|12-12-06, 11:22 AM||#1|
Join Date: Aug 2003
(SOLVED) FX6200 instability problem
I have a Shuttle AB40 motherboard (upgraded BIOS).
This system has been rock solid with a FX5200.
After plugging in a brand new FX6200 I suddenly could not use AGP anymore (now using Option "NvAGP" "0"). I did a benchmark (GLOBS) with the FX5200 with and without AGP4x, and it gave me the following fps:
Test AGP4x legacy PCI 1 112 105 2 305 256 3 323 316 4 794932 794378 (this is a non-drawing 'fake' test) 5 313 307 6 181 161
another problem started to occur after I plugged in the FX6200. Even without AGP I have suffered from "random" lockups. Those lockups left no trail in any log: the kernel dies very hard and has no time to write anything about that to a log file. Everything is dead, as in - no bus activity anymore: The video card continuoes to display the last screen - the soundcard continuoes to repeat the last thing in it's buffer - the keyboard stops responding (capslock doesn't work, SysRq doesn't do anything).
Fortunately, I found another anomaly in my kernel logs (note: at least on debian, it seems that /var/log/kern.log contains more useful info than /var/log/messages): Ever since I plugged in the FX6200 I started to see dma_intr messages for both harddisks: An IDE bus(?) problem.
Now normally, if one gets harddisk problems, one would expect a hardware problem. But since the harddisks work fine with the FX5200 plugged in, not to mention the fact that I get errors on BOTH harddisks (that use the same cable by the way) one is inclined to rule out hardware problems of the harddisk(s).
Nevertheless, if it were software problems then one would expect the problems to be related to the video card and running applications that use that card. However, I was able to reproduce the harddisk errors most reliably as follows: booting a rescue disk and then (in low res VESA ascii console mode) running a badblocks test. This invariably caused disk errors and as a result thereof the disk resetting to a state without dma (at which point one sees the speed of testing blocks drop considerably). This indicates a hardware problem with the harddisk, NOT a software problem with the video card if you get my drift.
So, once accepting that I had a hardware problem: how is it possible that the communication between the harddisk and the cpu is influenced by a different videocard? It seems to me that it could be only one thing:
The FX6200 is emitting more (or different) high frequencies. The high frequency electro-magnetic fields emitted by the FX6200 are not good to a lesser motherboard.
Because I had especially problems with my harddisk. I changed the way the cables run inside the box (away from the video card(!)). After thus fixing my cables a bit, my instability problems seemed to have greatly reduced. I ran badblocks twice without problems (before it never got till the end before) WHILE inside X and running glxgears
Apparently I was correct that it was a 'pick up' on the IDE cable.
Nevertheless, during all that time I still did have ONE dma_intr error. So, I turned off the machine again and now really recabled the machine inside. Putting the cables as neatly and cleanly as possible and far away from the video card etc etc.
Result: The machine froze several times before the linux kernel was even loaded, and once it got to that point I had a TOTAL disk failure: superblocks could not be found, my encrypted partition wasn't even recognized as encrypted. Many, many errors during boot about non-existing files, disk errors an so on, and finally I was dropped to a repair shell - yeah, right.
I repaired it now: I took out the FX6200 and put the FX5200 back in (cables still the same). Everything works fine now.
Get my email address from my homepage.
Last edited by alaric; 12-13-06 at 10:14 PM. Reason: Problem is solved.
|12-13-06, 10:30 PM||#2|
Join Date: Aug 2003
(SOLVED) FX6200 instability problem (with partial solution)
I am using the FX6200 right now with AGP4x !!!
No instability whatsoever
The final solution came from IRC; I was telling about the problems I had, and someone remarked: maybe your power supply is running at it's maximum capacity. This seemed very unlikely to me (I have hardly anything in the box), but it made me remark a fact that I had noticed after exchanging my card so many times: some of the capacitors on my motherboard had some 'dirt' coming out of them - and most of them were bulging... When I said this (it had not appeared THAT important to me, after all: my system worked perfectly (with the FX5200)) others started to laugh and tell me that my motherboard was already dead, it just didn't know it. It would take other hardware with it, so I better remove the parts now I still had a chance. And I was refered to http://www.badcaps.net/ for the background story (juicy espionage involved ). And indeed... this motherboard was manufactored in 2001. After reading that site, I couldn't turn off my PC fast enough.
The next day I went to town and bought 12 new capacitors (the original ones were 1200 uF, 6.3V -- but those turned out to cost a fortune. I ordered a few capacitors that were 2200 uF, 16V instead -- ok, those are 10mm diameter instead of 8mm - but I had no problems at all in mounting them).
To make a long story short. I just replaced all capacitors on my motherboard and ALL instability is gone. I can now use the FX6200 like I could the FX5200.
What can we learn from this? Well, I'd say that the following important conclusion can be drawn as being a fact:
The FX6200 induces more 'noise' on the powerline than the FX5200. I'd think it would load it more - but both are fanless. It's 100% certain now however that the FX6200 can make a system unstable that is stable with a FX5200, solely because it demands more from the system. If anyone else runs into stability problems with an FX6200, while a FX5200 works fine for you: have a good look at your capacitors!
Now the bad news.
This probably means I'll leave this forum as a daily visitor
Thanks to those who helped, especially Lonni,
PS I might be back after I bought that new monitor
Get my email address from my homepage.