Mike Coan wrote:

I didn't mention, and I probably should have, that I first tried this with an ECS Kn1 Extreme motherboard (also uses the nforce4 chipset) and had even more problems with 4 GB, although again 2 GB was fine. thinking it wasa a motherboard problem I then got the Epox 9NPA, which had very good reviews.

While it may have good reviews, there's always the chance of a bad one getting through. Can you take the computer back to where you bought it, to show them the problem?

On Sunday 19 February 2006 19:53, Rikard Johnels wrote:

Even memtest sometimes fail detecting Tried using the 2's independent? Install om A and swap it for B and reinstall? (Didn't really understand if you had...) Are they the same types? Or is it mixed manufacturers/models? Some MB's are very particular on the flavor of the sticks

Hi Rikard, Mike and James... It is extremely unlikely that the memory is fine and that two completely different motherboards are exhibiting identical symptoms -- even if they're using the same chipset. One possible scenario: Mixing and matching sticks from different manufacturers on high end machines like this is asking for trouble. Even when the specs "match" one brand can test and run OK, then the other brand can test and run OK, but both installed can generate a disparate "blend" of noise in the memory bus that confuses the surrounding logic. Each stick has a distinctive noise 'signature' and you want those signatures match... to be synchronized with each other... as closely as possible.

If both install perfectly, it MIGHT be the motherboard itself.

Another possible scenario: If POST is only recognizing and initializing 2.8 GB when all 4 GB is installed, what are the odds that the results reported by memtest are just wrong? regards, Carl

-----Original Message----- From: Per Jessen [mailto:per@computer.org] Sent: 20 February 2006 11:41 AM To: suse-linux-e@suse.com Subject: Re: [SLE] Cannot install 9.3 or 10.0 with 4GB ram, 2GB fine Carl Hartung wrote:

If POST is only recognizing and initializing 2.8 GB when all 4 GB is installed, what are the odds that the results reported by memtest are just wrong?

...

Pretty good I would say - basically POST (Power On Self _Test_) is failing, which should say it all.

Probably you will need to flash your bios, what make of board, model and bios revision do you have currently? Chadley

Dear all who replied,

...

On Sunday 19 February 2006 19:53, Rikard Johnels wrote:

...

Even memtest sometimes fail detecting Tried using the 2's independent? Install om A and swap it for B and reinstall? (Didn't really understand if you had...) Are they the same types? Or is it mixed manufacturers/models? Some MB's are very particular on the flavor of the sticks

[snip]

It is extremely unlikely that the memory is fine and that two completely different motherboards are exhibiting identical symptoms -- even if they're using the same chipset.

One possible scenario:

Mixing and matching sticks from different manufacturers on high end machines like this is asking for trouble. Even when the specs "match" one brand can test and run OK, then the other brand can test and run OK, but both installed can generate a disparate "blend" of noise in the memory bus that confuses the surrounding logic. Each stick has a distinctive noise 'signature' and you want those signatures match... to be synchronized with each other... as closely as possible.

...

[snip]

Assuming that all of the ram has passed memtest (and it seems that it has), I think I'd look at the bios on the main board - I know in the past there have been situations where unusual hardware configs were just plain not tested for by bios manufacturers, although there was no real reason why they shouldn't work. Manufacturer and version of the rom? Is there a newer version that you can flash and try?

