Hmm, you're not alone, seems like nforce4 are a bit buggy when dma is enabled. There's already a thread at the "linux kernel mailing list" (lkml.org) so post your MCE there ... (or you'll have wait ... and wait... and wait until someone hopefully fixes it)...

I have pretty much exactly the same problem... My machine is a:

AMD64 athlon 3200+, 939Dual-stat2 motherboard, geforce 6600gt pci-e, and a sata connected 200 GB HD..

The computer had been working for a few days with a dual boot windows XP/debian w/ kernel 2.6.14.2 and then it gave me an error:

CPU 0: Machine check exception: 4 bank 4 b20000000000070f0f
TSC 585602ff47
Kernel Panic - not synching: machine check

I did a full reinstall once but after a few days it gave me the same error!

I'm getting the same error on Fedora:
"CPU 0: Machine Check Exception: 4 Bank 4: b20000000000070F0F"
It used to work, even with the same kernel. Windows still works fine, no blue screens. If I disable SATA in the BIOS, it doesn't panic anymore. Wich makes me think it has something to do with that.
I reinstalled the kernel, and tried the previous version, but that didn't help.

Not that I'm happy to see I'm not alone, but it's nearly that... Curious that I find this page just wandering and suffering this problem since I upgraded to an Athlon X2 4200+ two weeks ago.

I'm still investigating what goes wrong (the new CPU is broken / the psu is not sufficient for that new CPU / the "old noname but dual channel" memory chips that were sufficient for the previous CPU shows at least their limits / the BIOS is buggy and do not configure the hardware properly).

mcelog tells me this is a Northbridge error. As the chipset is an ULi with integrated Northbridge and southbridge, I'm in trouble. However I managed to suppress the problem by reducing BIOS parameters on all fronts: CPU core speed at 195MHz instead 200MHz, RAM at 2*133MHz instead of 2*166MHz, CPU voltage at max 1.4V, and timings at max for memory. Now I'm still trying to restore some default values, I now that the CPU core, voltage and RAM speed are important for now (RAM timings seems to be irrelevant). RAM speed at 2*133MHz alone is not sufficient either. Of course, Memtest has not seen anything, and the temperature is indeed low (the machine crashes curiously at relatively low loads, and not at high ones!).

For the record, and as a quick and dirty FAQ for these errors, I give below my experience (I got this twice on two different machines, and it was different causes). So let's go:

A. General misbehaviour

If your machine randomly freeze (nothing alive in X11), reboots, either with or **without** any special activity, and if you have a recent processor (AMD or Intel), then you should consider yourself as a victime of the MCE (Machine Check Exception).

This is an exception raised by the processor to signal a hardware problem. Some can be recoverable, other not and crash the computer.

B. Diagnostic

To be sure:
1. Switch to text mode console (no framebuffer);
2. Reproduce the problem several times;
3. Look at the console output or in the result of the dmesg command.

If you get something like that:

HARDWARE ERROR
CPU 0: Machine Check Exception: 4 Bank 4: b200001000010C0F
TSC 18af428adee
Kernel panic - not syncing: Machine check

then you have a (serious) problem.

Technically, as the log says, this is indeed a hardware error. Not a software. But it can be triggered by any kind of usage of the computer. Let's be clear: you'll have hard time to find what is wrong, and you may eventually have to replace the whole hardware!

C. Getting details

There is a tool, mcelog, that can decode the exception code and give some clues. However, this is not always very usefull. For instance:

mcelog --ascii < crashlog.txt

Gives me:

HARDWARE ERROR. This is *NOT* a software problem!
Please contact your hardware vendor
CPU 0 4 northbridge TSC 18af428adee
Northbridge CRC error
link number = 1
bit57 = processor context corrupt
bit61 = error uncorrected
bus error 'local node observed, request didn't time out
generic error mem transaction
generic access, level generic'
STATUS b200001000010c0f MCGSTATUS 4
Kernel panic - not syncing: Machine check

This is better than nothing but finding the cause of the problem is still difficult. In my case, the Northbridge is integrated with the southbridge on my ULi chipset, so this can be virtually anything... Here it seems to be a memory error. But is it really the RAM, or the chipset?

D. Probable causes

Here is my top cause list. In general, this is not the CPU . This is:
1. Most probably, overclocking or high temperature in the case/near the RAM, CPU or chipset.
2. Very probably, the RAM chips or settings (mem testing tools may report no error but this is not sufficient, Linux/users can stress more memory than any tool).
3. Probably, the PSU that is insufficient (especially when the error occurs when accessing disks or after having added a hardware, or plugging a USB device).
4. Dometimes, cheap hardware on chipset/mobo.
5. Rarely, the BIOS wrongly configures something (my case since manually I managed to stabilize the box!).

E. How to fix it?

Iterate tries and tests!

