So I lost a couple of folding boxes last week. After convulsing on the floor in the fetal position for a day or two, took stock of what I had and what I could get to stop the hemorrhage of points per day. Here's what I've got (I hate epenis lists):

GA-P35-DS3L (rev 2.0)
e2160 (L2 stepping) at 3.2 GHz (400x8)
2x2 GB G-skill DDR2 800 on manual timing at 5-5-5-15
The Antec 380 power supply that came with an NSK 4400
Minja with the included 80mm fan
Ati 3850 w/S1 and a zip-tied 120mm
2 120mm's (one exhaust, one intake)

I transferred an e2160 to a new Gigabyte P35-DS3L with the goal to OC it and fold, fold, fold. Over 24 hours prime stable at 3.2 GHz (400x8). Finished the first WU running checkpoints just over 15 minutes per (what an improvement over the previous VIA board's 25 minutes!). Sometime in the middle of the second WU, windows crashed on it's own. OK, I just upped the core voltage a bit (1.36 idle via CPU-Z, Vdroop down to 1.344 or less often 1.328). Temps were circa 52-53ish on speedfan (68-70 via coretemp) with a Minja at full blast and whole bunch of 120mm's. A few hours into the session the desktop freezes. Restart, and install all those windows updates that I was being bugged about. I started F@H right back up, went to bed, and this morning checked to find that XP took another dive (back at login screen) Went from 20% to 40% on a WU, so probably around 5 hours before the restart.

Yeah, I know that F@H is really sensitive to OC's. What I would like is y'all's thoughts on what might be going on, and maybe some suggestions on what to do next. As I see it, there are several possible explanations:

1) OC isn't stable at 400 FSB, despite over 24 hrs of Orthos. Give it more voltage and retest, with Orthos, WPrime, or dual Prime95 instances. If not stable, then try for 355 x 9 or the like, or back it off to 3.0 GHz.

2) Windows update is to blame, turn it off completely.

3) Temps are to blame, something is overheating. (Orthos has higher load temps compared to F@H, but I dunno).

4) An Antec 380 watt PSU just isn't cutting it. Get a newer, bigger one.

Any thoughts or suggestions would be greatly welcome, not just for me but also for team 31574.

Thanks in advance, folks.

--KK

It shouldn't be the power supply. Your system isn't using close to 380W, I guarantee. Sometimes, with overclocks, you are just trying a little bit too hard. Try stepping it down to 3.0GHz. You wont' lose that much speed, but it just might be the sweet spot. You can also crank down your voltage a little, which saves power and heat.

You might try checking the NB and the memory to see how hot they are getting. You might also try putting the RAM timings on auto. Sorry, that's all I've got.

You left Windows Update on auto? Oh yuck that's terrible. Turn the damn thing and do occasional maintenance. See if that clears up your problem. I'd bet money the problem is here. On my Vista PC it won't even start while OC'd. I do occasional maintenance on my own. I reboot every 2 weeks, go back down in speed, install all the updates, and then return to my running configuration. Anti-virus was not too happy about OCing either.

You can always run Prime 95 more than 24 hours.

I think the EA500 was available with a coupon code for a measly $35. I'd buy a bunch of them if I was going to have a farm.

Total wattage is no indication of adequate clean power. Look at the NZXT 1000 watt power supply on jonnyguru.com. Its 12volt rails are a disaster. Then look at the EA500, which has great 12Volt rails. Bad 3.3 volt rails
will make memory look unstable. You didn't mention your voltages. You also didn't mention your NB temps. Can you hold and maintain a grip on your NB while it's running? If not it's probably around 60C or more.

Now we see what 2MB cache buys you as opposed to 1MB of cache on a C2D. My E6400 is running the same as you, but my Patriot Memory is running 4-4-4-12. Those 2 differences are worth maybe a little more than a minute (just under 14 minutes per checkpoint). The E8400 is supposedly 10% better than it's 65nm brother(sister).

But also - your asking for a 100% OC, which is asking a lot. You would find that a FSB of 333, setting the memory for DDR2-667 and 4-4-4-12 would probably result in a small performance loss. But I would still bet Update is what's messing you up.

