Jump to content
Sign in to follow this  
mrlobber

A voltage tolerance wall at 3.3-3.4 Vdimm on nF4 SLI-D?

Recommended Posts

Ok, here is my story.

 

First, let's start with my system specs I've run for many months just fine.

 

DFi NF4 SLI-D RA.02, bios 510-2

A64 San Diego 3700+ on Thermaltake Sonic Tower + 120mm Panaflo M1A fan

2x512Mb OCZ VX PC3200 (UTT CH-5)

various videocards, (X800XL, a PCI card temporary, now on X550 also temporary)

PCP&C 510W previously, now PCP&C850SSI

HDD's, optical drives, but no other relevant stuff (using Karajan audio etc).

 

The bios settings were (just a basic o/c without much time spent tweaking for the maxium cpu speed) as follows:

 

FSB Bus Frequency - 230

LDT/FSB Frequency Ration - 4x

CPU/FSB Frequency Ratio - 11x

PCI eXpress Frequency - 101Mhz

 

CPU VID StartUp Value - StartUp

 

CPU VID Control - 1.400v

CPU VID Special Control - Above VID * 104%

LDT Voltage Control - 1.30v

Chip Set Voltage Control - 1.60v

DRAM Voltage Control - 3.20v (using the 3.3V Vdimm jumper position)

 

DRAM Configuration Settings:

 

DRAM Frequency Set - 200 (DRAM/FSB:1/01)

Command Per Clock (CPC) - Enable

CAS Latency Control (Tcl) - 2.0

RAS# to CAS# delay (Trcd) - 02 Bus Clocks

Min RAS# active time (Tras) - 05 Bus Clocks

Row precharge time (Trp) - 02 Bus Clocks

Row Cycle time (Trc) - 07 Bus Clocks

Row refresh cyc time (Trfc) - 16 Bus Clocks

Row to Row delay (Trrd) - 02 Bus Clocks

Write recovery time (Twr) - 02 Bus Clocks

Write to Read delay (Twtr) - 01 Bus Clocks

Read to Write delay (Trwt) - 02 Bus Clocks

Refresh Period (Tref) - 3120 Cycles

Write CAS Latency (Twcl) - Auto

DRAM Bank Interleave - Enabled

 

DQS Skew Control - Auto

DQS Skew Value - 0

DRAM Drive Strength - Level 5

DRAM Data Drive Strength - Level 2

Max Async Latency - 7ns

Read Preamble Time - 5ns

IdleCycle Limit - 256 Cycles

Dynamic Counter - Disable

R/W Queue Bypass - 16 x

Bypass Max - 07 x

32 Byte Granularity - Disable(8 Bursts)

 

However, exactly the same settings were tested to be memtest stable up to 260x10 (just with 3.4 Vdimm and LDT link to 3x). The little problem always sitting in my mind was that, as soon as I upped the voltage to 3.5 Vdimm, memtest started erroring and crashing in tests #5 and #8, and I never was able to find out why (of course, memory and Vdimm jumper area actively cooled). However, since I wouldn't run 3.4+ Vdimm daily anyway, I left the issue not much further investigated, and to my thread I posted here on dfi-street.com long ago were no relevant answers anyway.

 

Well, all this was just an introduction :nod:, now the main story follows.

 

Not so much ago got a 1 Gb kit of old school Bh-5 Mushkin PC3500 LevelII which previous owner claimed to be memtest #5 and Superpi 1M stable on his DFIn F4 SLI-DR @ 260fsb 3.3Vdimm (now you probably want to say that means nothing, but read on ;) )

 

First tested on my P4GD1/Dothan 740 system for a while and was able to SuperPi 1M 2-2-2 up to 222fsb 2.9Vdimm.

 

Now decided to play on DFI nF4 (first just with 3.2Vdimm). Put the Mushkins in (kept the bios settings mentioned earlier), and bang! the system kept restarting right after post screen as soon as fsb exceeded 230. With 2.9Vdimm, the stability was up to ~221, just as expected, so the memory obviously worked fine.

 

Hmm, I thought, wtf. Played around more (swapped slots (from yellow to orange), switched the Vdimm jumper as well),everytime the same happened while under 230fsb the system run just fine. Put the VX back in and the reboots continued whereas previously I didn't experience any problems :confused:

 

Well, since clear CMOS was probably needed anyway at this point (if memory previously stable is not stable anymore at the same settings where it was before) I decided it's time to flash to a newer bios (after all, May 10th is almost 5 moths back, right? :D )

 

Chose latest bigtoe modded 704-2BTA. Flashed fine. With the settings above, the system again experienced the 230fsb barrier with either OCZ or Mushkin at both 3.2 as well as 3.4 Vdimm. Searched Xtremesystems for good xtreme old school BH-5 settings, tried those (mainly - Trfc of 12, Idle cycle limit of 016, various Tref values - Auto, 3072, 3120, 4708, 0638 etc, various drive and data strength combinations), the outcome was negative all the time.

 

Heh, I thought. Maybe this bios is not for my particular board after all (heard at least twice when people experienced similar reboots)? Flashed to 623-2 then. The same. Played around and around as described in previous paragraph, no stability. One moment, was suddenly able to boot up to 245 fsb (not memtest stable though), on the next reboot for tweaking - nothing again.

