Aging OC XP-M Barton requires more voltage?

[trexpesto]

On my original OC I was Prime95 stable at 11x216 @ 1.65V in BIOS (1.632V reported by cpu-z).

Now to pass prime at same OC I bumped the VCore to 1.7 in the BIOS, varies between 1.664 and 1.680 in cpu-z while priming. Higher than I'd like. 1.675V in BIOS lasted about 6 hours of prime.
(Also cpu-z reports Core Speed varying between 2377.1 and 2377.4 MHz running prime. I am running the RAM intensive Prime torture test.)

UNCHANGED:
Mobo: NF7-S r2
CPU: AMD Mobile 35W 2400+ 11x216
HSF: SLK-947U w/slight trim + TT SilentCat "52.24 CFM, 21 dB"
Mem: 2x512 Mushkin L1 PC3500 1:1 @ 2.3.3.7
PSU: Enermax 460W, 120G Seagate SATA, NEC 2010A

CHANGES: The system did have some instability on its previous settings (EDIT: not at first, maybe 18 months later), but I attributed that to the fact that the HSF and VGA cooler were clogged with dust, and the NB fan was intermittent. Both of which I tried to address:

1. New case has no vents to the back side of Mobo. Shouldn't matter.
2. Took off the stock northbridge cooler and replaced with zalman passive sink. The stock POS frequently didn't spin anyhoo. I used AS5, gets warm but not hot.
3. Added a VGA silencer for my 9800 Pro. Not OC yet.
4. Took off two 80 mm case fans and got a TT 130mm rear exhaust on mobo header, and a Nexus 92mm off of PSU.

It did show some promise of stability at lower multi (10x) and previous VCore, didnt prime that long. Also haven't run memtest for too long. But the inescabable fact is that stability is now there with the increased VCore and old ram voltage. Even under load.

My questions are: Why is this happening? Will it get worse? Can I get back to my old OC at old VCore?

 

[DAPUNISHER]

Lot of variables to account for, but I'd start by taking that TT 130mm rear exhaust off the mobo header. Lots of other considerations, given the voltage I don't lean toward the CPU starting to degrade, but the IEEE types will have better input on that type of issue.

 

[trexpesto]

Thanks for the suggestion -sounds like common sense. Just tried it, Prime lasted 6 minutes, so I plugged in one of my old case fans off the PSU, and pointed it at the NB just to make sure that is not overheating without the breeze from the 130.
Repriming now. Edit >> with 1.65V BIOS.

Any other suggestions?

 

[trexpesto]

Still priming.. Whoa that 130 was a .3 amp draw according to it's label, still I had the idea somehow that increasing VCore would only affect the CPU !? Wish that 130 had a PSU plug on it, bastriches. Got my strippers out.

Edit: Whaddya know: from Svc.com

Connects a 3-pin fan to a 4-pin power supply lead without using up the lead. This connector is necessary when using high powered 3-pin fans on most motherboards, or when there are no more 3-pin taps available on the motherboard.

 

[trexpesto]

OK, maybe not: prime just failed after 1 hour and 9 minutes.
Edit and failed after ~ 6 hours with 1.675V

 

[DAPUNISHER]

Did you try the small FFT test? How about memtest?

 

[maluckey]

AS5 doesn't last forever. You shouldn't panic about this. First take off the HS and clean it. I ran mine through the dishwasher. Next reapply AS5 and burn-in to set the compound. You should also unplug that fan from the header. It can cause instability.

CHANGES: The system did have some instability on its previous settings (EDIT: not at first, maybe 18 months later), but I attributed that to the fact that the HSF and VGA cooler were clogged with dust, and the NB fan was intermittent. Both of which I tried to address:

You stated that you were not stable at these settings. What makes you think that a year and a half later it would be more stable?

Finally, your voltages look fine. CPU-Z and the onboard sensors are not relly useful except as comparison of before/after. Your current voltage mirrors mine to be stable at a similar speed with a similar CPU. The default voltage for these CPU is the same as the regular chips. The main difference is that these chips are speed binned and can idle at 35W.

As far as case cooling...a new case may flow differently. It may not cool in the exact same way as before.

 

[trexpesto]

Thanks for the responses fellows. The update:

I have unplugged the fan from the mobo header. I replaced it with one of the old case fans running off of PSU molex, just like it used to do with no problems.
I have successfully run Prime mem-intensive overnight at 1.7 again hmm. As above, 1.675V lasted about 6 hours.