Thanks, y'all. I'll check the heatsink on the NB this evening, and then update with my voltages for the CPU, memory, and the NB. I am running the ram at 1:1, but it couldn't pass memtest at a 400 fsb and 1:1 unless I upped the volts on the NB to compensate. I never did do the "hand test" to tell, though in retrospect I probably should have. I'll post more tonight.

--KK

Have you reseated the NB heat sink with Artic Silver or are you still using the factory goop?

I got home late (hate the freakin' hospital). Surprisingly, the box hadn't crashed. Still folding strong from this morning's restart.

Looked around at work and discovered that speedfan 4.34 beta has better detection of the temps in the GA-P35-DS3L. Out with the old, in with the new. Compared to coretemp, the temps are much more similar than with speedfan 4.33. Core 0 is at 61-64 degrees, and core 1 is at 65-68 degrees under load. With a voltage increase to the core, on load it is now 1.36 V withour much variance. I haven't seen it dip into the 1.344 range sitting here tonight.

I took off the side of the case and watched speedfan graph a 5 degree decrease in temps. Touched the nb heatsink and burned my finger. That sucker is hot as blazes.

Right now, on 2653, the cpu cores are 63 and 66 degrees. How hot is too hot for the NB, and how can I figure out (without buying a digital thermometer) how hot it is? If it's roasting, I might invest in a zm-nb47f or the like. I haven't resetted the factory goop. It honestly didn't cross my mind. The recent GB mobos have had reports of good thermal goop applications under the NB heatsink. Anybody think it will really make a difference?

Voltages in speedfan are a crapshoot. The 3.3 rail is showing 3.23V, the 5 rail is 4.92, and the 12 rails register 4.29 and -16.97. That can't be right. Any programs that can show under load what the voltages are? The bios can do it, but then it wouldn't be under load.

Any ideas if 3.2GHz at 355x9 would run cooler than 400x8? If nobody has any ideas I'll probably try it next week and see.

I completely disabled windows updates, and I'll humbly take the advice of others with more experience and just download them on occasion from now on. I had it set to download and notify me, but not install. Considering that there were 92 updates installed last night, maybe that had something to do with it. I dunno.

The northbridge should be a little cooler if nothing else, or less hot is probably a better word.

The droop is normal. You want droop. No droop is bad. But 1.36 volts doesn't sound excessive.


Wow, those seem awfully high, I'm doing 400*8 and FAH is in the low 50's. Prime 95 rarely reached 60C. Once at 60C for a little while Arctic Silver seems to have cooked a bit and my temps came down. When I first did 400*8 FAH hovered in the high 50's. But you're above spec in any case.


Big bad red flags all around. My temps don't change with the cover on or off, except a little on the southbridge which sees less air movement. Clearly you need to:

1. Keep the cover off until you improve internal air circulation.
2. Improve internal air circulation.
3. You need to get the NB temps down. Mount a fan blowing straight at it somehow.


I'm note sure that will be enough. Sounds like yours is running 70C or so. Yes, you can do that, but you're cooking everything around them as well. Perhaps space/height is an issue, it so try this one but it still needs a fan.
http://www.sidewindercomputers.com/encnfoconoco.html
If not you can't go wrong with this one at this price. And still, you need a fan.
http://www.sidewindercomputers.com/thhrchco.html
But you should improve air flow through the case first, else you're just speeding up hot air.

My NB HS is like yours, the center fin has a "T" top which is a good place to mount more small HSs which I already had. If you have any I would add then ASAP. I stuck 2 copper one on the "T" and temps dropped like 5C immediately. I used the Switftech version of these, but these are cheaper right now.
http://www.sidewindercomputers.com/enbmfocobgar.html
This option allowed for better cooling with zero down time, I kept folding as I stuck them on, though I did lay the tower on its side for 24 hours until the thermal adhesive set up.