Thanks for all the suggestions, although I am not sure they will help. To clarify the memory situation. The memory is two sets of matched Dual Channel DDR from Rosewill. In other words, I purchased two sets of identical memory. Two sticks of 1 GB memory (2 GB) were packaged together and certified as Dual Channel DDR, with a lifetime warranty. I purchased two such packages for a total of 4 GB. Let's call the packages A and B. Package A passes Memtest whether it is in DIMM 1 &2, or DIMM 3 &4. Package B passes Memtest whether it is in DIMM 1 &2, or DIMM 3 &4. Since I couldn't install with 4 GB, I installed with package A in DIMM 1 & 2. I ahve not tried installing with just 2 GB in DIMM 3 &4, but it did pass Memtest in those slots. The motherboard, an Eox 9NPA Ultra, nforce4 chipset, came with a BIOS dated 6/29/05. The latest BIOS on the web site is dated 12/9/05, and I flashed the BIOS with that version. The motherboard thus has the latest BIOS available. The specs for the motherboard indicate that if all 4 DIMM slots are populated, then the memory must run at 166 Mz as opposed to 200 Mz. The board seems to recognize this as the POST says "128 bit Dual channel DDR at 166 Mz" When only two DIMMs are used the POST screen says "128 bit dual channel DDR at 200 Mz. I suppose I would like to know if anyone has successfully installed SuSE on a system with 4 GB and an Athlon 64. I see that the Opteron CPU is available now in Socket 939, not just Socket 940. Would that help? Is there a difference in the onboard memory controller between the Opetron Socket 939 and the Athlon 64 socket 939? Mike -- Michael A. Coan Woodlawn Foundation 524 North Avenue, Suite 203 New Rochelle, NY 10801-3410 Tel 914-632-3778 Fax 914-632-5502

On Monday 20 February 2006 10:35, Mike Coan wrote:

Thanks for all the suggestions, although I am not sure they will help.

At least we're trying!

To clarify the memory situation. The memory is two sets of matched Dual Channel DDR from Rosewill.

You haven't stated whether or not you bought the two sets of modules at the same time, but this really can be a factor, particularly on very high end machines that will be pushing the memory to the top of it's performance band. The 'noise signature' I mentioned in a previous post refers to the waveforms produced by the devices when they're operating in situ. The higher the chip quality and the more closely matched these (hopefully quiet) signatures are, the less likely you are to experience logic errors (including boot failures and random runtime errors) from noise on the memory bus. Rosewill isn't a manufacturer or even apparently a manufacturer's representative. This would make them a "reseller" or "distributor" which means they buy on the open market for resale. Since they're relatively new on the scene, even if the people behind it are seasoned professionals, they may not have the same manufacturer relationships and access to engineering resources that a more established company will have. I was in the memory business and can tell you from personal experience these resources are very important when you're dealing with high end machines. When the tolerances on a system are relatively loose and the memory isn't going to be stressed greatly, you can get by with lower quality. Otherwise, you're asking for trouble. If you have a "lifetime warranty" call or e-mail them and request an "advance replacement" exchange. Tell them all four modules must be identical, from the same manufacturer, production date *and* batch. Advise them that you need factory pretested modules built with name brand chips (Hitachi, Mitsubishi, NEC, Siemens or comparable.)

Package A passes Memtest whether it is in DIMM 1 &2, or DIMM 3 &4. Package B passes Memtest whether it is in DIMM 1 &2, or DIMM 3 &4.

We already know this. The problem is that the system won't accurately POST *or* install SUSE with both sets of modules in place. This is the specific problem that we're discussing.

The motherboard, an Eox 9NPA Ultra, nforce4 chipset, came with a BIOS dated 6/29/05. The latest BIOS on the web site is dated 12/9/05, and I flashed the BIOS with that version. The motherboard thus has the latest BIOS available.

As a consumer, you can't do better than that. But it might be worth opening a 'trouble ticket' with the board manufacurer describing this problem and see if they're recommendation is. I know... very difficult... but worth trying by e-mail anyway.

The specs for the motherboard indicate that if all 4 DIMM slots are populated, then the memory must run at 166 Mz as opposed to 200 Mz. The board seems to recognize this as the POST says "128 bit Dual channel DDR at 166 Mz" When only two DIMMs are used the POST screen says "128 bit dual channel DDR at 200 Mz.