I will try these:
* check my old case fan is running off of the optimal molex, not one used for anything else important.
* remove the new 92mm nexus and leave the case open.
* continue prime all day to see if the daytime ambient temps are a factor.
* run the small FFT test and memtest.
* replace the 9800 Pro with its new VGA Silencer with a stock Geforce4 mx440. That was the other major change. Has anyone heard of having to boost voltages (AGP??) when adding a VGA silencer? Whoa, it looks like it spins backwards? Counter-clockwise if seen from the bottom of the card looking at the fins. Makes for higher pressure? Do I have it plugged in correctly? Just tried it the other 3 possible ways, this is the only way it spins at all.

Maluckey, to clarify, the system was benched very stable at these settings, but ~ 18 months later, the dust got me. Cleaned it out and it was good again, didn't stress test it though.
I am considering the AS5 replace and reseat of the HSF except my CPU temps are not reported as all that high (AbitEQ, MMonitor), so I am holding off on that temporarily because it is a bolt-on SLK unit and everything is in the case with no mobo tray. Like an idiot, I didn't bench it outside the case first.

Thanks again for your ideas, Keep em Coming!

 

[maluckey]

The current draw from a 9800 pro with silencer is far and away higher than the old MX440. PSU voltage fluctuations as a result of this can make the +12v not stable enough, thus requiring slightly higher vCore.

 

[trexpesto]

Thanks for the info Maluckey. I had a 9800 Pro - stock - in there on my previous (1.5 year ago) stable OC, the MX is way older. I dont want to remove the silencer if I don't have to.

My Enermax is a monorail model 460W. I'll check it both at 1.7V and 1.65 with Sandra. Then switch out the vid card with the old MX440, check Sandra again at both Voltages while priming. If that is the problem, may consider wiring the silencer into molex after checking compatibility etc. I hate to run my whole MB at higher Volts just for that fan. I haven't even OC'd the 9800 yet.
------------------------------------
9800 @ 1.7 in BIOS, VGA Silencer on Low:
------------------------------------
Temperature Sensor(s)
Board Temperature : 26.0°C / 78.8°F
CPU Temperature : 45.5°C / 113.9°F
Voltage Sensor(s)
CPU Voltage : 1.65V
Aux Voltage : 2.66V
+3.3V Voltage : 3.34V
+5V Voltage : 5.11V
+12V Voltage : 12.28V
-12V Voltage : -12.86V
-5V Voltage : -5.45V
Standby Voltage : 5.04V
Battery Voltage : 3.63V

Result: Prime OK, 14 hour test

------------------------------------
9800 @ 1.65 in BIOS, VGA Silencer on High:
------------------------------------

Voltage Sensor(s)
CPU Voltage : 1.62-1.63V
Aux Voltage : 2.6 -2.66V
+3.3V Voltage : 3.34-3.36V
+5V Voltage : 5.11V
+12V Voltage : 12.28V
-12V Voltage : -12.86V
-5V Voltage : -5.45V
Standby Voltage : 5.04V
Battery Voltage : 3.63-3.65V

Result: Prime failed at 11 minutes
I believe the second sample was after Prime crashed. 12V looks solid. 3.3 and aux not so good.
I will run the MX440 tonight, have to go to work

 

[maluckey]

PSU voltages look good, if a bit high (modded or tuned?).

Your current voltages mirror mine. .

I find that Prime does not test well enough to be stable, and that huge maps of CIV III of CIV IV while running something like DivX in the background can find failure better and faster than Prime95. I ran 23 hours Prime95 without a hitch, and crashed in 10 minutes while gaming.

Newer video or chipset drivers can also be more or less stable and add stress to a setup. Some allowin more throughput when compared to older drivers. Th9is also adds stress.

The fact is that a so-called identical setup shares little in the way of OS kernel or drivers if you updated to current sets. It is hard to compare before and after when the setup isn't truly the same.

 

[trexpesto]

It's tough alright.. Prime failed after an hour with the MMX (no drivers installed) at 1.65 Voltages below. About the same. I suppose might as well run it at 1.7 to see if it lasts. First though, a round of small FFT Prime.

Hey Mal, if you don't mind, what are you running at for VCore, multi and FSB?
My PSU is totally stock.



