Official DFI NF4Ultra-D/SLi owners thread.

OK Rock Solid Stable at 3.079 Mhz (Phase change). Primed for over 22 hours.


SD3700at3079mhz.jpg


However I was getting slow performance, ie Superpi taking around 1 min for the 1m test.

So back to the drawing board.

I had my GSkill RAM at a Tref of 7.8ns (0016) as that was the only Tref that I could get stable at 2.7v

This is causing the slow performance.

So Tref is quite important for performance.

I am tweaking about to try & get my RAM stable at 3.9ns (Tref 3072). As Superpi 1M runs fine at that Tref but Prime errors after just under 7 hours at my current settings. Will mess with some of the RAM settings & raise the RAM voltage & see what happens.

I'm hoping to at least get 3.2 ghz out of my SD3700 eventually. Just a bitch to get the right RAM settings with this mobo.

This is using the 623.3 BIOS

I'll post back my results.
 
YooEntSinMeROYT said:
OK Rock Solid Stable at 3.079 Mhz (Phase change). Primed for over 22 hours.


SD3700at3079mhz.jpg


However I was getting slow performance, ie Superpi taking around 1 min for the 1m test.

So back to the drawing board.

I had my GSkill RAM at a Tref of 7.8ns (0016) as that was the only Tref that I could get stable at 2.7v

This is causing the slow performance.

So Tref is quite important for performance.

I am tweaking about to try & get my RAM stable at 3.9ns (Tref 3072). As Superpi 1M runs fine at that Tref but Prime errors after just under 7 hours at my current settings. Will mess with some of the RAM settings & raise the RAM voltage & see what happens.

I'm hoping to at least get 3.2 ghz out of my SD3700 eventually. Just a bitch to get the right RAM settings with this mobo.

This is using the 623.3 BIOS

I'll post back my results.
Click to expand...

The T Ref is only really slow on the very low numbers 0016 0032 etc, I messed for ages trying different values I did a 32m Super Pi at every T Ref value the only low performance were those few low numbers. The same goes with bandwidth for them it is also very low, the remainder there is very little difference between them.

I
Here's something for you to try, run either Everest or Sandra for bandwidth on one T Ref value ( I did it on everyone ) run each test 10 times and notice what happens on the memory read after the 7th test ie 8th 9th 10th, do a refresh before starting each test (press F5)
Then let me know your results, write them down and look at the pattern if you feel like it do it for a few T Ref values you will see a pattern I don't want to influence your opinion that's why I'm not saying what will happen.
 
kimandsally said:
The T Ref is only really slow on the very low numbers 0016 0032 etc, I messed for ages trying different values I did a 32m Super Pi at every T Ref value the only low performance were those few low numbers. The same goes with bandwidth for them it is also very low, the remainder there is very little difference between them.

I
Here's something for you to try, run either Everest or Sandra for bandwidth on one T Ref value ( I did it on everyone ) run each test 10 times and notice what happens on the memory read after the 7th test ie 8th 9th 10th, do a refresh before starting each test (press F5)
Then let me know your results, write them down and look at the pattern if you feel like it do it for a few T Ref values you will see a pattern I don't want to influence your opinion that's why I'm not saying what will happen.
Click to expand...

I may try that. I found that 0016 was the only Tref I could get Memtest+86 Test 5 to pass without errors at 280mhz. All other Trefs seemed to give errors.
 
YooEntSinMeROYT said:
I may try that. I found that 0016 was the only Tref I could get Memtest+86 Test 5 to pass without errors at 280mhz. All other Trefs seemed to give errors.
Click to expand...

I'm far from an expert on ram so I don't know what could be the cause of that, but my initial thought would be that it could be pushing the ram too far that could be why it works on such a low value poor performing T Ref value, it could be giving the ram the breathing space it needs to function, the problem is that there are so many adjustments that can be made and probably interact with each other it makes it very difficult.