The problem is POST is also telling you "2.8 GB" is installed... garbage in, garbage out. When I see an error as fundamental as that I treat *all* feedback from that point forward as suspect. Have you checked to see if there is a manual override that would allow you to force the memory bus to run at 166 Mhz? Have you tried any of the usual performance adjustments to slow things down a bit and see if the memory can cope?

I suppose I would like to know if anyone has successfully installed SuSE on a system with 4 GB and an Athlon 64. I see that the Opteron CPU is available now in Socket 939, not just Socket 940. Would that help? Is there a difference in the onboard memory controller between the Opetron Socket 939 and the Athlon 64 socket 939?

I think these considerations are premature. You haven't ruled out the quality of the memory yet. You've already tried a different board, too. Why not stay focused on the most likely culprit? regards, Carl

On Monday 20 February 2006 16:35, Mike Coan wrote:

Dear all who replied,

...

...

On Sunday 19 February 2006 19:53, Rikard Johnels wrote:

...

Even memtest sometimes fail detecting Tried using the 2's independent? Install om A and swap it for B and reinstall? (Didn't really understand if you had...) Are they the same types? Or is it mixed manufacturers/models? Some MB's are very particular on the flavor of the sticks

[snip]

It is extremely unlikely that the memory is fine and that two completely different motherboards are exhibiting identical symptoms -- even if they're using the same chipset.

One possible scenario:

Mixing and matching sticks from different manufacturers on high end machines like this is asking for trouble. Even when the specs "match" one brand can test and run OK, then the other brand can test and run OK, but both installed can generate a disparate "blend" of noise in the memory bus that confuses the surrounding logic. Each stick has a distinctive noise 'signature' and you want those signatures match... to be synchronized with each other... as closely as possible.

...

[snip]

Assuming that all of the ram has passed memtest (and it seems that it has), I think I'd look at the bios on the main board - I know in the past there have been situations where unusual hardware configs were just plain not tested for by bios manufacturers, although there was no real reason why they shouldn't work. Manufacturer and version of the rom? Is there a newer version that you can flash and try?

Thanks for all the suggestions, although I am not sure they will help.

To clarify the memory situation. The memory is two sets of matched Dual Channel DDR from Rosewill. In other words, I purchased two sets of identical memory. Two sticks of 1 GB memory (2 GB) were packaged together and certified as Dual Channel DDR, with a lifetime warranty. I purchased two such packages for a total of 4 GB. Let's call the packages A and B.

Package A passes Memtest whether it is in DIMM 1 &2, or DIMM 3 &4. Package B passes Memtest whether it is in DIMM 1 &2, or DIMM 3 &4. Since I couldn't install with 4 GB, I installed with package A in DIMM 1 & 2. I ahve not tried installing with just 2 GB in DIMM 3 &4, but it did pass Memtest in those slots.

The motherboard, an Eox 9NPA Ultra, nforce4 chipset, came with a BIOS dated 6/29/05. The latest BIOS on the web site is dated 12/9/05, and I flashed the BIOS with that version. The motherboard thus has the latest BIOS available.

The specs for the motherboard indicate that if all 4 DIMM slots are populated, then the memory must run at 166 Mz as opposed to 200 Mz. The board seems to recognize this as the POST says "128 bit Dual channel DDR at 166 Mz" When only two DIMMs are used the POST screen says "128 bit dual channel DDR at 200 Mz.

This is most probably your problem! The bus runs too slow when you populate all four slots. The specs for the RAMS states 200 MHz and is only certified for that speed. Trying to run them at anything less most yields possible timeout problems. See if you cant test the sticks on somebody's system that DOES run at 200 MHz at all times. Or see if the manufacturer of your motherboard has a "work around" for that timing problem. Or get a new set of RAM... --          /Rikard ----------------------------------------------------------------------------- email   : rikard.j@rikjoh.com web     : http://www.rikjoh.com mob     : +46 (0)736 19 76 25 ------------------------ Public PGP fingerprint ---------------------------- < 15 28 DF 78 67 98 B2 16 1F D3 FD C5 59 D4 B6 78  46 1C EE 56 >

On Monday 20 February 2006 14:11, Carl Hartung wrote:

What I'm suggesting is:

module A1 = slot 0:0 module A2 = slot 1:0 module B1 = slot 0:1 module B2 = slot 1:1.