Floffe is right, but not by much. Because your cpu temps are too high I would suggest you run 334*9 for now. Forget about 3.2GHz until cooling improves. Run memory at FSB:DRAM 1:1. That means at stock speeds your FSB is 266MHz (1066 quad pumped marketing crap). Set memory for DDR2-533 (266*2). Running synchronous is good. I assume you were doing this when your FSB was 400. At FSB 334 it should show up as DDR2-668. Don't worry it's slower than rated speed. You could probably change your memory to 4-4-4-12 at that speed. You won't see much of a speed decrease.


Is that a mini Ninja? You want maximum speed with a little....? No matter. Stick another fan on it 90 degrees to the first one, that will increase the air movement through it Better yet try strap on a 120 and have the excess portion of the fan blowing onto your NB.

You can't vary PS voltages so they provide little help. CPU, NB, and RAM voltages are useful. My board gives MCH voltages instead of NB. I leave that option on AUTO and it works fine. It would not surprise me though if your system ran a tad cooler and more stable with a larger power supply. Best to borrow one before buying.

These are issues I will have to deal with when summer comes. I'm in a second floor apartment and it's 60 right now. Yeah, heat is expensive but wearing sweats and using blankets is not. But come summer temps will surpass 85 or more than 10C higher. That will mean my temps will be too high and I have to deal with them. I suspect in addition to better cooling I will have to lower my FSB from 400 to 333, that's worth maybe 3-4C. I have a project case I am working on, modded for better air flow all around. I may even put a fan behind the motherboard where the cpu mounts. One fan at the top sucking out hot air near the front where air doesn't move much. The X2s don't seem to be an issue, especially with the Biostar 7050-M2, the only board where I didn't stick on any additional small heatsinks. It's a real champ.

Wow, aristide1. Thanks a ton for the detailed replies. I really appreciate the time and effort you took to respond in such a manner.

(Thanks too, floffe. I appreciate your suggestion as well-see below.)



The non-beta speedfan lists the core temps in the low 50's, like yours. With the requisite 15 degrees added on to it it would put the actual temp more in line with coretemp results. The newest beta of speedfan has this offset built in, and those are the temps that I reported above. If you're talking about using speedfan and monitoring temps in the low-mid 50's, then I think I'm probably in the same ballpark as you. If you are using the beta or coretemp and get those numbers, well... I guess I'll have to spring for a real Ninja.



Good suggestion. I'll keep the cover off and monitor the temps over time. Could be that I have lots of flow, but poor positioning of the fans. I'll report back after a few days.



Since the nb-47f is only about 10 bucks at microcenter, I might try it. If it stinks or doesn't change anything, I'll report back one way or another and probably get the enzotech copper heatsinks. Would you recommend a 40mm fan for it (shudder) or something else?



What kind of thermal adhesive did you use? Epoxy-type stuff or the tape-type stuff? You think aluminum heatsinks would be acceptable instead of copper?



Okay. I'll take yours and floffe's advice and decrease the FSB and raise the multiplier to 9. We'll see what happens to the temps.




Yep. Mini ninja. One of the threads I read before buying this board had a guy on there that used it and got some crazy overclock. In retrospect, sure, I probably should have gotten the regular ninja, but I liked the idea of having a heatsink that I could use in the future in a media case or the like. I'll look into adding a 120 versus just getting a real ninja.



I used memtest to check the memory when i first oc'd this board. I kept getting an MCH error around test 7 until I upped the NB voltage. More voltage to the NB is probably one of the contributing factors to it running so hot right now. If I go down on the fsb, I can very likely go down on the MCH voltage and probably run at tighter timings. Have to run memtest again, and retest next week.



Heh, I hear you about the heat. I'm trying to keep costs down right now. I've got the heat on about 66, which makes me wonder if the temps should be even higher. I would turn it down more, but any lower than that and I freeze to death while trying to sleep. I hear you about the AMD dual cores. My Opty 165 at 2.7 didn't get this hot. But, it doesn't move past checkpoints like the OC'd e2160.

No problemo <- Terminator accent. But also I want to make the FAH Subject here more interesting.


Hmmmm, now I will have to go back and check, I used Coretemp, don't recall which version. But my Tuniq Tower is never hot, even the heat pipes aren't that warm. I do recall that Arctic Silver 5 was about 2 degree better than most other brands.


