Sick Rig - r640 with current GPU - debugging hints needed

[tmc]

I may be a bit offtopic here, but I hope someone can give me a few hints in the right direction.

 

first off: I`m a former server technician for Cray Servers - I know a few things, maybe a bit rusty here and there with intel based technology.

 

why it is "a bit off": It aint exactly overclocking the system (i didnt change the clock) - it is stretching the performance lifetime. That Rig is just too fast to throw away when you`re kids can play on it. I want to keep it running at it`s edge.

 

I have a Fujitsu Siemens R640 Workstation (X5000 Chipset, 2xXeon E5160@3.0 Ghz, 32 Gb ECC Ram, 840W Fujitsu PSU, more in the links below) which I have used with a Asus AMD7970 4GDDR5 OC2 in the past, a ASUS R9 390-X 6GDDR5 OC2 until last week and a (temporary) R9 Fury Nano White this week. The 7970 once killed a power supply, but downclocking it helped to keep it stable. The more modern cards have less power consumption but never got stable.

 

The system runs fine through Prime95 & memtest86 for days. furmark ? 3dmark ? both crash almost instantly to cold freeze. calculating windows 7 performance index crashes during buffer width setup for dx3d 10 to a cold freeze. I switched out CPUs, GPUs, PSUs. I updated firmware and drivers with a test matrix - current windows 7 patches, AMD drivers beta as of this week, checked RAM. symptoms stay the same - i do something in dx3d 10 -> cold crash.

the problem with that crash is the machine will not even power on after that until i completely drain power from the mainboard (pull plug from mains, press start for 10 sec minimum - anything less and ethernet stays down or similar ghost efffects. diags say "nothing", so it cant even save PCI errors. temperature at the whole time: below 50C degree CPU, below 54C degree GPU.

 

GPU-Z, CPU-Z, CoreTemp - all fine. during crash nothing looks out of the norm.

 

TL;DR: Any suggestions which PCI-E / PCI-X tools to use for debugging e.g. power flanks on the PCI or what i may have overlooked ?

people in "normal" forums seem to be very confused once you ask for PCI debuggers, oscilloscope readings, etc.

(thats why I ask here)

 

 

I`m starting to doubt my sanity / debugging skills here or the GPU manufacturer`s keeping of the specs.

 

(should this run stable, i would love any hints on how to get the CPU up about 10% - one multiplyer - if anyone still remembers these...)

 

 

Links to the Board:

http://tech.zeiss.net-base.de/data/MIRAX/MANUALS/FSC/Celsius_R640/Mainboard%20D1808/A26361-D1808-Z120-en.pdf

(it is a customized version of this one, so you cannot just use a "off the shelf powersupply" - still trying to figure out the wiring for a current ATX one - might slaughter the dead PSU from 2 years back)

TYAN - Download BIOS: Tempest i5000XT (S2696)

[tmc]

the link to the board was wrong - i accidentially posted the one to tyans firmware - the firmware is different from what fujitsu offers.

here is the correct link to the non-FujitsuSiemens version of the mainboard: TYAN – System Boards: Tempest i5000XT(S2696)