2*TWINX1024RE-3200LLPT on ASUS SK8N

[leibold]

I'm pretty sure one of the 4 memory modules (2 packs of TWINX1024RE-3200LLPT) is bad and I already contacted the reseller to replace it. However I still would like you to have a look over the symptoms to see if you have any additional suggestions.

All modules are marked:

CMX512RE-3200LLPT XMS3200REv2.1

0443079-2

XMS3200R 512MB 400MHz 2326

 

The system is an ASUS SK8N v1.03 Bios 1004 with Opteron 148 revision C0 in an Antec 1080AMG case with 430W power supply. After getting some system hangs during the installation of SuSE 9.1 for x86-64 I ran memtest86 version 3.1a to troubleshoot the system. This reported plenty of errors and in order to better understand the memory situation I disabled ECC in the bios and also disabled interleave (I do want both of those features on during normal operation). Running one pass of memtest86 with only the standard tests results in 62444 total errors, however the errors were confined to one 512MB address area (1024-1536MB) and only effected 3 data bits (bits 1, 5 and 7 since the err-bits patterns where 00000080, 000000A0, 00000082, 00000002, 00000020). I also noticed that while a lot of errors where reported in each of tests 1,3,4,5 and 6 there didn't appear to be any in tests 2 and 7. The lower 20 bits of the error address also seemed to be always the same, but I forgot to write it down.

I did not overclock the system nor do I have any plans to

do that. The bios properly detected all memory timing parameters correctly and I left Vmem at 2.5V.

 

Just for curiosity, what memory chips are used in these modules ? Except for slot A1 which is exposed to the cold air from the cpu fan the heatspreaders on the other memory modules where getting noticably warm.

[RAM GUY]

Please make sure that you have the latest BIOS for your MB, and set the following settings on your BIOS;

Advanced Chipset:

Memclock Mode: Limit

Memclock Value: 2:1 (DDR400)

Bank Interleaving: 4/Enabled

Burst Length: 4

SDRAM CAS Latency: 2T

SDRAM RAS to CAS Delay (tRCD): 3T

SDRAM Active to Precharge Delay (tRAS): 6T

SDRAM Row Precharge (tRP): 2T

Master ECC Enabled: Enable

Jumper Free/Frequency Voltage Control:

Spread Spectrum: Enabled

CPU FSB Clock: 200

DDR Voltage: 2.7/2.75 Volts

All other settings should be set to default settings!

Then please test them one at a time with http://www.memtest.org and let’s make sure it's not some other issue! I would run the test for at least 2-3 passes to be sure!

[leibold]

Please make sure that you have the latest BIOS for your MB,

There are two newer bios revisions available for this board. A production release bios with version 1007 and a version 1008.4 pre-release beta bios. From the limited information available through ASUS it seems that the beta bios adds support for Opteron 150 and Athlon 64 FX-53(940). Which bios do you suggest, latest released or latest beta ?

 

and set the following settings on your BIOS;

Advanced Chipset:

Memclock Mode: Limit

Memclock Value: 2:1 (DDR400)

Bank Interleaving: 4/Enabled

Burst Length: 4

SDRAM CAS Latency: 2T

SDRAM RAS to CAS Delay (tRCD): 3T

SDRAM Active to Precharge Delay (tRAS): 6T

SDRAM Row Precharge (tRP): 2T

Master ECC Enabled: Enable

Jumper Free/Frequency Voltage Control:

Spread Spectrum: Enabled

CPU FSB Clock: 200

 

Why do you suggest to turn on ECC when testing the memory modules ? This could hide single bit defects since ECC would silently correct those.

With regards to interleave, this feature significantly increases performance but makes it harder to interpret the results of the memory test since the errors are distributed over a wider address range. Or is interleave confined only within the banks of a single memory module ?

I did not see any settings for burst length or spread spectrum and all other settings were as you specified them.

There were additional settings you didn't mention for the background ECC scrubbing of DRAM,L1 and L2 cache which I all left at the bios default setting of 'disabled'.

DDR Voltage: 2.7/2.75 Volts

This seems quite high. If these are 2.5V memory chips (you didn't answer my previous question about the type of memory chips used in these modules) then 2.75V represents a 10% increase (which typically is the maximum allowed). The Opteron datasheet (AMD 31412) specifies a maximum of 2.65V for the memory voltage (DC operating conditions). As I said in my first post, the memory modules are getting noticably warm even at 2.5V. I'm concerned that increasing the voltage would also increase the heat which in turn can be a cause of random memory failures and reduced lifetime.

Are you absolutely sure you want me to increase the memory voltage this high, even if there really is only one defective module and all other 3 modules are operating error free at the default 2.5V ?

All other settings should be set to default settings!

Then please test them one at a time with http://www.memtest.org and let’s make sure it's not some other issue! I would run the test for at least 2-3 passes to be sure!

I did run memtest86 (only the standard tests) multiple times and the errors were consistant (same number and types of errors each pass). Thanks for the link to memtest86+, I'll give that one a try too.

[RAM GUY]

The recommended voltage settings for our XMS modules will be 2.7/2.75 volts depending on the BIOS settings that you have. I am positive that I wanted you to set the settings on those settings, and to try and test them with http://www.memtest.org and let’s make sure it's not some other issue! I would run the test for at least 2-3 passes to be sure!

[leibold]

I upgraded to the latest Beta bios 1008.4. I had difficulties with the AFUDOS 2.11 utility and ended up flashing with the ALT-F2 bios update feature. The 1007 bios has a comment that AFUDOS 2.07 is required and perhaps that applies to the beta bios as well.

 

I repeated the testing with memtest86+ v1.30 and must admit that I like the improvements over memtest86. However it didn't change the fact that exactly one of the four memory sticks was bad and I had one of the two TWINX sets replaced by the reseller. I installed the new set today and it already passed a couple of times the memtest86+ standard tests. I'm going to run the extended tests over night, but I don't expect any problems.

 

I think I can answer some of my own earlier questions:

 

ECC: When ECC is disabled, the memory test will only test the 64 databits (0-63) and leave the data correction bits (64-71) alone and therefore untested. With ECC enabled all 72 bits get tested, however any single-bit errors will get automatically corrected. Depending on the memory controller, having ECC enabled may obscure the exact location of the error but as long as the memory test supports it (and memtest+ is clearly more up to date in this area) those corrected single-bit errors will still be counted. With the new memory in place (a total of 4 Dimms) and ECC enabled both the Errors and the ECC Errs counters remain 0 :-)

 

Interleave: I noticed an increase in the memory bandwidth that memtest+ reports when I ran with 4 Dimms instead of just 2 (in dual channel 128-bit configuration). It seems therefore that the Opteron memory controller interleaves not only within the memory banks of a single module, but also among the modules in the same channel (bank as well as node interleave). This is what I was hoping for when I bought 2 sets of TWINX memory. Time for Corsair to offer QUADX memory ?

 

Memory Chips: I'm not sure on that one, but I tried a guess in this thread: Enthusiasts Only -> Corsair Ram and Memory Chips used. If you don't mind please let me know if I got it right.

 

Thanks

Thomas

[RAM GUY]

I will be moving your POST to the Enthusiast only section, since this will be inappropriate section.