SSE & AMD Crashes Revisited

[nikhsub1]

Well, Audioaficionado and I have been PM'ing each other with an idea he had about maybe solving the SSE and AMD (T-Bred B's and Bartons) crashing at random when forcing SSE. His idea was (one I never thought of myself) to set the CPU usage down a bit from 100%. We thought maybe if there was some headroom, the CPU might not crash. Well, that theory is out the window. This morning one of my layers in teh farm crashed using SSE. I had it reset, logged in remotely and changed the config file to use 90% usage. It locked again about 20 minutes ago.

I was really hoping this was going to work, sad to say it is not, back to 3DNow! for me :mad:

[drshivas]

Is this type of crash a sure thing or a luck of the draw? I have a Barton (see sig) running with SSE (3.25 Beta) and it hasn't crashed once in the past month or so since I switched to SSE from 3DNow!.

[JDXNC]

No bad luck here either, my 2100 runs SSE and has never locked up because of it. I'm stumped myself as to what could be causing your problems, but if I think of anything I'll be sure to let you know right away :)

[bubba gump]

Its only I think in some of the older Barton or Tbred 'B' cores....for me its not that it locks up, its just that the damned thing kills the client :(

And isn't it something in the code that has the problems? Because I have been able to render stuff in 3dsmax using SSE fine....no errors on that stuffs (wonder if it would even give any)....

Hmm....I really hope Pande Group can get this sorted out to a software or hardware problem...if it hasn't already..

Fold and Frag on
Brian

*edit* this is very wierd...very wierd indeed, I can get it folding w/sse, and the client doesn't die when I changed two things....gonna try w/just having one changed in a bit but here is what I changed.

--Changed the CPU usage down to 85%
--Wild idea, but I changed the flag order from -service
-advmethods -forceasm to -forceasm -service -advmethods

I'm going to see if bumping the usage back up will effect it, or if I actually get a hard lockup instead of the client just crashing from SSE....

More back soon.

[JDXNC]

While we're on the topic of SSE, aren't morgan Duron's supposed to be able to run SSE? Whenever I run it on my 1.0Ghz Morgan it kills the client :confused:

[bubba gump]

And that seems to have solved the problem :o

If anyone else would mind trying this out...please change your flag order whether you use a shortcut or you use a service install of somesort (F@H service install or One-Click) and try changing it so that its like mine, -forceasm -service -advmethods or some other way where the -service flag is not first.

I have definately received an increase on this wu...

W/o SSE on a p683_TZ2_NAT_EXP
[23:45:53] Completed 1900000 out of 2500000 steps (76)
[00:21:07] Writing local files
[00:21:07] Completed 1925000 out of 2500000 steps (77)
[00:57:00] Writing local files
[00:57:00] Completed 1950000 out of 2500000 steps (78)

W/SSE on a p683_TZ2_NAT_EXP

[01:08:29] Writing local files
[01:08:29] Completed 1950000 out of 2500000 steps (78)
[01:08:29] Extra SSE boost OK.
[01:24:53] Writing local files
[01:24:53] Completed 1975000 out of 2500000 steps (79)
[01:41:05] Writing local files
[01:41:05] Completed 2000000 out of 2500000 steps (80)

This will be a crazy fix if this ends up working for everyone....no lockups yet, but maybe down the line there will be *crosses fingers*

Fold and Frag on
Brian

[JetMech]

Originally posted by JDXNC
While we're on the topic of SSE, aren't morgan Duron's supposed to be able to run SSE? Whenever I run it on my 1.0Ghz Morgan it kills the client :confused: I believe only the later Morgans have SSE (1.3).

[bubba gump]

Nope, all morgans have sse support, at least according to Gnerma in his thread for Gromacs...and the 1.3GHz Duron I have folds w/SSE just fine.

http://forum.oc-forums.com/vb/showthread.php?s=&threadid=130328&highlight=talk+AND+gromacs

Linky to gnerma's thread :)

Hope that helps a bit...

