Prime95 versus Folding stability

[KFB]

i am trying to get over 1.5GHz with my 1Gig Tbird, without a voltmod or watercooling [yet]. i can keep the idle temps around 40C and under Prime95 about 45C. folding maxes out at 48C. so far, folding (which is what i really care about) is stable, but P95 is reporting round off errors. any thoughts on why? i'm thinking not enough VCore voltage [i smell a voltmod real soon], but could it be memory timings or heat or PCI cards crapping out? has anyone else gotten a stable OC for Folding/general computing, but not for the P95 Torture Test?
thanks.

 

[Arkaine23]

prime95

So far I have not been able to run prime95 without errors at any speed/voltage, not even default. Usually it'll last 6-12 hours. At some point, usually when I'm not around, it'll just crash.

 

[Ploaf]

How long do you usually have to run prime to experience problems. I'm running the torture test right now without trouble.
I've had folding just up and quit when I OC'd too high. If that happened I would back it off. A lot.

 

[muddocktor]

Also, I think I remember reading at some forum, maybe hardocp, that the guy who wrote prime95 had released a version of prime95 around 3-6 months ago that had a flaw in it and was giving errors of some sort. This version was on the web for d/l for a little while and maybe you got it, instead of the older version or the newer, corrected version. I think that he had modified the program to run faster or something and his first revision had a flaw.

 

[Arkaine23]

prime probs

Yeah, I'll download a new copy and see if that helps. Its really disheartening to wake up and check the computer after maybe 6 hours of torture, only to find a black screen. At stock speed and voltage or even underclocked!

 

[KFB]

thanks for the replies.

ploaf: under the Torture Test, it will get a round-off error in the first or secnd Lucas-Lehmer iteration. what is weird is that it says:
round-off error (0.5) > 0.40
last time i checked, 0.5 IS greater than 0.4! unless the () means negative, then my comp is correct.

does anyone have a good link to the latest version of Prime95? i checked the web last night and all the links i followed were dead as hell. my current version is 19.1 ans the links said that 21.4 was out.

takiwa: are you sure about that? every other program besides P95 is stable on my rig. i think you are probably correct, but is there anywhere i can read up on it, that you know of? thanks.

arkaine: so you must say "screw P95." even stock speed gets errors? maybe i should slow my CPU down a bit and see if i still get errors. i'll be back....

 

[res0r9lm]

folding doesn't report errors like prime does...you are still getting them, though, and if you are sending your folding data in, it is corrupt, and will be useless for stanford to use...get prime 95 stable, and then fold your heart out! If prime detects a math error, then trust me, you ARE NOT STABLE to perform complicated mathmatical operations, like folding or SETI....

are you sure folding doesn't report error maybe you need to check your logs.

[14:06:10] Finished Work Unit:
[14:06:10] VerifyARCFile: Checksums don't match (frame 71). Failed verification
[14:06:10] Error: ARC file integrity could not be confirmed. Exiting
[14:06:10]
[14:06:10] Folding@Home2 Core Shutdown: BAD_FRAME_CHECKSUM
CoreStatus = 70 (112)
+ The core could not validate the current work unit for processing.

Deleting current work unit & continuing...
+ Attempting to get work packet

 

[Ploaf]

thanks for the replies.

ploaf: under the Torture Test, it will get a round-off error in the first or secnd Lucas-Lehmer iteration. what is weird is that it says:
round-off error (0.5) > 0.40
last time i checked, 0.5 IS greater than 0.4! unless the () means negative, then my comp is correct.

I've seen () used a lot to mean negative, but then I read a lot of stock reports. There's been a lot of negatives the last couple of years
Anyway. THis is where I dl'd it http://www.mersenne.org/freesoft.htm

I didn't have any error with it at all running the torture test so it must be good.

 

[KFB]

well Folding verifies ARC file integrity fine. maybe my WUs have mistakes in them, though, and i don't want that.
ploaf, thanks for the link. version 21 gets errors faster than the v19 i had. i am going to get this figured out so that i know my machine is 100% stable and error-free. after all, what good is a good OC if the CPU is doing FPU operations like the original Pentiums?! nothing less than flawless will do.

 

[muddocktor]

Well, maybe heat is getting to your overclock; 48° C is a little too hot for my t-bird to handle comfortably. I presently run it at 1400; used to run it in the 1450 range but it took a lot more vcore to keep it stable and it generated a lot of heat. I backed down to 1400 and I've been able to back the vcore down to 1.775 volts and totally stable. My t-bird is an AVIA 1200, which is a little older stepping than yours and it doesn't o/c as well as the newer steppings.

 

[KFB]

all i had to do was back it down to the speed that was stable for about 2 months - 1496
i really wanted to break the 1.5 barrier but 1496 is so close that i think i can call it 1.5

 

[Ploaf]

all i had to do was back it down to the speed that was stable for about 2 months - 1496
i really wanted to break the 1.5 barrier but 1496 is so close that i think i can call it 1.5

1.5 is fair enough.
Did it stop producing math errors at 1.5?

 

[KFB]

i actually ended up going to 1466 (133x11). no errors at all and i have had folding and prime95 torture test running concurrently for several days now. i posted this thread about an article which really helped me understand how Vcore increases help overclocking. definitely an undervoltage problem with my errors above 1.5GHz. i will voltmod real soon (i just don't want to fry my precious 'Bird!)

 

[res0r9lm]

i actually ended up going to 1466 (133x11). no errors at all and i have had folding and prime95 torture test running concurrently for several days now. i posted this thread about an article which really helped me understand how Vcore increases help overclocking. definitely an undervoltage problem with my errors above 1.5GHz. i will voltmod real soon (i just don't want to fry my precious 'Bird!)

the problem of raising vcore past 1.85 is cooling becomes an issue. I have found 1.92 to be a good compromise and 1.95 is getting close to the limits of aircooling.

 

[KFB]

res0r9lm: cool (intentional pun). watercooling will be mandatory when i really maximize my OC on my current rig. i wish i had a drill press.

 

[BBigJ]

I wouldn't worry too much about getting errors after 6 hours or so. They have to add a certain amount of randomness to the program to simulate the effects of thermal fluctuations, so the occasional error is just raising the effective temperature of the simulation. Not a big deal. However, if you are getting errors every hour or less, I would definitely back it down (or play with the voltage.) Fortunately folding doesn't require the accuracy of prime searching. I only test p95 for a few hours (~6) because I don't want to sacrifice that much folding time.

SBs