Failed Memtest with DSL3 and Crucial Ballistix 800Mhz

[rockyjohn]

New build:
GA-P35-DS3L with BIOS updated to Fb8
e8400
4x1GB Crucial Ballistix PC6400 (from mobo list)
Voltage to 2.2v, timings to 4-4-4-12

Like others, wish I had known about this forum before buying Crucial memory (although it has worked well for me in past) for this board. I reviewed quite a few similar posts, but none seem to address my issue when testing seems to point towards a motherboard issue - or maybe some other settings that need to be changed.

The computer works fine but repeatedly failed Memtest86+v2.01, almost always on test 6 with similar if not exactly the same error report. I ran two tests for 20 hours with 31 passes and 29 errors.

I reversed the order of sticks (1-2-3-4 became 4-3-2-1), putting each in different channel and reversing order in channel. Repreated the test. In 11 hours, completed 18 passes with 5 errors.
Tested each stick individually and all passed.
Tested one stick in each slot; it worked for three slots but errors on slot #3. The last two tests taken together seem to point to a motherboard issue.
Set timings back to 5-5-5-18. Repeated with one stick in slot 3; passed. I was hoping this would be a compromise fix. But when I installed all four sticks; completed 9 tests with 28 errors.

Tested with Prime95 v25.6 using blended test to stresse memory - ran for 1 hour, 4 passes, no errors.
Increased block size to increase stress on RAM; completed passes, no errors.
Changed timings back to 4-4-4-12, ran test 4 passes, no errors.

Based on some posts by Lsdmeasap, reduced memory voltage to 2.1 and increased FSB and MCH by +0.1 to increase stability.

I then ran Memtest again, test #6 only, 7 passes with 5 errors .

Now what? Do I have a mobo problem - or are there other settings I should change or other things I should try?

Here are my current settings:
Robust Graphics Booster Auto
CPU Clock Ratio 9X
Fine CPU Clock Ratio 0.0
CPU Frequency 333x9
CPU Host Clock Control Disabled
x CPU Host Frequency (Mhz) 200
PCI Express Frequency (Mhz) Auto
C.I.A.2 Disabled
Performance Enhance Standard
System Memory Multiplier (SPD) Auto
Memory Frequency (Mhz) 800 800
Advanced Timing Control
DRAM Timing Selectable Auto
CAS Latency Time 4
Dram RAS# to CAS# Delay 4
Dram RAS# Precharge Delay 4
Precharge Delay (tRAS) 12
ACT to ACT Delay (tRRD) Auto=3
Rank Write to READ Delay Auto=3
Write to Precharge Delay Auto=6
Refresh to ACT Delay Auto=52
Read to Precharge Delay Auto=3
Static tRead Value Auto=6
Static tRead Phase Adjust Auto=1
***System Voltage Optimized***
System Voltage Control Manual
DDR2 OverVoltage Control +0.3*
PCI-E OverVoltage Control Normal
FSB OverVoltage Control +0.1
(G)MCH OverVoltage Control +0.1
CPU Voltage Control Normal
Normal CPU Vcore 1.21250v

[Lsdmeasap]

First off, it could just be the use of a Beta BIOS. I see you have to use one so you may be out of luck until a non beta is posted. But just had to tell you that because EVERY beta I have used caused me ram errors. You could try F8a, it supports 8400 as well, as does F8f, but then again all three being beta could give you ram errors

I would keep the voltage to 2.1 or less unless you have a Fan on the ram, as for MCH you may need .125 or .15 for 4 sticks, so try that as well. As you have read, I personally think GA and crucial have issues, but some disagree and get them working fine, Lets see if we can for you.


Try these settings >>>

Robust Graphics Booster Auto
CPU Clock Ratio 9X
Fine CPU Clock Ratio 0.0
CPU Frequency 333x9
CPU Host Clock Control [Enabled]
x CPU Host Frequency (Mhz) 333 << manually set to 333
PCI Express Frequency (Mhz) 100
C.I.A.2 Disabled
Performance Enhance Standard
System Memory Multiplier (SPD) 2.40
Memory Frequency (Mhz) 800 800

Advanced Timing Control
DRAM Timing Selectable [Manual]
CAS Latency Time 4
Dram RAS# to CAS# Delay 4
Dram RAS# Precharge Delay 4
Precharge Delay (tRAS) 12