Fold and Frag on
Brian

[drshivas]

I have a 1.2 Morgan running SSE fine as well. There was a tremendous boost in WU time with it enabled:)

As for my flag order, it is this: -advmethods -forceasm

I launch my FAH client with a shortcut in the startup folder, and not as a service.

[Arkaine23]

I think you all are forgeting that Stanford makes no guatuntee of stable WU's for people using -forceasm.

[bubba gump]

Yes, very true Arkaine, but before the tbred b's and bartons we had stable SSE usage, so I think the main goal is to try to get these back to as much as we can w/these new loverly overclockable chips :)

Hopefully we can get this matter solved...or at least diminished into smaller areas...hopefully...

[nikhsub1]

There is a 'test' WU that will crash your AMD rig within 5 minutes... with the forceasm flag. You can download it HERE (http://home.broadpark.no/~lsevalds/FaH.zip) but make sure you set the date on your computer to like May 1st of this year BEFORE you run it. Also, you need to create a shortcut to the exe so you can run the flags. The thread on this is 15 pages long and is found here: http://forum.folding-community.org/viewtopic.php?t=4444&postdays=0&postorder=asc&start=0

This WU locks EVERY SINGLE ONE of my T-Bred B CPU's but my Pentiums fly through it like nobody's business.

[Audioaficionado]

Bummer about my theory not panning out :( I was hoping we'd all got lucky this time.

I'll have to try out that 'known bad WU' on my new Barton as soon as 'slow SVC' gets me my AS-5 so I can mount my heatsink.

I'll try out those alternate flag positions too and see if it makes a differance.

I'm running s2k3 as a workstation OS so I'll see how it works next week. (after I get my parts from SVC) Grrrrrr...

[JetMech]

Originally posted by bubba gump
Nope, all morgans have sse support, at least according to Gnerma in his thread for Gromacs...and the 1.3GHz Duron I have folds w/SSE just fine.

http://forum.oc-forums.com/vb/showthread.php?s=&threadid=130328&highlight=talk+AND+gromacs

Linky to gnerma's thread :)

Hope that helps a bit...

Fold and Frag on
Brian Sorry. In the AMD Duron Model Data sheets no mention is made of SSE instructions in specific terms while 3D Now!, Enhanced 3D Now! and MMX are. So much for assumptions.

[bubba gump]

Hmm...interesting..was this just for 1GHz Durons? Becuase I know my 1.3 (which is a newer one...I think :P) can run SSE...

Good luck getting the AS5 from SVC man, too bad ya couldn't get the 12gram from Jab-tech :)

Gonna try that WU soon nik....lemme just finish one I know I can :-P

Fold and Frag on
Brian

[JDXNC]

Well that said maybe I just ought to get a new mobo(would that have anything to do with it?) and an Applebred :)

[Fast420A]

I think that finding 1500+ and up Pally's is the way to go for a farm that's going to be at stock speeds. That's what I have with no crashes to far. The board doesn't even have any features but I am going to play around with some wire tricks sometime.

[Chris]

Lets hope stanford fix the problem, its loosing them and us lots of production!

[scoobiedoo]

LOL,

Rumor has it there will be no SSE for any cpu at all in the future.

but that's just a rumor.

scoobie

[Fast420A]

I heard that rumor and hope it stays just that, A rumor.

[bubba gump]

Well that WU seems to give me a hardlock...still no hardlocks on this WU w/SSE running successfuly :D

Hopefully I won't get many bad WU's....

Fold and Frag on
Brian

[walaka7]

i had a pally eat dirt on a 102x yesturday, and im also hearing of intels crapping out on the 10xx's as well. so it would appear, to me anyway, that there are some issues with their newer batches of wu's. My dually dumped two (tbred) for the first time ever :eek: one 102x went south and took the other client with it (348) then it kept thinking the cores were corrupted until it downloaded 5 of them, then it just gave up and went to "sleep" (sounds like my co-workers)

[Nickme90]

Quick question. How do I go back to just 3D-Now?

