Just upgraded an XP2100 to a 2800XP. My memory is a pair of 512MB Twinmos PC2700 cas2.5 sticks.
These have been perfectly working with my XP2100 using 333MHz setting.
However with the XP2800 I can only get a stable system by using the facilty to up the core voltage by 5% (Gigabyte 7vaxp ultra). This puts the core voltage upto 1.8V. Consequently temperatures get upto about 70C after several hours of Prime 95.
I have a Hiper 420W PSU, Arctic Cooling Copper Silent 2 fan, 4 case fans (2 in,2out).
Without raising the core voltage Prime 95 will only run for 2-5 minutes. I have tried most of the slower memory timing settings - Cas 3 even makes the PC difficult to boot (blue screen).
Stock vcore = 1.65v X 5% = .08 1.65 + .08 = 1.73v. A vcore or 1.8v will produce temps like that especially without a excelent cooler and I would not put to much faith in the temps reported by m/b sensors which are known to inaccurate by as much as 20%. The slower memory timings should make the boot process easier??? Have you tried to raise the memory voltage up to 2.7v?
( Quote ) Possession of a powerful computer is no substitute for a working brain. A working brain is no substitute for possession of a powerful computer.
The voltage is read by mbm as 1.7V when it is set to normal operation. Hence +5% takes it upto around 1.8V.
It seemed to be slightly better at normal voltage afterwards - do cpus bed in ?
I have adjusted the cooling (changed the fans for higher speed fans) and it's better again but it is still not reliable.
I wondered if it could be the PSU although the voltages are staying well within spec and the system until recently had a PCI Haupauge WinTV card and a Geforce 4200Ti (now a mx400) - these must surely have reduced the strain on the PSU but not the 12V line. The 2100 worked fine with these and 2800 only consumes about 4A more.
There are 5HD,DVD-R/W,CD-R/W,DVDROM, SCSI II card (nothing attached), Nothing on USB or firewire.
Following disconnecting 3HD and the optical drives it seems to manage to do 1 hour plus of prime95.
Core 0 1.68-1.71V Core 1 1.33-1.34V +3.3V 3.31-3.34V +5V 4.91-4.96V (this rises to about 0.05V when idling) +12V 12.40-12.59V (this drops 0.1V when idling)
Temps Peak at 42C core (off die) 36C somewhere on the motherboard.
Changed PSU to a 600W Magna unit (not great but it's all I could find).
Prime 95 will now run at normal voltage (1.7V according to MBM) for 1-2 hours before failing. Temp reaches 46C.
Tried +5% voltage setting (1.8V) and it remains stable temp reaches 55C.
I believe when cpus are made they are graded on their ability to operate at a particular speed. It is often possible to operate faster by overclocking the processor - the best overclocking is most likely to work if the core voltage is raised slightly. Is it possible that I have a slightly duff 2800 that should be a 2500 or less?
Any idea what your RAM temps are getting up to? - to the best of my knowledge there is no way to measure these...
However, check the airflow around your RAM, in particular where teh CPU heat sink dumps out on the RAM side.
I'm wondering if your RAM is overheating and then causing a failure that translates into a shutdown.
This is mostly conjecture of course, but I have seen and used the RAM heatsinks in the past, if you are really driving this puppy your RAM could be getting too hot.
Check for fine dust caked on the outside of the RAM and things such as the IDE cables blocking the air flow around the RAM
I don't know exactly how the Vcore for the CPU and the memory timing/voltages work together....
As to your conjecture about a slightly duff 2800.... I've worked in a cihp manufacturing line, and I can see no reason why you may in fact not be correct. The behaivyour certainly seems to back you up. Get the serial number off the CPU - it should be on a sticker on the fan itself if you have a boxed retail CPU/HS&F from AMD, if not then you will have to go to the chip itself. Then call AMD tech support and don't let them go until you find out what that chip really is.
You could also have them test it for you....requires sending it to them, if it's bad, they will possibly replace (don't recall what you said about in/out of warranty)