------------------------------------
MMX440 @ 1.65:
------------------------------------

Voltage Sensor(s)
CPU Voltage : 1.62 - 1.63V
Aux Voltage : 2.66 - 2.64V
+3.3V Voltage : 3.33 - 3.36V
+5V Voltage : 5.11V
+12V Voltage : 12.22 - 12.28V
-12V Voltage : -12.86 - -12.77V
-5V Voltage : -5.45V
Standby Voltage : 5.04V
Battery Voltage : 3.65V

note: variations are while prime running
Result: Prime failed at 58 minutes

 

[CP5670]

1.7V should be no problem for XPs. My old XP 3200 had a stock voltage of 1.65V and these mobiles are handpicked versions of the same thing, as far as I know.

 

[maluckey]

Hey Mal, if you don't mind, what are you running at for VCore, multi and FSB?

As I am typing this I am at 12x x 200 and 1.712 vCore (1.75 in BIOS), 1.4v vdd, 2.7 volt vDIMM @ 2-2-211 in dual channel, interface enabled and nVidia IDE drivers (2.7) installed. I have CIV IV in the background at the moment. and am checking the web in-between turns.

I need 1.85 vCore in BIOS to run at 2.5 ghz (in BIOS at FSB 200) to stablize for gaming, and 2.0 vCore for 2.6 Ghz (in BIOS at FSB 200) for stability in gaming.

I can stabilize at 209 x 12 at 1.95 vCore in BIOS as well but need 2.9v vDIMM for stability.

The above setting have been the same since day one.

 

[SniperMerc]

Bump the CPU V to 1.75. and if you have an adjustment for the Northbridge voltage, give it a bump too.

I have a mobile 45 watt 2600+ @ 200x11.5 for 2500 mhz and been running it at 1.75 or 1.8 during the last two years (depending on what configuration ive been running) using a Zalman CNPS 7000 A-ALCU hsf with AS 5.

I do have some PC 4000 Corsair ram but faults would happen about once or twice a day if running anything over 200 mhz on the fsb so just backed it down to 200. Prime would run at 215 fsb for 24 hours but for whatever reason I was getting some weird system faults like games would just crash to desktop with no crash report, that sorta thing.

I suspect it's actually a problem with the northbridge needing some more juice but no adjustment for it onboard and I didn't feel like modding the board to do it.

 

[maluckey]

SniperMerc,

It's worth a shot, but some NF7-S v2.0 versions do NOT like the vdd to be above 1.5v. It depends on the board I guess. I can game stabilize this rig at 212 x 11 with 1.80 vCore in BIOS (with a 35W 2400+) without a hitch and with 1.4v VDD.

Sometimes a vCore pencil trick does wonders for stability. That's another worthy ten seconds spent. Sometimes it helps.

 

[SniperMerc]

I'm actually kinda suprised its taking that much voltage to stabilize it in the first place, the 35 watt chips were a bit better than the 45 watt like I have.

One thing to remember is that the stock fsb on those chips is only 133 so even at 200 mhz you still get a 67 mhz oc on the fsb.

At any rate, post yer results.

 

[trexpesto]

Ya, small FFT failed after 8.7 hours, ya. vdimm hmmm? Ima try memtest overnight @ 1.65. Loving your input.

 

[maluckey]

I'm actually kinda suprised its taking that much voltage to stabilize it in the first place, the 35 watt chips were a bit better than the 45 watt like I have

In reality, the 35w and the 45w rating is only for idle. At full tilt they are identical. These chips were binned to the lowest that they could idle at, not the average voltage at rated speed. That is the reason that the XP-M 2600+ was a 45w chip. It could be stable at default voltage, yet not able to idle at 35W.

Both of these CPU (35W and 45W) were easily able to run stable at 2.5 gHz without issues on air cooling.

I would also like to bring up the fact that sooooo many of the older claims made by "enthusiasts" about the 35W XP-M were snake oil or simple bragging. Many of the claims were nothing more than a way to boost the authors "E-penis" by posting screens at 2.7 Ghz and claiming that it was stable. There is even a large thread on Xtremesystems about faking benches for this purpose. Most of those claims could never run Super-Pi without crashing, and certainly never game with those settings.

In the end what you get from these fakers is so much bad information that it makes finding the truth difficult. Screeenshots and benchmarks are fine, but that is NOT stablility. Prime95 is NOT stability in itself, though if you fail Prime95 you are not stable.

 

[SniperMerc]

I was able to get into windows at 2.7 ghz with mine, at 2.0 Vcore but I would need to remove my 6800 GT to do it lol, also, I consistantly could run benches at 2.6 ghz but again without my 6800 GT in system, damn card pulls a lot of amps, but I definately wouldn't call it stable at 2.6.