Do I have to replace the 3.25 beta client with the 3.24?

Cause If I just take off forceasm it drops back to standard loops not 3dnow.

I swear this problem has got worse in the last 2-3 weeks. Before that it might have locked once every week, now I finding it doing it like 3 to 4 times a day.

Thanks for any help.

[Audioaficionado]

Without the flags I always get 3Dnow with the 3.25 beta client unless I pick up a tinker. Then I don't even get 3Dnow.

[walaka7]

yeah nick that is a way to do it.. run 3.24 with flag and let it revert back to 3dnow, however you would have to run the first unit without fail before it reverts back. as far as running no flags, its hit and miss on getting 3dnow on the 3.25 client

I feel for ya, my dually after 10 months with NO wu loss all the sudden dumped both clients within 20 minutes of each other, and my yatta which has been kind to me for over a year hard locked on the same project at the same frame 3 times in a row

[Fast420A]

On my dad's rig which is VERY quick to lock up with SSE, I had to drop the 3.25 and use the 3.24 with just the -advmethods flag.

[man_utd]

Blah, I don't know why but My WU number has gone down the drain on P4. I realize that this is the wrong place for that, but I also have a 800 Mhz Slot A Athlon, wondering if it has SSE.

Either way, with the P4 I had 3 WUs done in a day 3 weeks ago, now i am barely getting 1 out every day and a half.
Any ideas on how to fix?

[Fast420A]

The Slot A 800 Mhz Athlon doesn't have SSE, I have one myself that takes about 45 minutes a frame with 3DNow on the Gromacs.

[Flashfx2]

nikhsub1, i'll try that test core sometime, probably tonite. I don't think my rig has ever crashed on random and I've ran the sse boost ever since the 3.25 beta has came out. It's a tbred B BTW.

[Audioaficionado]

Originally posted by man_utd
Blah, I don't know why but My WU number has gone down the drain on P4. I realize that this is the wrong place for that, but I also have a 800 Mhz Slot A Athlon, wondering if it has SSE.

Either way, with the P4 I had 3 WUs done in a day 3 weeks ago, now i am barely getting 1 out every day and a half.
Any ideas on how to fix?

Just check how much CPU resources FAH is using. If it's in the high 90s, you're ok as there's no other app stealing resources.

Other than fine tuning your OS/system there's nothing else you can do if the WUs they're sending are getting weird on us. It'll probably get better in time as newer WUs get developed.

[Arkaine23]

Yeah, a lot of it really just depends on the WU's themselves. New units may not have this problem...

[muddocktor]

I've had a few hard locks on 1 Tbred B and 1 hard lock on my Barton from these p10xx WU's myself.

Nik, I have a scenario here that will really make you scratch your head though. I just got a Chaintech dually board from Newegg's refurbs last week and put it in place of the KG7 board in my daughter's rig. I used the XP2400 that was used in the KG7 and robbed the XP2100 out of my KHA+ system to put in the board. Both procs have been hard coded to a 16 multi and I did the L5 mod to enable SMP. Now, that XP2400 would never run a gro WU reliably using SSE in the KG7, even at stock speed. I installed it into the cpu2 socket on that mobo, with the 2100 being in the cpu1 socket. Since I've gotten that machine up and running a week ago, both procs have been rock solid and no lockups at all, running at 2208. I haven't changed anything else besides the mobo and adding the 2100 to that rig either.

What in the heck would let that 2400 that would hard lock processing with SSE in the KG7 suddenly enable it to run SSE while processing Grommies now?:confused:

[Audioaficionado]

Before you locked it at a 16x multi, it like all Tbreds and Bartons was an unlocked processor.

Hmmm...