1. Check the temperature of the CPU, chipset, RAM, the whole case. Do not overclock.
2. Check your memory with memtest. Try to lower memory banks speed, deactivate dual channel mode. Test with different RAM.
3. Remove some disks/PCI cards, unplug USB devices, use self powered USB device or a powered hub. Test with a different graphic card - not a gamer one. Test with a different (more powerfull) power supply unit.
4. Reduce speed of core components in the BIOS, increase voltage, increase delays of RAM/PCI components (I'm here ;-).
5. Test with a different mobo, a different CPU. If you arrive here, you are near to change the whole PC, and you should consider it to spare time, money, and have newer and better hardware. I'm not here :-o

Hope this could be usefull for anybody wondering what this mess means...
Best regards

C. Getting details

There is a tool, mcelog, that can decode the exception code and give some clues. However, this is not always very usefull. For instance:

mcelog --ascii < crashlog.txt

Gives me:

HARDWARE ERROR. This is *NOT* a software problem!
Please contact your hardware vendor
CPU 0 4 northbridge TSC 18af428adee
Northbridge CRC error
link number = 1
bit57 = processor context corrupt
bit61 = error uncorrected
bus error 'local node observed, request didn't time out
generic error mem transaction
generic access, level generic'
STATUS b200001000010c0f MCGSTATUS 4
Kernel panic - not syncing: Machine check

This is better than nothing but finding the cause of the problem is still difficult. In my case, the Northbridge is integrated with the southbridge on my ULi chipset, so this can be virtually anything... Here it seems to be a memory error. But is it really the RAM, or the chipset?

D. Probable causes

Here is my top cause list. In general, this is not the CPU . This is:
1. Most probably, overclocking or high temperature in the case/near the RAM, CPU or chipset.
2. Very probably, the RAM chips or settings (mem testing tools may report no error but this is not sufficient, Linux/users can stress more memory than any tool).
3. Probably, the PSU that is insufficient (especially when the error occurs when accessing disks or after having added a hardware, or plugging a USB device).
4. Dometimes, cheap hardware on chipset/mobo.
5. Rarely, the BIOS wrongly configures something (my case since manually I managed to stabilize the box!).

E. How to fix it?

Iterate tries and tests!

1. Check the temperature of the CPU, chipset, RAM, the whole case. Do not overclock.
2. Check your memory with memtest. Try to lower memory banks speed, deactivate dual channel mode. Test with different RAM.
3. Remove some disks/PCI cards, unplug USB devices, use self powered USB device or a powered hub. Test with a different graphic card - not a gamer one. Test with a different (more powerfull) power supply unit.
4. Reduce speed of core components in the BIOS, increase voltage, increase delays of RAM/PCI components (I'm here ;-).
5. Test with a different mobo, a different CPU. If you arrive here, you are near to change the whole PC, and you should consider it to spare time, money, and have newer and better hardware. I'm not here :-o

Hope this could be usefull for anybody wondering what this mess means...
Best regards

C. Getting details

There is a tool, mcelog, that can decode the exception code and give some clues. However, this is not always very usefull. For instance:

mcelog --ascii < crashlog.txt

Gives me:

HARDWARE ERROR. This is *NOT* a software problem!
Please contact your hardware vendor
CPU 0 4 northbridge TSC 18af428adee
Northbridge CRC error
link number = 1
bit57 = processor context corrupt
bit61 = error uncorrected
bus error 'local node observed, request didn't time out
generic error mem transaction
generic access, level generic'
STATUS b200001000010c0f MCGSTATUS 4
Kernel panic - not syncing: Machine check

This is better than nothing but finding the cause of the problem is still difficult. In my case, the Northbridge is integrated with the southbridge on my ULi chipset, so this can be virtually anything... Here it seems to be a memory error. But is it really the RAM, or the chipset?

D. Probable causes

Here is my top cause list. In general, this is not the CPU . This is:
1. Most probably, overclocking or high temperature in the case/near the RAM, CPU or chipset.
2. Very probably, the RAM chips or settings (mem testing tools may report no error but this is not sufficient, Linux/users can stress more memory than any tool).
3. Probably, the PSU that is insufficient (especially when the error occurs when accessing disks or after having added a hardware, or plugging a USB device).
4. Dometimes, cheap hardware on chipset/mobo.
5. Rarely, the BIOS wrongly configures something (my case since manually I managed to stabilize the box!).

E. How to fix it?

Iterate tries and tests!

1. Check the temperature of the CPU, chipset, RAM, the whole case. Do not overclock.
2. Check your memory with memtest. Try to lower memory banks speed, deactivate dual channel mode. Test with different RAM.
3. Remove some disks/PCI cards, unplug USB devices, use self powered USB device or a powered hub. Test with a different graphic card - not a gamer one. Test with a different (more powerfull) power supply unit.
4. Reduce speed of core components in the BIOS, increase voltage, increase delays of RAM/PCI components (I'm here ;-).
5. Test with a different mobo, a different CPU. If you arrive here, you are near to change the whole PC, and you should consider it to spare time, money, and have newer and better hardware. I'm not here :-o

Hope this could be usefull for anybody wondering what this mess means...
Best regards

This problem just appeared on my wife's PC (gentoo). I have some old RAM (nice in it's day), 1 stick of which has been suspect for a about a year. Recently I opened up the case to clean everything out and my thought is that I may have put my modules back in, in a different order than they were initially. Regardless, memtest86+ is currently running and I have quite a few errors being reported. This memory is not currently being overclocked but it had been in the past. When the test is done running I will swap the RAM modules and hope for the best but I think I will be buying some new RAM. Hopefully, if I play my cards right I can smooth talk the wife into an upgrade. Cheers!