 

Finally flashed back to 510-2. Loading the saved settings with CMOS Reloaded and playing around, just decided to try the Trfc value of 12 once again without much hope. Ouch! Suddenly, with the Mushkins, I was able to use the Trfc = 12. Changing the LDT ratio to 3x, finally stabilized at 255x10 @ 3.3 Vdimm :cool: , with the settings displayed in the screenshot:

 

I shall admit that while the memory timings were not changed except for Trfc = 12, the advanced DRAM configuration settings this time looked this way:

 

DQS Skew Control - Auto

DQS Skew Value - 0

DRAM Drive Strength - Level 8

DRAM Data Drive Strength - Level 3

Max Async Latency - 7ns

Read Preamble Time - 5ns

IdleCycle Limit - 016 Cycles

Dynamic Counter - Enable

R/W Queue Bypass - 16 x

Bypass Max - 07 x

32 Byte Granularity - Disable(8 Bursts)

 

So after those 3 days of painful memory testing, I had finally arrived to my new safe base where to start another takeoff for higher fsb's :angel: (moreover, the previous owner's claims turned out to be almost true, those 5fsb difference might be attributed to various other factors) . But now I run into the problem which I mentioned at the very beginning: 3.4 Vdimm and above would do nothing for memory speed but just produce memtest errors. I would understand BH-5 not scaling well above 3.6Vdimm, but for 3.4 and 3.5, at least marginal speed increases should be gained, shouldn't they? Yet it doesn't happen, but just errors come out. Moreover, the errors come out even at lower speeds, for instance, 245x10 @ 3.5Vdimm, sometimes the board starts the rebooting cycle with such settings. Multimeter measuring directly at the memory slots confirms the Vdimm voltage seemingly doesn't fluctuate even when set to 3.7Vdimm. If the mysterious fsb wall's at various bios tested described earlier might be attributed to not-a-proper-ram-settings-combination, what to say about this Vdimm wall?

 

This is not an overheating issue, because it happens also after the system is switched off and let to cool. Moreover, just as I said earlier, both memory is cooled by a 120mm fan as well as the 5Vdimm jumper area - by 80mm fan.

 

At this time, I've tested 3 PSU'S (the PCP&C's also a Chieftec 410W native 24pin ATX2.03, 2 CPU's (with the previous San Diego I owned, the system showed the same behaviour), these two memory kits (OCZ VX and Mushkin LVLII), even migrated through 3 videocards, and only the motherboard has been the component which unites all these cases. Unfortunately, my DDR Booster I got exactly for the behaviour testing (to see whether the high voltage intolerance on both of my memory kits is board related), was DOA, so it's in RMA right now. However it might still be I've missed some important settings somewhere, although with all my knowledge I can't image where could that be.

 

So I'm asking for your thoughts, if you had enough patience to read all that I wrote here :shake:

Share this post


Link to post
Share on other sites

Nice article :) When i use over 3,4V on me memory the only Tref value (with the VX) that doesnt give me errors in memtest is 1168. Not a clue why they stop scaling though, me VX will go up untill i hit 3,7v. I use these settings, maybe theyll help:

 

Genie BIOS Settings:

LDT Voltage Control - 1.30v

Chip Set Voltage Control - 1.50v

DRAM Voltage Control - 3.40v - 3,70v

 

DRAM Configuration Settings:

DRAM Frequency Set - 200 (DRAM/FSB:1/01)

Command Per Clock (CPC) - Enable

CAS Latency Control (Tcl) - 2.0

RAS# to CAS# delay (Trcd) - 02 Bus Clocks

Min RAS# active time (Tras) - 05 Bus Clocks

Row precharge time (Trp) - 02 Bus Clocks

Row Cycle time (Trc) - 07 Bus Clocks

Row refresh cyc time (Trfc) - 16 Bus Clocks

Row to Row delay (Trrd) - 02 Bus Clocks

Write recovery time (Twr) - 02 Bus Clocks

Write to Read delay (Twtr) - 01 Bus Clocks

Read to Write delay (Trwt) - 02 Bus Clocks

Refresh Period (Tref) - 1168 Cycles

Write CAS Latency (Twcl) - Auto

DRAM Bank Interleave - Enabled

 

DQS Skew Control - Auto

DQS Skew Value - 0

DRAM Drive Strength - Level 8

DRAM Data Drive Strength - Auto

Max Async Latency - Auto

DRAM Response - Normal

Read Preamble Time - Auto

IdleCycle Limit - 16 Cycles

Dynamic Counter - Disable

R/W Queue Bypass - 16 x

Bypass Max - 07 x

32 Byte Granularity - Disable(4 Bursts)

Share this post


Link to post
Share on other sites

i find this very interesting because i run 3.3v 24/7 (because i can, i guess) but upping it to 3.5v even though i have more than adequate cooling causes errors.

 

what do you suggest I do? this is really bothering me!

 

EDIT and i always see Disabled (8 Bursts)... I can only do 4 bursts as the highest..?

Share this post


Link to post
Share on other sites

loc.o, thanks for the insight. Tref of 1168 doesn't do anything for me, just hurts the bandwidth by ~2%.

 

Played around more, this time, looked around in Windows as well. Was able to complete Superpi 1M at 260x10 @ 3.3Vdimm and 264x10 @ 3.4Vdimm. However, with 3.5Vdimm, Superpi 1M fails even at 255x10. Any ideas what's going on with the Vdimm on the board?

Share this post


Link to post
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

Sign in to follow this  

×
×
  • Create New...