Set to the highest value below, then later you can move them lower ONE AT A TIME testing after each move
ACT to ACT Delay (tRRD)_________3-4
Rank Write to READ Delay________ 9-11
Write to Precharge Delay_________ 3-5
Refresh to ACT Delay______________ 48/52-62
Read to Precharge Delay__________3-6
Static tRead Value_______________ 7-10
Static tRead Phase Adjust________ 1-31 << Leave on AUTO or Zero/0 Or 1-31

System Voltage Control Manual
DDR2 OverVoltage Control +0.3*
PCI-E OverVoltage Control Normal
FSB OverVoltage Control +0.1
(G)MCH OverVoltage Control +0.1 << May need .125 or .15
CPU Voltage Control Normal
Normal CPU Vcore 1.21250v

Here is/was likely also you main issue I saw >>>

Static tRead Value Auto=6
Static tRead Phase Adjust Auto=1

You cannot use a phase adjust value with 6, as it will make tread value run at 5 most of the time. I doubt you will even get 4 sticks to run with a tRead value of 6, so best to use 7 or above and do not use a value AT ALL for phase adjust

[rockyjohn]

EUREKA.
I followed your suggestions and they solved the problem. I ran Memtest, first for one hour and 7 passes on test #6 (which created all of the failures before) and then for 7 hours and 10 passes on all tests - all without error.

The MCH OverVoltage Control I left on +0.1 as I could only increment in units of 0.1 so I could not go to 0.125 or 0.15.

I now get a warning on my BIOS screen - the System Voltage Control header changed to a flashing red warning "System Voltage NOT Optimized". Is it safe to continue ignoring this?

Regarding the BIOS version, I went to F8b thinking that since F8a, the version I needed for Wolfdale was still beta, any immediate problems Gigabyte had with it might have been fixed in the release of F8b. However I now see this might have ignored any new issues created in F8b. Since it now seems to work ok - at least thus far with the memory changes - do you think I should stick with it or change to F8a to be safe? If they make final changes to F8a to make it non-beta, will they just take the beta label off it or issue it as a new number? I assume if they issue a higher number non-beta I should flash it, but if they just take the "beta" designation off, will it have changes that I need to flash?

How can I measure the effective speed of my memory to have some idea or measure of the impact of the current setting changes and the future ones as I roll back the higher numbers?

Regarding the unit by unit rollback of the higher settings, any suggestions on which I should try first? Any special pairings or relationships that I should be cognizant of?

Thank you for all of your help. I hope Gigabyte appreciates all the RMA's you've saved them from. I am more than a little surprised that when I purchase a mobo from them for a CPU they say it supports, I find it only supported with a beta bios. And then to select memory off the Qualified Vendor list by part number and have these problems is not what I expected - particularly given what appear to be known issues. Perhaps you could suggest (if you have not already) that they remove this memory from the list. I know this board is being replaced by an energy efficient one, does it seem to have the same problems with the Crucial Ballastix memory? I see Crucial does not even list this board (it lists rev 1.0 but not 2.0) in its memory selection tool.

I bet you have made a lot of friends on this forum. Ever think of running for President? We could use a good man - oops - person - who knows how to troubleshoot and tweak things to make them work. Someone who knows how to make the all the pieces work together.

rockyjohn

[Lsdmeasap]

Originally Posted by rockyjohn

EUREKA.
I now get a warning on my BIOS screen - the System Voltage Control header changed to a flashing red warning "System Voltage NOT Optimized". Is it safe to continue ignoring this?

Great to Hear!!

For sure, that is just there to remind you incase you set things Very high or do not know what you are doing. It will always say that if you manually set many things. NO need to worry at all

As for the BIOS, like you said each may have fixed something the other did or did not, so maybe F8f would be better to try? But you may just be best to leave it for now until a final comes out. It will likely just be F8 with no letter

You can use Everest Ultimate to do bandwidth tests, The free trial does not offer all sections of the tests for you to view but it is a good tool to use and see some changes. The benchmark you want will be in Tools, then cache and benchmark in the upper menu>>>>>>
EVEREST Ultimate Edition Download - Softpedia

SiSandra Software is another great tool for this
SiSoftware Zone

As for the settings, I would start with rolling back Static tRead first as it has the most impact on performance. Act to Act next, then read to precharge, then write to precharge, then refresh to act, then write to read delay. Then finally I did not list it but you can try 4-6 on write to read delay and see if it benchmarks any better then 9-11, sometimes it does and sometimes not

