Originally posted by: coolamasta
Right then, been playing with memory and fsb settings and found that running the crutials @ 1067mhz with the cpu @ 1334 5:4 it wasn't stable in memtext86 after about 20 mins but I was on 2.150v for mem...
Decided to try your settings:
4:5 CPU : RAM
VCORE = same
FSB = 1,336
DDR = 835
VDIMM = 2.125V
1.2VHT, NB, VTT = 1.35V
SB = Auto
Timings 4,4,4,10,1T, tRC = 14
and after a quick hour mem test and 2 hours or so on prime95 all seems fine.
what is better for memory latency and bandwidth? 667mhz (1:1) or 835mhz (4:5)
will see how this goes for the next few days and report back
Oh and although I set the memory to 4,4,4,10,1T, tRC = 14 in the bios they show as 4.5.5.16.1T tRC 20 in CPU-Z, whys that?
Thanks for your help again mate, you are a legend with this board
I'm flattered.
The CPU-Z data comes from the eeprom SPD / EPP area of the memory. It shows the guaranteed latency settings at those respective speeds -- IF you are accessing the tab window "SPD". If you are accessing the "memory" window, then something is really screwy there. My CPU-Z shows the "set" latencies just as I set them.
This is most strange, considering that you used my settings to get at least 2 hours stable. I usually test a setting configuration for between 6 and 13 hours. 2 hours indicates to me that if it doesn't last for half a day, you're pretty close to it, and it's "achievable."
Latency requirements mean looser settings as you increase the FSB. So at 667, you are likely to find settings with CAS=3 to be good. There would then be a range over which those settings continue to be stable as FSB increases, and the increase in FSB would require an increase in VDIMM or memory voltage to maintain the tight latencies. Conversely, as you increase the FSB (we're talking about 1:1), you can also loosen the latency settings, quite possibly making it tenable to decrease voltage a tad.
A review article that got me started with the Crucials -- so that I didn't have to experiment blindly -- suggested that settings of 3,4,4,10 were good through around 775 Mhz (DDR); 4,3,4,9 was supposed to be good between 775 and something like 840.
Now that I look at the Anandtech review on Crucial memory, the other reviewer may have had it wrong about 3,4,4,10, but he was probing memory speeds in the 740 to 775 range, and didn't mention lower settings -- as i recollect (and I could be wrong). It was a German review, and my fluency in German is limited -- Ich spreche kleine Deutsch, und viele Jahre seitdem meinem Tage auf Universitat.
but the article had tables of latency results at different CPU : RAM ratios and FSB speeds.
The best Everest memory benchmark "read" result I got at between DDR 710 and DDR 712 was below 9,700 -- more like 9,650 if I remember -- with a 1:1 ratio.
With 4:5 and DDR 835, I get about 9,500. But with FSB = 1,400, 4:5 with DDR 875, 2.15 to 2.175V, the bench result is 9,750. I think that's with the processor at 3.15 Ghz.
While I can get it rock-stable -- now to think of it, my notes show this:
3.15 Ghz
350 Mhz CPU_FSB
1,400 FSB; 875 Mhz DDR (RAM 437.5 = 5 * (350/4)) || CPU:FSB_ratio = 4:5
VDIMM = 2.125V
VCORE = 1.40625V
1.2V_HT= 1.35V
NB_Core = 1.35V
CPU_VTT(FSB)= 1.35V
SB_Core= 1.55V
So the voltage requirement may only be 2.125V.
Now . . . . . the CPU Power-User Magazine reviews of DDR3 modules when put through their paces are blown away in Everest benchmark comparisons with these settings.
Also -- I think I'd bump up the VCORE as shown above to 1.41875V. The "Droop" under load will bring it back to between 1.36 and 1.38V. I know for sure that the higher setting was stable for many hours. Here's the remainder of my notes on those settings:
Timings: 4,4,4,10,1T,tRC = 14
Certified, 8 hrs sFFT; 6 hrs LFFT; 5 hrs Blend @ VCore=1.40625V and 9 hrs Blend at VCore = 1.41875V
Despite the way I phrased my "Certified, . . . . " line, I believe I manually terminated every one of those tests to show 4 x {0 errors, 0 warnings}
Also, many thanks to Idontcare for providing the ftp link to BIOS version 1303. But again -- why does it not appear on the BIOS download page with the other versions 1301 and 1305?
I thoroughly enjoy the Chinese, and taught Chinese graduate students for ten years. Halcyon days. But there is a stereotypical trait -- often described by the word "inscrutable." I cannot figure out why they've obscured the 1303 version, whether this is a random inconsistency, or some . . . . deliberate choice. For now, I'll stick with version 1301. When I upgrade to an E6850 or a 1,333FSB Penryn, I'll flash the BIOS again with v.1305.
Facebook
Twitter