Venice hates Prime95.

[perry_78]

I'm having fun with the SP2004 version of prime. After some time, the GUI disapears and the time priming goes up by thousands of hours in the taskbar...

Not to mention random re-starts with 100% stable G.skill LAs, even with 2T and bank interleaving enabled...

[Gautam]

My A64 wasn't even prime stable at stock. Didn't stop me from running 2.7 gigs daily though.

[One_Hertz]

if it is not prime stable doesnt that mean your cpu will degrade quicker over time and die a premature death?

[Gautam]

Yeah, might take only 7 years at that rate. :p

[saratoga]

if it is not prime stable doesnt that mean your cpu will degrade quicker over time and die a premature death?

Not really. It could mean you have bad RAM, or some other thing not related to the CPU. Mainly it just means that every so often a certain number of operations return the wrong value for some reason. So in effect the code asks the CPU what 1+1 is and gets 3 or some other incorrect value.

If it doesn't happen all that often it could take days or even weeks to accumulate enough mistakes to crash the computer, since most CPU cycles don't go to the OS and of those that do, most don't really matter (IE would you even notice if theres a mistake reading the input from the mouse and the OS calculates you moved 1 pixel to the left of where you really did? Or if the counter that keeps track of time loses a single milisecond?). But the CPU is still not working properly, which can be really bad for processing data that depends on accurate calculations (any form of distributed computing, some codecs, etc).

[dnottis]

Vcore dip at the wrong time on the 15min run most likely.

I agree with ya there - but I'm been logging using Speed fan and I'm not seeing drops in the logs. Also, I've bumped vcore a bit higher and it will still be inconsistent.

[frostedflakes]

Wait so prime indicated that your CPU doesn't do math properly, but you're folding with it anyway? Can you think of a reason turning in bad data to a distributed computing project is a terribly mean spirited thing to do?

I don't see what the problem with this is - the F@H client doesn't pick up any errors. If the program receives a value that's known to be incorrect, the core will shutdown and send the data that has been collected up until that point back to Stanford.

Also, do you really think Stanford would be stupid enough to send a particular WU to only one person? Hell no. They expect that a few of the WUs sent out may come back with errors, or for whatever reason may not get done, so clones are sent out to multiple computers.

But like I said, F@H hasn't run into any errors...

[Ubermann]

Seems like prime is ~100mhz stable below 32m SuperPi.
Should i say its primes fault and say its stable at 32m Superpi anyway ? no..
I noticed if you play much online and comp isnt 100% stable then ping usualy jumps up and down and makes me die.
Download pingplotter from http://www.pingplotter.com/download.html and set it to 1sec interval.
Then start clockgen at stable clock.
Raise 1fsb at a time from that and watch pingplotters graph.
Its amazing.

[MUCHO]

ftp://mersenne.org/gimps/p95v246.zip

Supposed to be a beta version of Prime that takes care of problems associated with the Venice core.

For all I know it could just be a trojan but I found it in this thread...

http://www.xtremesystems.org/forums/...ght=prime+beta

I have d/l it and tried it but I've not tried out my Venice yet...

[dnottis]

Looky what I found - dated June 9th 2005...

ftp://mersenne.org/gimps/p95v2412.zip

Dean

[ReelMonza]

Now let's just wait a StressPrime update...

[TheMeatFrog]

There are levels of stability. Do you want your computer to be stable enough to cruch numbers all day and night? Then prime stable may be necessary. If all you do is game then get your rig game stable. The level of stability you need is determined by the functions you use most.

Holla back! Best post of the entire thread!

[saratoga]

I don't see what the problem with this is - the F@H client doesn't pick up any errors. If the program receives a value that's known to be incorrect, the core will shutdown and send the data that has been collected up until that point back to Stanford.

How would the client run into errors? Its not like it runs every WU twice before handing it in. Unless your machine is so unstable that it crashes, the client will dutifully compute the incorrect results, and send them in.

Also, do you really think Stanford would be stupid enough to send a particular WU to only one person? Hell no. They expect that a few of the WUs sent out may come back with errors, or for whatever reason may not get done, so clones are sent out to multiple computers.

Any data sent in is compromised, so it will hopefully be discarded. At best you're not actually helping the project (and wasting their time/bandwidth). At worst, you could be hurting the project (two people both send in bad data).

[blackjok3r]

Well Ive got that venice version of prime and managed this:

[HousERaT]

anyone using this new (Venice friendly) version of prime? Does this mean I have to wait for a San Diego friendly version too?

[Paa`]

Im sure ive said this before but these "version's of prime" aint friendly to certain cores.

The only difference ive noticed from old version to the so called winchester/venice version's is that all test's including ram use 100% cpu

I guess this is because the memory controller is on the cpu die now.

Unless u stress the cpu to 100% you wont ever fine your max ram overclock..

Hope that makes some sense :x

[dnottis]

Well Ive got that venice version of prime and managed this:

Did you have problems with other versions of prime tho?

[Damien]

So will this new "venice-friendly" prime be winnie-friendly as well? Im running 300x9 @ 1.55v, which fails prime within 10 minutes, but has yet to crash on me since I got it. Even fails prime at stock within 25 mins.

[DeNs]

I can run SuperPI 1M about 30mhz~ faster then what is "Prime stable", and i can run 80-100mhz (!) faster then "Prime stable" on the WinRar benchmark! go figure. Prime95 just seems to stress A64s alot.

I've always found that Spi is a lot more sensitive with my winnie than Prime is - I won't get through a 1M iteration test if this thing is just that little bit over its stable point (2550MHz). Thing is though it can prime for hours at those settings, eventually bombing out after anywhere between 20 minutes to 3 hours.

I always run a 1M then a 32M run, then run StressCPU for 12+ hours for complete stability. 24 hours if I don't need to use that PC

My P4 can run Spi at insane speeds, but can't hold a stable OC at some of the higher frequencies (4.1GHz+). It really varies.

---dens

[prime95]

Prime95 version 24.12 is not designed to be winnie-friendly or venice-friendly. I've made several optimizations to make finding primes faster. These changes will work the FPU harder. I'm not sure if they will put more or less stress on the cache and memory subsystems.

[Magnj]

my Claw will do 5 32M pimods in a row, loops of 3d01, loops of Frac test, Virus and Spyware scans, games, game stress tests. Multi tasking ect. But will not pass prime 24.12. Crashes in under a minute. but it has not restarted once, not one hiccup. So its good enough for me

[dnottis]

Prime95 version 24.12 is not designed to be winnie-friendly or venice-friendly. I've made several optimizations to make finding primes faster. These changes will work the FPU harder. I'm not sure if they will put more or less stress on the cache and memory subsystems.

At least it's not so finiky. I've run it a few times and it's at least consistent vs. older versions...

[Damien]

Just started this about 30 mins ago, still running. Before it'd fail within 15 mins, so this is a good sign (running 295x9 @ 1.55v atm)