[RoadWarrior]

My Tbred A has been stumbling along on tinkers since late august because I was tearing my hair out with it munching units all the time. Part of that was heat and now in the cooler weather I should experiment again . However, before that it went like a month on SSE gromacs with no problems, then it was locking up once a week, then it was every couple of days, then it was every few hours... I am wondering however if I damaged it because I accidentally overheated it one day. Also a suspicion of mine is that the problem is worse on VIA chipsets, specifically the KT266, 266A 333 and 400 with DDR ram. I say this because it seems that people with Tbreds in older boards, and cheap farm setups with SDRAM seem to have surprisingly fewer problems than those who have them in "faster" VIA boards. However, I notice a few of you in this thread seem to be running on nforce and having problems, maybe it's a wierd DDR issue.

One thing I did notice on my rig was a dramatic difference in stability between running cas 2 and cas 2.5, you'd expect 2.5 to be "safest" right? well actually on my machine cas 2 works far better!?! I have also noticed a few other poeple coming across this on Tbreds and Bartons, cas 2.5= crashy, cas 2= stable, wierd huh?

Anyway, as muddocktor has also observed, I think half the answer at least lies in the mobo.

This problem is extremely hard to evaluate since gromacs SSE crashes will happen when the CPU is taken too far, I tend to ignore anyones protestation that their rig is 100% stable if overclocked and only tend to consider seriously a problem as significant if the problem also occurs at stock and under speeds.

regards,

Road Warrior

[Fast420A]

My Dad's frequent crasher is a Tbred B 2100 on a VIA 400 chipset. My not very often crasher is a 1700 Tbred B on an Nforce2 chipset.

[walaka7]

Well ive been reading here and at the "community" Im seeing reports if problems in the 10xx series in sdram with pally, sdram with tbred a/b DDR with tbred a/b barton and even some intel users have been having issues. It is quite true that all hardware acts a little different and some might be fixing their issues by changing configurations. My question is this though: when does this stop being a "hardware/overclocking" problem and a project problem. Right now their argument is that since hundreds or thousands are being correctly done and turned in that the problem must be elsewhere. That just makes me scratch my head and wonder. THey have no idea who's turning in the good ones (my guess is P3 users as they seem to have the least amount of issues) and i dont know if they can tell how many are going south (i guess if they can count their uploads verses returns they could get an idea, but they may be just counting hte partial credit ones which would certainly throw the numbers was off). So my suggestion for now is this: IF you are having problems with a wide variety of projects, then by all means , investigate your hardware config and make apropriate changes. the mobo has just as much to do with SSE op as the proc. IF your problem is only with certain projects, then report it to the community: http://forum.folding-community.org/ . make sure you save your log file and copy your folder with F@H in it (exe, client, everything) . If you still decide to make hardware changes to camoflauge what I believe is a project issue, then by all means have fun. but they may come out with a different project in the near future that wont like that config either, so proceed at your risk on that. I myself am waiting for 110 to come back up and try to duplicate the problem on the same project. Then im gonna make a decision. Both of the rigs of mine that locked up or failed wus, have been fine for a couple of days now. im not aware of a hardware problem that phantomly comes and goes on a rig that runs 24/7. Good luck to you all as lost production is bad for everyone involved

[nikhsub1]

Mud that is truly weird. I think it (the crashes) have to do with the whole system and not just the CPU. I have NEVER had an Intel choke on a gromacs unless of course I OC'd way to far. I can not figure it out myself, one of those 'chase your tail' scenarios.

[muddocktor]

Nik, the only thing I see that is changed from when that rig was using the KG7 board (besides a fresh install of WinXP versus 2K it used to have) is that the 5v rail is a lot higher and more stable than with the KG7. The Chaintech board uses all 3 psu connectors; the ATX connector and both the 12v and 3.3/5v P4 connectors and my 5v rail is now rock solid at 5.00-5.03v. With the KG7, which only uses the ATX connector, my 5v rail would vary from around 4.91v idle to 4.82v when loaded. This is with using the same Enermax 651P-VE psu with both boards.

