot, 2.7v seems to be the perfect ammount of voltage to take my TCCD sets to 300HTT 1:1 I always suggest you start with lower voltage and work your way up, not to exceed 3.0v This voltage may vary from set to set, but this is where my sets respond the best.



So here is 25 passes of Memtest86 test 5 without any errors. For the sake of this quide we will call it stable. I would advise you to run at least a few hours worth to test for full stability.

And here we are in Windows:







Again i would just like to point out this guide is only intended as a starting point for those on the NF4. Your results will vary, and these settings may be very far off from what works for you. If you have anything you would like to add i urge you to PM me or post in this thread and i can add it to the post. If you have any questions feel free to ask them and i will do my best to help you.

GREAT Guide, I vote sticky! This will help many many people out for sure..

Sticky indeed!

what does superpi 1M and 32M look like ?

eva2000 said:

what does superpi 1M and 32M look like ?

Click to expand...

Didn't test them at those settings actually. My intent wasn't to provide my best settings/scores, just settings i knew were 300HTT capable with air/normal daily use. I can do that though. My best pi is 25sec if you want to do a comparison for best score? If you have some better settings that are 100% stable, please let me know.

Thanks man!

If this is going to be a sticky I suggest a link on were to download it. I know you can use google but just for quick reference I think it would be good to have a link.

good guide sucka, i appreciate anyone that can snap good clear pics..... for everyone screaming sticky, Onepagebook did one of these last month.

Free bump, this needs some more views with all of the TCCD questions asked.

very nice guide. so many people struggle with the elusive ddr600.

Thanks guys. There are several guides out there, but with the ammount of questions i got after my Venice thread i thought it would be nice to share what i did for those numbers. I even did it on air for you guys

Bump - why is this not a sticky yet?? im gonna print this out to try and get my gskill LE's to do 275fsb...as atm i can only get them to do 250 and its holding my san diego back...

I got it all working except I had to make the ras to cas delay to 4 otherwise I get like 2,000 errors per run. Can I do one of those burn in techniques and just have it constantly run at 2.5-3-3-6 anyways until they can handle it?

Sucka said:

I even did it on air for you guys

Click to expand...

lol @ this.

great guide though sucka its going to steer alot of people with this goal in the right direction.

azianese said:

I got it all working except I had to make the ras to cas delay to 4 otherwise I get like 2,000 errors per run. Can I do one of those burn in techniques and just have it constantly run at 2.5-3-3-6 anyways until they can handle it?

Click to expand...

Some people will claim burn in works on TCCD, i haven't tested that theory myself. RAS to CAS at 4 gives you about a 200mb bandwidth hit. See if you can't loosen some other things up to gain that stability and maintain high bandwidth. Tref and drivestrenth and async can really make or break it.

Nice job Sucka.

Working 300 x 9 @ 2.5-4-3-6 in tccd now.

SOY_SET said:

Nice job Sucka.

Working 300 x 9 @ 2.5-4-3-6 in tccd now.

Click to expand...

Let us know how it goes

I'm pretty sure I narrowed down the instability directly on the ras to cas delay. It makes the difference between 0 errors and like 2,000-8,000 errors per run and the number appears to just go up with every passing run so I am guessing I'll have to keep it that way. Would a 3200 be able to have all the same settings except have the multiplier at 10 to get to 3 ghz on air?