Addendum: I got curious and looked at Intel's DDR architecture whitepaper and also threw some queries at Google. Here's a very interesting summary and I think it can be trusted as the info originates from Crucial's engineering dept.: http://www.buildorbuy.org/dualchannelddr.html "Dual Channel memory configurations will provide increased performance over Single Channel configurations. Note: Memory channel speed is determined by the slowest DIMM module populated in the system. Rules to Enable Dual Channel Mode To achieve Dual Channel mode, the following conditions must be met: 1. Matched DIMM configuration in each channel 2. Same Density (128MB, 256MB, 512MB, etc.) 3. Same DRAM technology (128Mb, 256Mb, or 512Mb) 4. Same DRAM bus width (x8 or x16) 5. All either single-sided or dual-sided 6. Matched in both Channel A and Channel B memory channels 7. Populate symmetrical memory slots (Slot 0 or Slot 1) Note: Configurations that do not match the above conditions will revert to Single Channel mode. The following conditions do not need to be met: 1. Same brand 2. Same timing specifications 3. Same DDR speed Carl

On Mon, 2006-02-20 at 10:35 -0500, Mike Coan wrote:

The specs for the motherboard indicate that if all 4 DIMM slots are populated, then the memory must run at 166 Mz as opposed to 200 Mz. The board seems to recognize this as the POST says "128 bit Dual channel DDR at 166 Mz" When only two DIMMs are used the POST screen says "128 bit dual channel DDR at 200 Mz.

Thow things jumped out at me: 1. "128 bit Dual channel DDR at 166 Mz" Last time I checked, and correct me if I'm wrong, Athlon64 CPUs don't have dual channel controllers, only the Opterons do. So based on this, I guess the nForce4 has its own dual channel controller that it uses instead of the on-CPU one? This leads me to the next thing: 2. The board detects different speeds based on different modules. This almost always means the memory controller is either faulty or just poorly implemented. This happens on the best of boards. I purchased a top-of-the-range ASUS board for my AthlonXP, A7V880 (cost almost twice as much as the Gigabyte equivalent) and I had this exact problem. It misdetected my CPU and memory. At first I thought the memory was faulty, but after trying it on another board, I had the ASUS board swapped out. The new one had exactly the same problem. Evenutually I got the Gigabyte equivalent, and with the same memory, it's running 100% Another I noticed in the above situation, is on the ASUS board my performance seemed to increas if I drop dual-channel mode and lowered the memory speed - 100mhz single channel was the best. I would suggest you try disable dual channel mode and see if that makes a difference. Lastly, just a thought, are you installing the 32bit or 64bit version? I don't know the details, but as far as I know 32bit machines can't handle (or can handle but cannot properly use) all of 4GB of Memory. Anyone know more about this? Hans

On Tue, 2006-02-21 at 08:46 -0500, Adam Tauno Williams wrote:

On Tue, 2006-02-21 at 15:06 +0200, Hans du Plooy wrote:

...

Lastly, just a thought, are you installing the 32bit or 64bit version? I don't know the details, but as far as I know 32bit machines can't handle (or can handle but cannot properly use) all of 4GB of Memory. Anyone know more about this?

There are lots of 32bit machines with far more than 2Gb or RAM, it works (and uses it) just fine. You can get IBM xSeries Xeon machines with 16Gb of RAM or higher.

Oh, I thought 32bit intel systems couldn't address that much memory. Either ways, I stand corrected :-) Hans

On Tuesday 21 February 2006 11:41, Mike Coan wrote:

Epox tech support indicates they have 4 GB installed running Windows XP and 2000 using Corsair RAM and also Kingston RAM. I would hate to buy 4 GB of Ram in the hopes that it fixed the problem. it is actually cheaper to buy on Opteron single core processor.