Well some do get Crucial to work, but I suggest against it everytime from helping many users I say it is not compatible at all. Probably why you say you dont see it listed at Crucials site! Yeah, "E" versions have the same issues. Like I said though it is not a Official thing at all, just my opinion based on MANY threads read, threads helped with and personal experiences in trying to setup systems for customers. As you know MANY people LOVE crucial and for good reason, they USED to be the best (Again my Opinion) but they have since started using different chips then they used to in the past that got them their great Reputation, still Micron as they are they same company but different chips then those that they used to use (D9's). And in the past few years new or old chips both seem to DIE often, with low or high voltages, timings or for just no reason. But even the older chips still are not good with GA boards I think.

Good luck in your timing tweaking, just keep in mind it is not a simple task and takes some time and effort to find the sweet spot. Take advantage of the BIOS profiles feature and save you a stable high area of settings to use when you dont have time to play and save a separate one that you are working on getting better timings in.

F11 on the Main BIOS page is Save, and F12 is Load a profile. All you do is save the one you are using now on the main page. The go in and make some changes and go back out to the main page and save again using a different name and then you will have 2, you can use up to 8 or more depending on the board. I love this feature it saves much time in getting things done, more people should use it for sure!!

Again, good luck and post back if you run into any issues that stump you.

Thanks for the kind words, Good to hear after some of the negative replies I often get here by random posters!

[rockyjohn]

My tweaking has not been as productive I had hoped.
I have not touched the primary timing settings - 4-4-4-12, nor these:

CPU Host Clock Control
CPU Host Frequency (Mhz)------------333
PCI Express Frequency (Mhz)----------100
Performance Enhance-----------------Standard
System Memory Multiplier (SPD)-------2.4

DDR2 OverVoltage Control -----------+0.3
FSB OverVoltage Control -------------+0.1
(G)MCH OverVoltage Control ---------+0.1

Day one went well - slowly changing from stable to day one - as noted below with default (unstable) for reference. At end of day one I ran Memtest overnight for overnight for 9 passes without errors.

I made several successful adjustments on morning two - also shown below - then the I started losing ground.


-----------------------------Default---Stable----Day1-----Mid2
ACT to ACT Delay (tRRD)--------3--------4--------4--------4
Rank Write to READ Delay-------3--------11-------6--------5
Write to Precharge Delay--------6--------5--------5--------5
Refresh to ACT Delay----------52--------62-------58------58
Read to Precharge Delay--------3--------6--------6--------4
Static tRead Value-------------6--------10--------8--------8

I lowered Rank Write to READ Delay to 4 and had Memtest error.
I returned it to 5 and test error - even though this had passed a 3 pass test before.
Raised to 6 and retested - failed again.
Not wanting to push this higher than the overnight pass configuration, I left it at 6 and and raised Read to Precharge Delay back to 5 and it passed. Lowered Static tRead Value to 7 and it failed. Raised it back to 8 - where it had passed one trial before - and it failed. Left it at 8 and lowered Read to Precharge Delay to 4 to make all settings match previous good configuration and it failed. Now I am going to raise it back to 6 to match the configuration I ran last night and run it tonight - and hope I have not lost any ground. The failed tests are summarized below, starting with the good mid-day and showing changes.


____________________________-------------CHANGES--------------_____
--------------------------Mid2---Fail--Fail--Fail-Pass--Fail--Fail--Fail--Final
ACT to ACT Delay (tRRD)----4------------------------------------------4
Rank Write to READ Delay----5-----4----5----6-------------------------6
Write to Precharge Delay----5------------------------------------------5
Refresh to ACT Delay-------58-----------------------------------------58
Read to Precharge Delay---- 4--------------------5----------------4----4
Static tRead Value----------8--------------------------7----8----------8

Any suggestion on a better course for trying to reduce the settings?

Rocky
-

[Lsdmeasap]

Well I dont know what you are doing, I dont think!

Act to Act need only 3-4 to pass, anything over 4 should not fail at all unless you have something else wrong first

This does not work here? >>>

ACT to ACT Delay (tRRD)----3-4
Rank Write to READ Delay----9-11
Write to Precharge Delay----5
Refresh to ACT Delay-------52-58
Read to Precharge Delay---- 4-5
Static tRead Value----------7-8

I know you posted much info, but I think I just am confused about how it is laid out. I am sure it is clear to you, but with the two tables I got lost.