Fans have 2 purposes:
1. Move air in or out of the case.
2. Move air hovering around heatsinks and hot areas.
On occasion with a duct a fan can do both functions. I have yet to try.

I have:
1. The stock fan in the Tuniq Tower at about 1/4 speed, about the 3rd dot. I can tolerate it up to this speed.
2. The stock fan in a Seasonic S-12 430
3. One low speed Yate Loon at 12 volts.
Not silent, not too bad for somebody already losing his hearing. All fans are 120mm.


But the whole point of the Sidewinder ads was that they weren't all that much over $10. And do you have any Arctic Silver 5?


I have epoxy but haven't used it yet. The thermal sticky tape has worked well so far for me. I always lay the case down and let gravity do its thing for a day.

ANY heatsink on that "T" will help. Hell, people use pennies as heatsinks. I recommend pre1980 all copper, not copper clad. But real heatsinks are typically made to have lots of surface area. This is why when anyone throws away pc parts I grab any and all heatsinks. If you have room you could stick one of these on each flat side of your NB for $2.50 total. You do need to watch your clearance.
http://www.sidewindercomputers.com/miso.html

I plan on using this on my southbridge. Then I will take the original southbridge hs and epoxy it to the side of the NB. Now that's recycling.
No 40mm fans. Hang an 80 in the neighborhood with some bungy cords or something. Stick the 80 in your side duct.

If you change your NB HS you may need a rubber for protection.
http://www.sidewindercomputers.com/micoprru.html


If you maintain 3.2GHz in any case don't expect cpu temp drops. Shoot for 1MHz passed the normal divider frequency. Boards differ, at 333 you could be at the high end of the divider or at the beginning of the next divider. 334 eliminates the question, as would 401. (Sometimes I can't believe how much technical info is out there on this stuff.)


Sure somebody got great temps with this and a triple thick Delta fan running at 7000 RPM. I would say cpu temp is your first priority. If need be add another 80mm fan to the Minja. You'll probably see results in minutes.


My NB HS was 60C stock! You added a little more to a bad situation. So what? Move the air, probably worth at least 5C. Don't sacrifice being able to run Prime 95 successfully for a small amount of NB voltage, it won't break the bank.


I'm all for saving money. You can always epoxy pennies to the edges of your nb-47f. If you read it I got a huge increase when going from 266 to 333, but only a small increase going from 333 to 400. Reason is every 66MHz the board resets the NB divider back to a safe speed. An article published long ago said that if you took a pc and simply dropped in a cpu that was twice as fast, and kept everything else the same, it would only feel about 20% faster. To feel twice as fast everything has to be twice as fast, and that just won't happen. When I OC'd my X2 I had to slow my memory at one point. Man, I took some hit, but I had to if I was going to continue to increase the FSB. I think each "setback" becomes more costly. So I have no plans on slowing the memory again in order to pursue an even higher FSB, it's just not worth it. The only numbers that counts are FAH points and time between checkpoints.

I'm under a comforter with several inches of goose down, I think. If I used that and the temps were set at 66 I would wake up soaked with sweat. It takes a little while to get used to these temps, and I will raise them when it gets really cold out, ie below zero.

Added - I recall "issues" with Ninja's and the pushpin business. Some people found out they work better with more downward pressure. To determine this put the board horizontal and then put some more weight on the top of the Minja. Try some quarters or an Ike dollar coin. If that works you will probably have to alter the securing mechanism to screw down. It's not a big deal, just a PITA.

Sorry to hear about your problems. We've been chasing each other up through the rankings. Wondered what happened. Hope you get everything back up and running soon.

Keep folding, its good for the neighborhood.

My temps are 54/56C worst case with Coretemp .95.4
Room temp is 60F, so add 3C more for appx your room temp (a guess).

Thanks, Dave. Had to give up one of the boxes because we're moving. I'm starting to save up for a quad so I'll be back to where I was in a few months, hopefully.

I'm using coretemp 0.96.1.

