Is my system stable?

[cimurray]

I built my new rig about 6 months ago and other than giving it an initial thrashing up to 4GHz OC on air when new I run everything at stock settings. I use Flight Simulator X a lot and after an hour or two flying the system locks solid. Everything else is fine - spreadsheets, word processors, windows media player, itunes etc. I tried re-installing all of the drivers but that made no difference. So I did a complete re-format and reinstall of XP taking it all the way back to SP3. Tried FSX which ran for a few hours then locked solid.

Wondering if the problem was hardware I ran the memory for 12hrs on Memtest86+ with no problems. But when I run prime95 on both cores it will fall over after four or five hours. I don't like this at all - but am I being too fussy or do I have a problem? And if there is a problem any ideas on how to fix it? (Temps - idles at 51C but maxes out at 63/64C according to Asus probe)
Specs:
Asus P5E3
E8400
8800 GTS 512
OCZ 2GB DDR3 PC3-10666c9 Gold
Tuniq Tower 120
Vcore - auto

[CrowdAds]

[ctal]

Those temps sound quite high for an E8400 running at stock speed and voltage. I'd first re-measure the temperatures using Realtemp, which figures to be more accurate than other monitoring programs for the 45nm (and 65nm) Intel chips, as you'll gather by reading this:

Real Temp - Documentation

If the airflow through your case is good, and your room temperature is typical, if Realtemp indicates much above 30 degC at idle I would clean off and reapply the thermal paste under your heatsink. And I would follow these directions for applying it, even if you use something other than Arctic Silver:

http://www.arcticsilver.com/pdf/appi..._dual_wcap.pdf

Regards,

-- Al

[cimurray]

I've done some reading and some testing as suggested. Real Temp 2.11 shows both cores idling at 39C compared to Asus Probe showing 50C. In addition RealTemp shows TjMax = 95C and the score for both cores on the "test sensors" is +4 which seems to indicate they are both working and reading correctly.
OK, so now I know a bit more about the true temp on my cores - how do I stop the crashes under full load after four or five hours or is this considered to be normal?

[ctal]

That still sounds too high to me. I have an E8400 that idles in the low 30's while overclocked to 3.8gHz and with Vcore increased to 1.35V. I also have an E8500 overclocked to 3.8gHz, with stock Vcore, that idles at 24degC, both per Realtemp.

These cpu's are widely regarded as very cool running.

I'd suggest re-doing the thermal paste, especially if you did not apply it previously per the technique shown in the Arctic Silver paper I linked to.

Regards,
-- Al

[cimurray]

Update - Having thought some more about my problem I slowed my processor down by lowering the multiple from nine to six. In other words it is now running at 2GHz rather than 3GHz. I have just completed a sixteen hour run of Prime95 without incident whereas before it would crash between four and five hours. The only major difference I have been able to notice is that my system is cooler than before - Real Temp has both cores at 49C under full load which is 4 or 5 degrees lower than before and seems to be sufficient to make the difference.
My working theory is that I need a new case. I built a brand new machine at the beginning of this year but I kept my old case which is a Lian Li 65 that used to house a P4C800 Deluxe with a 2.8 P4. The case is pretty full with my current set up and I had to shoehorn the CPU cooler (Tuniq 120) into the box and it just fits in no more. I don't think there is sufficient ventilation or cooling going on so I know what I want from Santa this year - a bigger box with bigger fans and greater airflow. So I'll just run the processor at this lower speed until then.

[wonderwrench]

Originally Posted by ctal

That still sounds too high to me. I have an E8400 that idles in the low 30's while overclocked to 3.8gHz and with Vcore increased to 1.35V. I also have an E8500 overclocked to 3.8gHz, with stock Vcore, that idles at 24degC, both per Realtemp.

These cpu's are widely regarded as very cool running.

I'd suggest re-doing the thermal paste, especially if you did not apply it previously per the technique shown in the Arctic Silver paper I linked to.

Regards,
-- Al

I can see the low 30's but not 24C. Whats your real case temp measured with a accurate thermometer? I'd guess higher than 24C. Maybe your room temp is 15C?

[ctal]

Originally Posted by wonderwrench

I can see the low 30's but not 24C. Whats your real case temp measured with a accurate thermometer? I'd guess higher than 24C. Maybe your room temp is 15C?

Room temp is 21C per a reasonably good household thermometer. I didn't open up the case, but placing the same thermometer directly over a top blowhole fan gives a reading of 23C, and placing it next to a rear exhaust fan gives 24C.

It is a very large Lian Li PC-A70B server case, containing seven 120 mm case fans. The cpu is cooled by a large OCZ Vendetta 2 hsf, which includes a 120 mm fan.

Perhaps the 24C cpu reading is a bit inaccurate, but I suspect not by very much.

Regards,
-- Al

[zapionics<alt>]

It cant be 24c with an ambient of 21c... minimum rise over ambient would be at least 15c at idle IMO...in the best possible case... what does the Intel spec say about that?

[ctal]

Originally Posted by zapionics<alt>

It cant be 24c with an ambient of 21c... minimum rise over ambient would be at least 15c at idle IMO...in the best possible case... what does the Intel spec say about that?

There is no clear answer to that. Please read the Realtemp writeup that I referenced in post 2. Among many other things, it explains that the thermal sensor in the 45nm (and 65nm) cpu's were designed to be most accurate at very high temperatures, to accurately trigger protective mechanisms, and are not particularly accurate at low temperatures.

Realtemp is designed, as explained in the writeup, based on actual measurements taken on an E8400 with an infrared thermometer. Other programs are not, and instead are based on assumptions/guesses as to a value of Tjmax that Intel has not documented (they are supposed to document it for the next stepping, which will be released in coming months).

Realtemp also recommends a calibration procedure that can further improve its accuracy. I didn't perform that on my E8500 system, with the 24C number (at stock Vcore and with a conservative overclock), but I did perform the calibration procedure on my E8400 system (that has a significant Vcore boost and a more aggressive overclock). Following that calibration procedure, I measured numbers in the low 30's at idle; I would expect the E8500 to be significantly below that due to the lower Vcore and lesser overclock that I am running it at.

I suspect that your 15C best case projection of Tcore vs. Tambient is based either on experience with earlier cpus that ran hotter, or were not cooled as well as in my case, or, if they were E8xxx chips, they were measured with different programs that were presumably less accurate for the reasons stated.

Regards,
-- Al

Edit: Keep in mind also that, at idle, SpeedStep is lowering the cpu clock speed by a factor of 1/3 for the E8400, and by a bit more than that for the E8500. The desktop processors of a few years ago did not have that feature, which is one more reason why they would have run hotter at idle.

[cimurray]

Two months after my initial posting I think I've got it licked! I upped the voltage on my memory from 1.7V to 1.8V and it hasn't missed a beat since - not one lockup. As 1.7V is the recommended voltage for my memory I can only assume my motherboard has been slightly undervolting from new and causing the occasional lock up. Oddly enough it only ever happened in Flight Simulator X.

[PeterT]

FSX is very demanding and memory hungry, so I'm not surprised that it found the weak spot. Glad you got it licked.