BTW whats yer vdimm at?

 

[maluckey]

I can also post screenies at 2.7 gHz, but not for long!!! Crazy fast temperature rise!!

I run at 209 x 12 (2-2-2-11) @ 2.9 vDimm in BIOS. 200 x 12.5 @ 2.7 vDimm (2-2-2-11). Max stable is 2.9 vDimm @ 221 x 10.5 Mhz FSB (2-3-3-13). I need HUUUGE volumes of air to cool at that level.

2.6 gHz is doable (and game stable) with a 120 CFM fan on my CPU and an open case. My PSU is strong enough, but the voltage and heat is crazy for air cooling. 2.5 gHz is the highest gaming stable with a closed case, that I can do any gaming/work with.

 

[trexpesto]

Memtest ran all night OK at 1.65V; 9 hours 21 passes. Is that conclusive?
Still have the MMX440 in there.
The dimms are on stock voltage - 2.6 - since they are PC3500/433. 2-3-3-7
216-217 FSB is good for 1:1 with the memory, something I want to keep. Actually in BIOS it is 3:3.
Northbridge voltage is something to look at. It probably did overheat a few times due to the sticky fan on the stock cooler. I used to spin it up with my finger. The new sink doesn't get too hot with a case fan blowing on it at least.

1.7 VCore worked fine was my impression. I guess now I need to test it more.
Mal, point is well taken about OC "Proofs". I wonder about the repeatability of these tests.
That is, if you can pass Test A (Prime, memtest, etc.) in X hours, how likely is it that you will fail before that when re-running the test. For instance, small FFT test lasted 8 hours where ram-intensive test lasted 1 hour. Does that mean anything (RAM or Northbridge problem) or does it need repeats of each to really say? The tests run in cycles anyhow, so they are their own repeats, correct?

So, having the 1:8 ratio of ram-intensive:small-FFT Prime times, memory and NB are most suspicious.
Options to continue:
* Slack the memory timings and retest ram-intensive Prime.
* Increase vdimm to from 2.6 to 2.7V and retest
* reinsert the 9800 w/VGA Silencer and rerun memtest.
* move the fans around, plug a few more in and retest.
* put the VCore up to 1.7 and test - checking early assumptions and "looking under the lamppost"

What do you think?

 

[trexpesto]

I have a screen at 1.84 VCore 217.2 x 12 (cpu-z). It tested and played OK with case open, but put it back to 1.65, worried about longevity. Socket A is going to be a collectors item soon. 100$ Buy-Now price on ebay for 2500+ XP-M Barton.

 

[maluckey]

trexpesto,

First thing that I would consider is the RAM voltage. I run original BH-5 and am never stable at 2.6v vDIMM. What chips is your RAM using?

As far as Prime95...Small FFT is pretty much CPU only. Large FFT also stresses RAM. Video games are pretty much the best all-around tester. They also test AGP and PSU more thorougly due to the increased system/subsystem involvement.

If you have acces to CIV III or CIV IV then run the game at huge maps with maximum civilizations and see if you crash. If not then you are likely stable. These games are hard on overall system performace (esp. late in the game). I find that my Prime95 stable rig crashes after thirty minutes of CIV IV. Thats when I adjust accordingly.

Also, newer chipset and video drivers allow more throughput, alter settings and/or timings, thus stressing all the components a little more than before. That is one of the reasons that the nVidia IDE driver is not often used for high FSB speeds. The incresed throughput makes more stress on all of the components. I can only run the 2.7 IDE driver to 209 mHz FSB with any sort of guarantee of stability. For higher FSB I revert to the default Microsoft driver.

 

[trexpesto]

Maluckey, thanks for your replies, makes for better morale.

Prime failed @ 1 hr 9 min on Blend (memory-intensive) test with slack timings 2-4-4-9 as opposed to 2-3-3-7. Still default voltage. I'll increase VDimm and test incrementally, maybe later chipset voltage. Mine defaulted to 1.5V and I have never changed it yet. Also, I can retest at 1.7 VCore

Good question about the RAM chips, I still have the heat spreaders on, but I got them after the BH5 era was mostly over.

I am trying to keep my vid card out of the equation, as I know the system as a whole is not stable now, to try and keep the minimum number of variables. I know I have updated the chipset drivers and BIOS a while back, guess I can try and restore the old ones if I can figure out which. PS: I cleared the CMOS before doing most of this testing.