Wally, I agree with you there; it's like Stanford and some of the mods/admins at the community automatically assume that any problem such as the lockup problem is automatically a hardware based problem and not as a result of some flaw in their precious codework or work units. There is one admin over there that really torques me up with his snide, snotty, sarcastic answers that he sometimes gives, and is a detriment to the project as a whole, IMO. The few lockup problems I've had lately have all been due to the 10xx series of wu's and haven't had any problems with the other grommie work.

RoadWarrior, have you checked your ram stability lately? I had hell with my Asus dually for a while with spontaneous reboots, hard locks and kernel dumps until I figured on checking out my ram. I had a stick of Corsair registered that went bad; wouldn't even run memtest86 far enough to even start to check the ram and would crash the program and hard lock.

Audioaficionado, they're not locked, even now. It's just that these MPX boards don't have a way to change the multiplier manually except the K7D and it can only go up to a 12.5 multiplier in bios. I could take either proc out of this board and throw it into my NF7-S and change the multi in bios to what I would want.

Fold on!:D

[Audioaficionado]

Originally posted by muddocktor
Wally, I agree with you there; it's like Stanford and some of the mods/admins at the community automatically assume that any problem such as the lockup problem is automatically a hardware based problem and not as a result of some flaw in their precious codework or work units. There is one admin over there that really torques me up with his snide, snotty, sarcastic answers that he sometimes gives, and is a detriment to the project as a whole, IMO. The few lockup problems I've had lately have all been due to the 10xx series of wu's and haven't had any problems with the other grommie work.



If they keep up doing that kind of crap, all the PR and spin doctors in the world can't repair the damage. We're doing this as a goodwill gesture and favor trying to help with their research. Alienate the folders and who else is going to spend the time, money and effort to keep all the DC clients working?

[RoadWarrior]

Yeah mudd, I was thinking maybe I had a RAM problem while investigating all angles, threw memtest86 at it and it passed no errors half a dozen times at least, ran a few more RAM checkers/tests on it too with no result.

[Audioaficionado]

Originally posted by muddocktor
Both procs have been hard coded to a 16 multi and I did the L5 mod to enable SMP.

Originally posted by muddocktor
Audioaficionado, they're not locked, even now. It's just that these MPX boards don't have a way to change the multiplier manually except the K7D and it can only go up to a 12.5 multiplier in bios. I could take either proc out of this board and throw it into my NF7-S and change the multi in bios to what I would want.

Fold on!:D

If you didn't mod any bridges other than L5, how did you hard code them? Socket wire trick?

The NF7 is a special overclocker's board and can do things most others only dream about ;)

[muddocktor]

I had previously modded both the 2100 and the 2400 to 16 multipliers a while ago, so I could maximize their folding potential in the KG7 and KHA+ boards I had them in. Those old boards don't have a 1/5 pci divisor and I don't like to run my pci devices too far out of spec. The 2400 was really easy to mod, I only had to blow 1 bridge to change the multi from a 15 to a 16. The 2100 required a little more work in that I had to fill a couple of laser pits in the L3 bridges so I could connect the middle L3 bridge and I had to blow the 2nd L3 bridge on it. All that is changed on these procs though is the multiplier used though. You can stick them into a nf2 or Kt400/600 board and still have multiplier control. Any board that makes full use of the 5 bit multiplier control should be able to access all the working multis in bios AFAIK.

[walaka7]

Originally posted by muddocktor
Wally, I agree with you there; it's like Stanford and some of the mods/admins at the community automatically assume that any problem such as the lockup problem is automatically a hardware based problem and not as a result of some flaw in their precious codework or work units. There is one admin over there that really torques me up with his snide, snotty, sarcastic answers that he sometimes gives, and is a detriment to the project as a whole, IMO. The few lockup problems I've had lately have all been due to the 10xx series of wu's and haven't had any problems with the other grommie work.



Yeah ive been a bit dissapointed so far in the reponses i've gotten to my posts. Maybe its just me though. Im just not one to go changing anything until at least a half arse hypothesis is derived. To suggest a rig thats been flawless for 10 months has a bad proc just doesn't make sense to me. But i have not been there an ample amount of time to form an educated opinion of the "community" and i will maintain an open mind, just not an open case.. not yet.