Okay, here's the update. I knocked back the FSB like you suggested to 335. Memory is running 1:1 at 4-4-4-12. Was getting 17:35 min per checkpoint on WU 2635 compared to just over 15 min at 400x8. Left it running over the weekend when I drove back to MN. Got back late Sunday/early Monday AM and found the box had crashed again. Took off the mini ninja and looked at the heatsink. You were absolutely right about the "issues" with the retention mechanism. Whoever thought of this wonderful system for intel was obviously working for in secret for AMD. It looked like the middle of the sink didn't make good contact, which is odd as hell since I checked, double checked, and then rechecked the retention mechanism when I installed the thing outside of the case. I was using arctic silver 5. I really like the stuff. Usually put on a very thin layer with a latex glove or some such (saran wrap works in a pinch) and then set the hsf on the board, take it off, make sure it contacted the integrated heat spreader, then re-attach once again and double check (?triple check?) the whole kit and caboodle uno mas. Since I had everything out of the box, I decided to check out the NB heatsink as you suggested. Was still hot, enough to burn my finger if it was on it for 2 seconds or more. Removed it, and thick putty-like goop lay on the inside. Egads. Went down to microcenter the next day and paid for the NB47F and fitted it to the NB. Wanted to go with the Sidewinder deal, but the credit card was over the limit. Way over the limit. Installing the NB47F was a bear in its own little way because the thing has to fan out anterior to posterior instead of superior to inferior due to the CPU heatsink and the pushpins were really hard to get to. Suddenly realized that I didn't have any AS5 anymore because the movers had packed up everything in the apartment (and by everything, I do mean everything--found out there was no TP in the apt the hard way). Well, I couldn't go another week without a folding box, so I went back to microcenter and bought a real ninja with my operating money for the week. I may be subsisting on ramen and beans (and toilet paper), but at least I will be folding. With the included 120mm fan from the real ninja, 120mm intake, exhaust and the back pci covers removed, coretemp 0.96.1 reports core temps at 56 and 59 at full load folding. Kept the multiplier at 9, FSB is now 345. Actually increased my times to 18:40. Decreased the Vcore and NB voltage a couple of notches, no apparent ill effects. Next step, to increase the FSB and see what it does to stability and to the temps.

Oh, and it was 58 this morning in KC on the way into work at 6. By 8 it dropped to 19. Guess I'll be sleeping in jeans tonight since I forgot to bring an extra blanket last weekend when I came back from MN. I'm not turning up the heat.

You are one hardcore folder

I think people have installed stuff like coins in between the cpu and the ninja. Others have modified them to use the ThermalRight through the board attachment mechanism.

Decreasing voltages can cause errors that only Prime 95 will pick up.

OT - My E6400 is running at 400*8, 3.2GHz and uses 150 watts.

Yes NB47F is a bear to install, but makes a big difference.

Mr. K Killa,

You need to read this.

http://www.madshrimps.be/?action=getart ... rticID=567

Thanks, aristide. Looks like a lot of work to bend the fins, but the reviewer did notice some drop in temps, so it might be worth it. Not crazy about putting two fans on in a push-pull configuration, though. The real Ninja decreased temps by a decent amount at the same noise level compared to the mini Ninja, so I'll probably stick to one fan for the time being. I suppose I could remove the rear fan from the case (Antec SLK 3700 clone) and do the push-pull thing, though, with no increase in sound.

The beta speedfan expired, and I had to install 4.33. This version doesn't jibe as well with coretemp 0.96.1. Speedfan now shows variations of 48-53 (min-max) with the proc set to 401x8 and the vcore at 1.36 after vdroop under orthos. Coretemp shows circa 63-67 degrees. No difference with the case open or closed. Almost 18 1/2 hours stable under those settings, though. Touching the new NB cooler, it is cool to the touch.

With some more time, I'll experiment more with the setup and report back. Maybe the through-the-board mounting system would work better....

I'm not bending no stinking fins. But I thought some of the other info may have been helpful.

do you think lapping makes a real difference?

Well I suppose that if you were in a borderline situation, like your recent heat levels, then its a real difference. But if you are in the low 50's already why bother?

