New RAM == unstable

[tanderso]

I've previously posted this in the linux desktop forum, but could not find an answer. Please help if you can.

I've added 1G of RAM and now my system is unstable. I have a Soyo Dragon Plus! with three DIMM slots capable of supporting up to 3G according to the manual. I have three DIMMs equallying 1G -- 1 x 512, and 2 x 256. They are all PC2100. Any combination of two DIMMs in any combination of slots works without problem... but adding the third one causes memory errors. Therefore, I conclude that it is not a hardware problem. I've run docmem (a FreeDOS-based bootdisk memory checker), and it passes all three DIMMs with no problem. Therefore, I conclude that it is a linux problem. I've compiled the 2.4.20 kernel with high mem = 4G so that it recognizes more than 960M, and it does recognize it in dmesg and top. However, I still get segmetation faults with stack traces randomly in all software.

Has anyone installed more than 960M of memory and had memory errors as a result? Any idea what the problem is?
Sincerely,

Tom Anderson
CEO, Order amid Chaos, Inc.

http://www.oac-design.com

 

[marsd]

http://www.linuxhardware.org/features/01/11/30/1626230.shtml

There is a post on memory issues with 512 mb
DDR dimms in the feedback section.
Evidently flashing the bios with the newest bios upgrade is the cure.

Good Luck

 

[tanderso]

I'd like to believe that the BIOS could be the problem, however that doesn't make any sense. For one thing, BIOS detects the full 1G, and passes the boot tests. Secondly, the FreeDOS-based docmem reports zero errors in all of its memory tests. And thirdly, my current BIOS revision is 2BA2, and there are only two updates since then: 2BA3 addresses Athlon 2100+ support, and 2BA5 addresses AGP, Smartcard, and large HDD support. Also, the 512M DIMM works fine alone or in conjunction with one 256M DIMM, but not two. Therefore, I must be skeptical about updating the BIOS to solve this problem. I may do it as a last resort anyway, but I'd rather not compound problems by making unnecessary changes, as I've already done by recompiling the kernel.

What I would really love is a specific technical answer as to why this problem occurs in linux (but not FreeDOS), rather than hearsay and anecdote. Does anyone else run Mandrake or any other distribution with a Gig or more of memory?
Sincerely,

Tom Anderson
CEO, Order amid Chaos, Inc.

http://www.oac-design.com

 

[tanderso]

I find it really hard to believe that nobody in this forum is running linux with a gig or more of ram. If you are, please let me know if you've had any problems with stability or not.
Sincerely,

Tom Anderson
CEO, Order amid Chaos, Inc.

http://www.oac-design.com

 

[MrTom]

I run 1GB or more on plenty of servers, as i'm sure a lot of people do and there are no stability problems at all with the Linux kernel. You shouldn't even need the 4GB option as without it it should be OK up to 2GB but either way it's not going to make a difference.
I still think it sounds like a hardware problem that maybe is only apparent under high load - which is why docmem doesn't show anything. Try running with all the RAM but with as few things in the PCI slots as possible (in case PSU is overloaded), make sure the DIMMs are kept cool, disable any fancy BIOS options like low latency timings or memory interleaving, etc...
Have you tried running any other OS on it, like windows or another live linux distribution like knoppix?

 

[tanderso]

Well, with the Mandrake default kernel, it only recognized up to 960M. With the new kernel and the 4GB option, it now recognizes the full 1G, but it has the stability problem either way.

I tried underclocking the RAM to 100MHz and it reduced the number and frequency, but there were still random seg faults. Timings are specified by SPD, but I tried manually turning them all to the most conservative, and still the same problem either way.

I only have an SBLive in the PCI slots, and a GeForce2 in the AGP. I highly doubt it's a PSU problem since the system doesn't crash when I spin up my DVD drive which must draw much more power than a stick of RAM. Also, 350W is plenty for this system.

It's not only apparent under high load either, as Panel apps, Mozilla, and other programs crash randomly (if X even starts) while the system is idle.

The case has three fans, and the temperature is always relatively low. I've also now removed the side panel which reduced temperature even further. Still same problem. I really don't think it's a hardware issue.

I've only tried my Mandrake install and the docmem FreeDOS. Maybe I'll try one of those bootdisk RAMdisk distros and see if the kernel compiles without error. I'd love to hear any other ideas about what might be going on though.
Sincerely,

Tom Anderson
CEO, Order amid Chaos, Inc.

http://www.oac-design.com