Hi all, after several days test and effort, now, when you see that thread topic,
you might be thinking about OPB is showing his cherry picked again now. However,
let me seriously tell you right now, this is 100% not cherry picked but just randomly selected
from the 30 pairs out of 100 pairs I have done for the SPD programming. I was quite surprised that
characteristics of Micron 5bf is totally different from the other memory modules. Initially, I was very very frustrated
when I saw Ozzimark can get over 290fsb 1:1 1t. For me, no matter how hard I tried,
I have never got pass 285fsb except use upper divider when I ran fx60 under single stage(cold bug).
Due to last time, some users also struggle and complain that their ram got weird spd just showing PC3200, so
that's why I got to do the QC and do the SPD. After get used into it, I figured out this IC is, well I can't say it's King of 2GB set, but
I strongly believe that all this results are based on the non-cherry picked and non-bias bench. If you really wanna say something,
that might be on my PSU or Venus. Well, but I tell you what, most of these pairs passed DDR600, 1:1, 3-3-3-3-7-12...
only required 2.78v~2.80v under ultra-D. Also, I found that the DS/DDS of Ultra-D is way more effective than Expert or Venus !!
but for instance if you wanna run 4x1g 2t , you gotta use Expert or Venus for sure. Let's go ahead and discuss all these myth of Micron 5BF.
Let's start from 2 dimm DDR600, Venus/Expert platform first:
Conditions:
Before we start, you will need to make sure, there is a beta bios 406 for all DFI nf4 series boards, go flash that bios first!!
Here I must appreciate Tony really did a great great job, this bios play a big big role of 5bf can "easily" get on toDDR600 with that unbelievible
tight parameters.
Basically, we can look at the picture below, honestly, I was thinking about ozzimark has mentioned on the 5bf thread about we can try 2t first,
so after I finished the SPD programming on the first pair, I set 300x11 under bios and certainly CPC option I set was disable. After I boot into windows
I also think if I want to run 2t, why don't I just have all the setting are the tightest waywithout probles; therefore, initially, I did not set TRFC as 12 just like you
see in the screen below. but I did set tras and icl as 8(cuz I don't want to jump up too much from 16cks) with TWTR as 1; and wow...it did not fail..
But we all know , you will need to run that several times for 1m to make sure it's right. During this moment, when it was runing, I intentionally switch the 2t
command from enable to disable (1t), and it also passed without any problem.I was really excited at that moment when I stared the 1t at the CPUZ 133 memory
timing confirmation. so then I keep those timing as refernce and try to get the paten on 5bf. Certainly, the best way to confirm is running every length of super
pi. so let's see here.
From this screeny, we can see from CPUZ, my primary parameters setting is 3-3-3-0-7-8; I at least ran 5 times by using this setting.
Please read through all the notes.
super pi 1m:
2m, are you guys aware of what 2 timings I changed? Yes, I changed Tras to 2 and ICL back to 16clks; honestly,
256 really do nothing and even failed to complete calculation.
Now, what about 8m? yup, I change TREF back to 3120(200mhz, 15.6us), Tras to 2 and then ICL as 0 clock; during this 8m running, I got couple
errors, such as " not exactly run" or "not convergent in SQR"...that all happened around the 6~7th iteration. Therefore I thought that 2.78 won't be enough for
running at DDR600 with those tight timing! (think about the TRFC as 12), so I restart and add from 2.84 to 2.87v(it's off about 0.04v compare with mbm5
detected value), , no wonder, it just running smoothly all the way end.
Now, here is the insteresting part about 16m, when I finished 8m running, I accidentally thought, holy xxxx, maybe I might not have to give that vdimm for
more 0.03~0.05v, so I went back to check my setting and think if I get ICL relax a bit and set Tref as 100mhz/1.95us, I might have chance to do
16m finish. Alright , here we go, this result verified my assumption as well.
Now, it's the time to challenge the most difficult one though. Honestly, I got about 4 times fails before I understand the trick.
First of all, from the results and ideas above, we can apparently sort out that most timings on the a64tweakers table are basically fixed;
consequently, those variables related to our discussion are:
TREF, TRAS and ICL
Certainly, not just these 3 timings, before this step, you gotta completely understand some key factors, such as how to set DS/DDS, and remember, trrd can be
set as 3
as well as TWTR can be changed to 2 "in order to stablize a bit" (actually really not much! haha ); my concept is " Performance can NOT be hoding back due to
lack of stability."
THerefore, I changed ICL back to normal value 16, so of course I can't just backward, then I also put 2 as TRAS to balance that along with Tref switch to
133mhz, 1.95us to complete this 32m calculation.
** well, I still think I need to be fair to show the camera pictures, 4 of them. If anyone live closed to Columbus, You are more than welcome to pay me a visit
and ramdomly choose any two sticks , I can do my best to show you how to!
Now the CPUZ verifiedmy 32m pic got verified number on that)
**Also, we might want to check out about bench results from the Everest ultimate :
well, how about Opteron? To be contiuned...
************************************************** ****************************
Here are some results just verified what I said from above, ultra-D looks like no need that much vdimm and not that confused DS/DDS setting
than either Expert or Venus
before I post my 1m, 8m, 16m, and 32m, I also sure no problem to share the bios setting here:
actually, you can see the timings I used under ultra-D are even tighter then Venus
1m:
So for the 2 dimm DDR600 1t , we now have rough idea how this ram is if you treat it correctly. let's see next, full dimm
4x 1gb on DDR617 stability test.
TO be next...
************************************************** ************************************************** *************************************
4Dimm 1g x4 fully stable for DDR617 ; current WR
Actually , you can see I was pretty closed to what I figured out the paten of 5bf later on, because what I thought at that moment of running 4 dimms
is this merely just for 2t timings, so I can let the timings as tight as it could be, so if you look at that carefully even DDR617 2t timing like that will not be too
bad though.
CPU verified:
windows captured screeny:
Digital camera pics , took during running .
Again, hopefully, this will clear some confusion,!! If you want to play micron 5bf, you gotta put a lot of effort, I mean a lot. Seriously.
OPB
Originally Posted by Philosophy
Bookmarks