You know I was kinda surprised. I upgraded one of my X2s to a Black Edition. I pulled off the old 4000+'s hs and:

1. The AS has set up to a soft solid. I had heard it was kind runny, but not from what I saw. Perhaps mine has been reformulated?

2. I expected a high spot or 2 on the cheap standard heat sink. What I found was there was no clear spot of contact, there was an evenly thick layer of AS across the whole thing. Almost like the clamping mechanism didn't put enough downward pressure on the thing, which if you've seen the AM2 clamping mechanism, is kinda hard to believe. Its much better than push pins.

This made me curious and I got to check 2 different but standard hs's that come with the 4000+, mine and my roommates. With a razor blade I discovered neither one was flat. More like a little wavy. But the damn X2 puts out so little heat, even humping away at 2.8GHz that it's still good enough. But it definitely is a good candidate for a lapping if money is an issue. Add a bigger fan and Im sure it would be outstanding.

Well the statement that did it for me was

I had a sneaking suspicion after reading and reading and surfing and reading and surfing...you get the idea...that it might be my windows install that was causing the machine to crash after folding for a couple of days. So, I decided to try to install kubuntu. No go. The cd wouldn't even boot to the live desktop to install. I used the 64-bit version of 7.10. It's funny, because I was able to install SuSE 10.2 on my old Opteron 165, but this machine has 2x2GB of ram, so that may have something to do with it. I dunno. I just spent the last 2 days / nights reinstalling windows after changing to a single hard drive. Updates off (thanks, arisitide). We'll see how it goes over the next few days and I'll report back.

If you couldn't get kubuntu to boot, I suggest you try Mandriva 64-bit.
It has always booted out of the box on all the compaq and HP boxes I have had at work, and also on all the home-built ones I have had. I too had some issues with the 64-bit ubuntu.

I will understand if you are a bit put off by the idea of doing yet another install (although if you d/l the DVD version, your install should take less than 90 minutes), but I know you are a dedicated folder, and the SMP client rocks in Linux.

And I can provide some technical support, since I have been using Mandriva and its predecessor Mandrake for nearly 8 years.

Regards

Mike

I hope you realize that you can not install anything properly when you OC. You need to drop to stock speeds, install, and then return to OCing.

That may be true of Windows, but does not hold necessarily for Linux. At least, not with mild overclocks of 20 to 25%. I am hosting various versions of Red Hat, and the new Mandriva Spring 2008.1 in VMware under Mandriva 2008.8 x86-64, and I installed the guest OSes without removing the OC. And I have had the host installer start a couple of times on me, when I left the install DVD in the drive through a reboot.

Still, hopefully the killa's windows box will now fold more stably, and this will all remain moot.

Mike

I agree, but who knows what Kansa is doing?

I'm folding as fast as I can for team 31574.

Plico celerrimus.

The update is that the box has been stable for circa 48 hours so far at 401 FSB x 8 with ram at 1:1. Circa 15:20 checkpoints on the current WU (2653, run 0, clone 181, gen 55). The previous WU was also a 2653 that gave 16:17 checkpoints. The real test is going to be if it stays up all weekend. I'll be back Monday night to check on it and report back.

BTW, anybody know how to see all the notifications? Often times the time stamp on the left hand side obscures some of the text when it is displayed. I'm running with the -verbosity 9 flag.

Oh finally, something that Google can't find.

Well, the box locked up again, and I originally thought that it had something to do with the servers being down over the weekend. I restarted the client and the next day it locked up again.

So, realizing that uptime is more important than a fast, fast, fast checkpoint time, I raised the multiplier to 9, dropped the FSB to 357, and dropped the timings on the ram to 4-4-4-12 from 5-5-5-15, though it is running at 1:1. End result is that I went from ~15.5 minutes per checkpoint to ~17.4 minutes per checkpoint. We'll see if that makes a difference.

I did do some looking into network issues this past week, and I wonder if my cable provider is interrupting the net on the weekends to discourage torrents and the like. Others have reported that the client will stall or freeze when this happens in the middle of a WU. Any thoughts on net access and stability of the client?