[scoobiedoo]

hehe i pm'ed vijay over at the community forums...haven't recieved a reply yet....as for the stability issues...i'm guessing that since most of these wu's have been people using the -advmethods aka the beta tag....you probably are getting a good amount of beta testing wu's that are just super sensitive or not stable at all.

this is just my hypothesis on the matter at hand...maybe we should all remove the -advmethods tag and see what comes of it then as the regular client will get gromacs without that tag now.

scoobie

[Audioaficionado]

Originally posted by muddocktor
I've had a few hard locks on 1 Tbred B and 1 hard lock on my Barton from these p10xx WU's myself.

Nik, I have a scenario here that will really make you scratch your head though. I just got a Chaintech dually board from Newegg's refurbs last week and put it in place of the KG7 board in my daughter's rig. I used the XP2400 that was used in the KG7 and robbed the XP2100 out of my KHA+ system to put in the board. Both procs have been hard coded to a 16 multi and I did the L5 mod to enable SMP. Now, that XP2400 would never run a gro WU reliably using SSE in the KG7, even at stock speed. I installed it into the cpu2 socket on that mobo, with the 2100 being in the cpu1 socket. Since I've gotten that machine up and running a week ago, both procs have been rock solid and no lockups at all, running at 2208. I haven't changed anything else besides the mobo and adding the 2100 to that rig either.

What in the heck would let that 2400 that would hard lock processing with SSE in the KG7 suddenly enable it to run SSE while processing Grommies now?:confused:

Do you think that L5 mod might have something to do with it?

What does the L5 mod enable or disable that could make it more SSE stable?

[walaka7]

The L5 mod only makes the board recognize the proc as mp. There isnt any known connection between that and any function of the proc itself.

Now there has been a noticable change in atmosphere at my thread in the community. And im not quite so dissapointed anymore. The general consensus is still hardware, but they appear to no longer be taking a hardline aproach to it. All of my rigs are banging out 344's and have been for the last 24 hours. I will certainly update as soon as i start getting 10xx's again. Im sure most of you already know this, but there is a simple little proggie that one fo the mods over there listed that stresses SSE and the whole proc, for that matter to the max. I actually did see a 2 degree rise in core temps after running this proggie for a bit. it is found at teh bottom of this page http://home.comcast.net/~wxdude1/emsite/

[Audioaficionado]

Why won't the dual mobos just run the XPs as MPs if that switch doesn't affect the way the CPU behaves?

[walaka7]

becasue when the procs are tested for smp use, the ones that either fail, or not get tested at all get the bridge cut and are sold as xp's. Not all xps will work as smp. Thats teh risk you take when you jump the bridges.

[Audioaficionado]

If it's just an info string then the BIOS could run SMP if it could be set to ignore the non SMP status. They just don't want to support any problems with the XPs that won't do SMP, so you won't see any MP mobos setup that way.

BTW (more on topic) I just fired up my Abit NF7-S & Barton and have it folding gromacs SSE. I'll see if I have a good CPU or a problem one soon enough. Rather than do a custom OC, I just picked the XP3200 option and left everything else at default. My other choices were XP2500 or Custom.

I'm running FAH SSE 100% CPU load @ stock 1.6vcore/200/2200 @ 23c/41c. That SLK947U, quiet 92mm Zalman fan and AS-5 seem to be doing a great job :thup:

Even though the package says this is a Barton XP2500, the core is most likely an XP3200 so technically I'm not even overclocking this system yet. Sweet :cool:

[walaka7]

/me crosss fingers

[Audioaficionado]

Hmmm...

Had my first lockup. Not sure what caused it but I was finally able to restart after several ctrl+alt+del keystrokes so the WU wasn't lost. Now I'm set at 95% CPU usage to see just what happens overnight.