With all due respect, Mike, I gave up contributing to this thread because we were starting to run in circles. You simply refuse to consider or discuss going back to Rosewill and telling them the memory they sold you is of poor quality. That memory is *not* functioning correctly. I looked at the manual for your mainboard. It should automatically detect and run with that 4GB exactly as you've installed it... multiple times, already. Instead, you install the four sticks and it: - "sees" and reports only 2.8 GB during POST - refuses to boot the SUSE installer. --> And this is the *second* mainboard to choke on that memory.<-- The solution is as clear as the sun on a bright summer day. Get some decent RAM for that "Ultra" high end system. Your problem will go away, you can get on with your life and new projects. Of course, "IMHO," "YMMV" and all the other usual caveats apply. ;-) Good luck & regards, Carl

On Tue, 2006-02-21 at 15:23 -0500, Mike Coan wrote:

"Yes it will support it as well. We have tested with Corsair XMS ans dual package they offer. Although running in all memory slots with ddr 3200 Nvidia chipset have some issue with it running all memory banks with 1Gb modules."

Even Epox admits there are problems with all 4 memory bank using 1GB modules and the nvidia chipset. The other motherboard was the same nvidia chipset.

Maybe try two 2GB dimms? Hans

Mike Coan wrote:

I realize that POST indicating 2.8 GB of RAM is a problem, but I am not sure why it is a problem with the RAM and not a problem with the motherboard or the chipset. Both motherboards use the exact same chipset.

You're right - it could be one or all of those. The question is - how do you diagnose it and eliminate all but one? First of all, I would not even consider the chipset. You might as well suspect poor tracing on the PCB. You can't diagnose that. Only once you've excluded the motherboard and the memory, can you perhaps point the finger at the chipset. Or the tracing. That leaves motherboard or memory - you've already tried swapping the motherboard, right? When that memory config fails on two roughly identical motherboards, chances are it's the memory.

The BIOS recognizes all 4 GB.

This is most probably only an indication that the serial EEPROMs are working.

Memtest recognizes all 4 GB. Memtest runs for 13 hours with 4 GB installed and gives no errors.

Is this the most recent version of memtest86+? When POST fails, it does seem odd that memtest does not. Maybe memtest86+ isn't exercising the memory to the full extent.

Even Epox admits there are problems with all 4 memory bank using 1GB modules and the nvidia chipset. The other motherboard was the same nvidia chipset.

Wait - the manufacturer says "it might not work"? This is when you switch to a different manufacturer. I.e. you send the board back and get a refund - if of course the board specs clearly say the board WILL work with 4 x 1Gb of any decent make. /Per Jessen, Zürich -- http://www.spamchek.com/ - managed anti-spam and anti-virus solution. Let us analyse your spam- and virus-threat - up to 2 months for free.

On Sunday 19 February 2006 21:10, Carl Hartung wrote:

Hi Rikard, Mike and James...

It is extremely unlikely that the memory is fine and that two completely different motherboards are exhibiting identical symptoms -- even if they're using the same chipset.

I hate to interject this but the first thing I thought of when the OP said he couldn't install with 4GB was that 4GB requires an option be used in the kernel for 'large memory'. Could be the install kernel wasn't compiled for 4GB (unlikely but possible) or that there is some bug in the code for 4GB. (in the install kernel) Just my $.02 But I agree that memory is the first thing to think about.

Carl Hartung wrote:

...

Could be the install kernel wasn't compiled for 4GB (unlikely but possible) or that there is some bug in the code for 4GB. (in the install kernel)

It would be nice to have a method to test this. Any ideas?

/usr/src/linux/scripts/extract-ikconfig will extract the .config from a [b]zImage file. The config option is CONFIG_HIGHMEM4G or something like that. /Per Jessen, Zürich