[rockyjohn]

Sorry for the confusion.
First, the numbers you were seeing as Act to Act Delay where on the wrong line of the second table - they should have been on the second line for Rank Write to Read Delay changes. The error must have been my memory acting up - oops except that I wrote it on my other PC while running Memtest on the subject machine.

The first table shows the first changes and lists the entire configuration for each trial. The last column of the first table is the first column of the second table. After the first column in the second table, I entered only the numbers that changed for each trial - I added a header over those - and then I added my final config. Hope that makes it easier to understand.

My current numbers appear reasonably close to the config you listed above - but I thought it was generally the higher end of the numbers and it was reasonable to expect better. My original Memtests appeared to have a single repeating error. My new settings have 5 of 6 settings higher than the original defaults with the changes totalling about 15 clock cycles.

I ran the Sandra benchmarks, but still have not had a chance to learn what they really mean. However I also ran it on my old Dell and compared the numbers for its 3 GB of 667 MHz memory with the new 4 GB of 800 MHz and I thought the new would be comparatively higher. The Dell is the second one -

Benchmark Results
Int Buff'd iSSE2 Memory Bandwidth : 5.91GB/s-------------4.76
Float Buff'd iSSE2 Memory Bandwidth : 5.90GB/s-----------4.73
Results Interpretation : Higher index values are better.

Performance vs. Speed
Int Buff'd iSSE2 Memory Bandwidth : 7.57MB/s/MHz------9.15
Float Buff'd iSSE2 Memory Bandwidth : 7.55MB/s/MHz-------9.11
Results Interpretation : Higher index values are better.

In what order would you suggest I try to reduce my settings from where they are now ?

Rocky

[Lsdmeasap]

OK, well I just want to let you know what you think is "Default" the value shown on the Left and what you thought you got with AUTO is not what you get. it is what is loaded in the BIOS only at that time, and often changes within the BIOS itself and within windows.


No, lower the number the better on most all settings except Rank Write to Read Delay, that one differs from setting to setting depending on the chipset, ram and other settings. but the rest, the lowest you can get stable is best


I am not sure about SiSandra, as I dont use it as much these days, but this is what my 2x1 800Mhz Mushkins at 900 Mhz 4-5-5-12 get in SiSandra>>>
Int Buff'd iSSE2 Memory Bandwidth : 9.84MB/s/MHz
Float Buff'd iSSE2 Memory Bandwidth : 9.96MB/s/MHz

Can you compare cache and memory benchmark in everest? Tools, then cache and benchmark. I would just think you do not have the correct best timings as of now is why they are so low. Act to Act And tRead likely are why, can you not use 3 on Act to act, and 7 on tRead?

Set act to act to 3, tRead to 7 and raise the rest all back up for now, then lower them (All the that you just raised slowly again) leave act to act and tRead set to 3 and 7

[rockyjohn]

I have made a lot of progress following your suggestions. My current settings are:

ACT to ACT Delay (tRRD)	3
Rank Write to READ Delay	11
Write to Precharge Delay	5
Refresh to ACT Delay	52
Read to Precharge Delay	3
Static tRead Value	6

Leaving Rank Write to Read Delay at 11, I was able to move everything else back down to the upper then lower levels you listed and the last two beyond that back to the default settings. So it looks like Rank Write is the culprit. I will now set it lower and see how far I can get and remain stable, unless you suggest otherwise.

I am going to download the trial version of Everest and will test tonite.

I tried to download free PC Mark 2006 - but after a download of almost 1 GB it would not open.

I ran Sandra again and the tests reported about a 10% improvement - not the amount I was hoping for - and I am not sure how or what it is reporting either. I tested memory at PCPitstop.com and it reported the memory at 10.4 GB/s compared to a "normal" of 11.26.

[Lsdmeasap]

Yes, try and lower it and see how it affects you. Like I said it does vary from ram to ram and board to board what that setting will do.

So you got tRead to work at 6 eh, nice!!! Here, is PCMark 05 this is what I use (But not the free version) >>>
|MG| PCMark05 1.2.0

Just so you know, here in Everest is the only test I use Everest to bench with and look at (I have never used any of the other Benchmarks in it) I just use 3DMark 06 and PCMark 05 >>>