[nikhsub1]

Audio, I have tried varying the % of CPU useage, does not seem to help. I have also tried an L5 modded CPU, and that locked too. I have some T-Breds that hardly ever lock, like maybe once every 3 or 4 months and others that will do it daily with SSE forced. No rhyme or reason to it yet.

[Audioaficionado]

Mine made it trough the night alright. I'll see how it goes as the weeks unfold (no pun intended).

I'm thinking it's just like the overclockability of any individual CPU. It quite often is just the luck of the draw. I've had the exact same model number of everything that the other guy had and yet he could OC 200MHz higher than me. Theres a factor in these CPUs that isn't on the spec sheet and it determines wheather we can run F@H with SSE or not.

If Stanford wants us all to use SSE for more production, then they'll have to figure out what it is in their WUs that is triggering this instability. I hope this isn't going to be a problem in the AMD64 or Opertron processors otherwise I'll have to migrate over to Intel all together. I'm not going to waste the electrical power expenses on 3Dnow when I can get twice the production with SSE.

[dz]

I've been running SSE for 15 units so, it's locked twice (in a 30-minute period) but that was a temperature issue I think. I'm not even sure if it was the CPU or vid card that locked, I've got the vid card cranked up quite a bit, and it was warm in here (too much for the heat pipe and junk thermal compound I guess)

[Audioaficionado]

I ran my old XP1700 DLT3C cranked up to 215/2260 all summer long at 55c and had no FAH SSE problems.

This Barton isn't really overclocked yet and is running 42c fully loaded.

It's all a matter of chance whether you're affected right now until someone can find exactly what the cause is and fix it. It might not even be a stability issue as lots of problem processors are not even overclocked.

[nikhsub1]

Originally posted by Audioaficionado
I ran my old XP1700 DLT3C cranked up to 215/2260 all summer long at 55c and had no FAH SSE problems.

This Barton isn't really overclocked yet and is running 42c fully loaded.

It's all a matter of chance whether you're affected right now until someone can find exactly what the cause is and fix it. It might not even be a stability issue as lots of problem processors are not even overclocked.
The funny thing is that all my OC'd T-Breds crash LESS than the ones running stock. The farm for example, all stock 2400's in an airconditioned room (24/7 AC) and these will lock like nobody's business and temps are under 40C load, all with fortron 300W PSU's etc. The OC'd machines really seem to crash less often.

[Audioaficionado]

Originally posted by nikhsub1

The funny thing is that all my OC'd T-Breds crash LESS than the ones running stock. The farm for example, all stock 2400's in an airconditioned room (24/7 AC) and these will lock like nobody's business and temps are under 40C load, all with fortron 300W PSU's etc. The OC'd machines really seem to crash less often.

I guess that means we all need to overclock to the max :D

[dz]

I wish I could. Last time I tried this thing at 1.75 Vcore it wouldn't boot at all when cold.

[baru]

yeah, ive been having problems in the last day. rig keeps locking up when i run it with the -forceasm flag. i guess i'll just finish off the unit with only -advmethods. it's painful... all that production lost. but i didn't have too many problems with my 2400 locking up before that. hopefully it's just the wu.

[walaka7]

well my dually just finished the exact same project that horked it the first time (same run/clone/gen) with no issues, and hasnt had a problem since that one time. I have changed absolutly nothing. no clock changes, no bios changes, i have not even opened the case or even changed the clients all i did was restart the clients and teh system once (teh system restart had nothing to do with F@H, just some config changes for network) Go figure

[baru]

finished that work unit i was having trouble with... it's odd, i tell you. i was having trouble with a project 344 wu, and now i'm almost done another pj 344, with no problems so far. **knocks on wood** must have been a bad wu or something....we'll see.

[Arkaine23]

I had a few reboots yesterday, but then it was also pretty hot and my cpu was up to 58C!

[Audioaficionado]

356 p344_gnra_unf_1 SSE on both my Barton and the Shuttle XP1700. So far, so good.

Barton is 42c and system 25c.