If I were you I would lower the ram using a divider then play with one setting at a time and see what you can improve, I must say though before you rely too much on bandwidth settings do the above test, I spent a week doing it and covered every scenario and my results were ? I'll not tell you as if you do it and the same happens to you as me then we can do some head scratching, as I said I don't want to give you a hint but it should be very obvious whats happening after about 15 mins or so.
 
YooEntSeenMeROYT - have you seen the various posts on here, XS and DFI Street regarding Tref?

I'm surprised 0016 was the most stable.

It is my belief that the commonly referred to table of values is incorrect and has been compiled from what various Windows programs interpret the Tref to be.

In actuality, what many people believe is true (including me) is the following:

e.g. @ 200 Mhz, one cycle is 5ns.

Ideally, the memory wants refreshing every 7.8us (generally speaking).

To get this 7.8us, you need to use the correct Tref value, which is 1560 cycles (1560 x 5 = 7800 ns = 7.8us).

Of course, there will always be exceptions...

If too low a Tref is used, the memory will be refreshed more often, leading to increased heat and stability problems, giving errors.

If too high a Tref is used, the memory won't be refreshed enough and risks losing data, again giving errors.

Another thing to try is swapping the sticks round. As you might imagine, one stick is probably slightly better than the other and it is known that the slot closer to the CPU will help the poorer stick keep up with the stronger one. (Trace lengths and signalling blah blah).

HTH?
 
monkeygabriel said:
try 4708 for TREF.
Click to expand...

Any reason?

My suggestion would be 2064 or 2336, based on the fact the memory is @ 280 Mhz, giving a cycle time of 3.57ns. In order to give the memory the required 7.8us refresh time, 2184 cycles are needed. The BIOS doesn't offer this, so the two values I have given are the closest and should, in theory, give the most stable / least errors.
 
Last edited:
Jimbo Mahoney said:
Any reason?

My suggestion would be 2064 or 2336, based on the fact the memory is @ 280 Mhz, giving a cycle time of 3.57ns. In order to give the memory the required 7.8us refresh time, 2184 cycles are needed. The BIOS doesn't offer this, so the two values I have given are the closest and should, in theory, give the most stable / least errors.
Click to expand...


i was meaning if trying to find max speed.with my ocz pc4800 that was the most stable tref for me @ 310MHz. i.e. i could only get to 310MHz with it set to 4708.
 
Since getting my ultra-D board, I keep having to wait like 30 seconds before I can switch on my computer. Is this normal?

I have an Atec 480W NeoPower.
 
Use a boot CD from here. (My personal webspace so I gotta take it down soon as I don't want to use up all my traffic). and burn the BIOS with the AWDflash onto a separate CD. Then boot with the Boot CD and swap the disks over and place the BIOS CD in. I'm not sure if you can flash direct from the CD or if you have to copy thr files to the RAMDRIVE (a virtual drive stored in your RAM but if you can flash from the CD direct, type this at the prompt: <awdflash N4Dxxx-x.bin> and press 'Enter' [without the < >'s]. This will flash for you in you follow the instructions.

To find out which BIOS you have, run CPU-z and look at the BIOS date under the motherboard tab. Anything before 10th March is too old and you ought to flash.
 
Many thanks.

I'll do that when I get home.

Just going to email them, but it seems when I click send, I get a java error. :(
 
When installing XP on to a Seagate SATA NCQ drive with this motherboard, are there special drivers I should load when it asks me to?

I know you can proceed without any additional drivers but from what I've read it seems that a SATA drive may not be treated properly if you do this, stuff like NCQ not being enabled and the drive being put into 'IDE Mode' or something.
 
wush said:
When installing XP on to a Seagate SATA NCQ drive with this motherboard, are there special drivers I should load when it asks me to?

I know you can proceed without any additional drivers but from what I've read it seems that a SATA drive may not be treated properly if you do this, stuff like NCQ not being enabled and the drive being put into 'IDE Mode' or something.
Click to expand...

You don't need any special drivers during install, but you'll have to install the forceware drivers including the IDE to get options such as NCQ working in windows.
 
You must log in or register to reply here.
Back
Top Bottom