How many LinX loops do you run?

How many LinX loops do you use to determine stability?

  • 20 - I do the default, because the default is always best.

    Votes: 18 52.9%

  • 50 - I do a little more than suggested.

    Votes: 3 8.8%

  • 100 - I believe in stability.

    Votes: 4 11.8%

  • 200 - My stability is better than your stability.

    Votes: 2 5.9%

  • 500 - I am [H]ard!!!

    Votes: 4 11.8%

  • I don't use LinX because my OC isn't stable.

    Votes: 3 8.8%
  • Total voters
    34
Status
Not open for further replies.
Zoson

Zoson

Supreme [H]ardness
Joined
Aug 4, 2001
Messages
6,113
I'm assuming everyone does All available memory.
I used 100 loops, then just recently switched to 200 loops because my 4675MHz overclock would fail on loop 156 every time, while my 4400MHz overclock passes.

 
Why not just run it for 1000 loops, or let it loop constantly 24/7? I'm sure somewhere down the line at loop 2,834,945,107,383 it'll fail and you'll have to wait another 6 months to see if your OC is stable. *rolls eyes*

Really now. Anything over 50 is excessive. There's no point to be made by testing that much. No one taxes their system that hard.
 
HydroBudz said:
Why not just run it for 1000 loops, or let it loop constantly 24/7? I'm sure somewhere down the line at loop 2,834,945,107,383 it'll fail and you'll have to wait another 6 months to see if your OC is stable. *rolls eyes*

Really now. Anything over 50 is excessive. There's no point to be made by testing that much. No one taxes their system that hard.
Click to expand...

I can't agree with this mainly because the failure was so predictable. LinX was doing something in its 156th loop that was different, or the cumulative effect of running up until then caused the failure(perhaps by cooling limit). It's only a matter of time before I find something that does the 'same' thing and crashes my machine.

I've consistently seen failures after the 50th loop due to the fact that my watercooler takes longer than 50 loops to reach peak load temps. And also because long bouts of gaming without testing past that seemed to yield random crashes.

The bottom line is that I don't want my computer to crash for any reason related to hardware instability, unless that piece of hardware is actually failing.
 
HydroBudz said:
Really now. Anything over 50 is excessive. There's no point to be made by testing that much. No one taxes their system that hard.
Click to expand...

+1 here.

Zoson said:
I've consistently seen failures after the 50th loop due to the fact that my watercooler takes longer than 50 loops to reach peak load temps. And also because long bouts of gaming without testing past that seemed to yield random crashes.
Click to expand...

What are you doing, playing Crysis at max settings for 36 hours strait? Maybe you need to add a new rad to your loop.
 
I just run 200 or 250 and let it go while I'm at work. Come home at lunch and after I'm done at work to see if it's dropped yet. I did find a setting that seemed stable, but when I would let it run, it would make it about 50 passes and fail. Besides, it gives me a lot of passes to compare and see if I may be able to drop voltage or if I need to up it.

Also, I use either 24000 or 25000 for the runs to at least have some kind of consistency when comparing settings since I use LinX to fine turn my voltages. The GFlops variance will get smaller and smaller the closer you become to target voltages for at least x58.
 
I do 50, and then I run Folding@Home 24/7. If IBT/LinX doesn't catch any instabilities, F@H will eventually.
 
i selected 20.. but i usually dont do more then 10.. woot for ADD and the fact that i run boinc 24/7 on all my systems.. so if boinc fails.. then i know its not stable, lol..
 
Wow, does everyone take such a horrible stance on stability nowdays?
If it crashes it's not stable...
The point of testing stability is so that F@H won't have to catch a bad work unit, and waste all your CPU time that was spend doing the work...
 
I don't run it at all, it never passes more than 4 loops. Do I care? Nope. My system has never crashed during normal usage. Been that way for almost a year and a half.
 
criccio said:
I don't run it at all, it never passes more than 4 loops. Do I care? Nope. My system has never crashed during normal usage. Been that way for almost a year and a half.
Click to expand...

So pick the last poll option and hit vote.
 
I don't run LinX because I have better shit to do than let my computer heat up my entire house then go on forums and brag about how long I can run LinX.
 
Last edited:
RADEoN said:
I don't run LinX because I have better shit to do than let my computer heat up my entire house then go on forums and brag about how long I can run LinX.
Click to expand...

Nice edit, I saw your original post.

Way to make excuses. Bottom line is if you're not testing with LinX or IBT(which both use Linpack)... Your stuff probably isn't stable and the second you do something that actually taxes your machine, it will crash.

Not to mention that since Linpack is crossplatform, you can compare your results on different OS's.
 
Zoson said:
Way to make excuses. Bottom line is if you're not testing with LinX or IBT(which both use Linpack)... Your stuff probably isn't stable and the second you do something that actually taxes your machine, it will crash.
Click to expand...

Not true. I can't run these stress tests without my machine failing eventually, but like I said above, during normal usage ( x264 encoding, gaming, etc...) it hasn't ever crashed and I built it over a year ago.
 
Zoson said:
Wow, does everyone take such a horrible stance on stability nowdays?
If it crashes it's not stable...
The point of testing stability is so that F@H won't have to catch a bad work unit, and waste all your CPU time that was spend doing the work...
Click to expand...
Which is why I test with Linpack before I run F@H :rolleyes:. Maybe you should actually read my post.
 
Zoson said:
Nice edit, I saw your original post.

Way to make excuses. Bottom line is if you're not testing with LinX or IBT(which both use Linpack)... Your stuff probably isn't stable and the second you do something that actually taxes your machine, it will crash.

Not to mention that since Linpack is crossplatform, you can compare your results on different OS's.
Click to expand...

Here's my edit in less words:

The creator of LinX admitted he's new to programming, and apologizes for any bugs.

Link

Note that I'm rather new to programming, so there may be some bugs in there. Should you find one, report it and I'll do my best to fix it.
Click to expand...

Doesn't seem like it would be too great at checking stability...
 
I don't use LinX as much as Prime95, but when I did I use 20 runs.

I would love to know what, other than maybe F@H, you could possibly do that would equate to the same stress level as running LinX 100+ times.

I'm with the cynics, I think people who run it so much for stability purposes do it because they either need to know it is rock stable, or so they can brag about it on the forums. Unless you are encoding video for 12 hours at a time or playing multiple clients of Crysis simultaneously, you aren't going to be simulating the workload of LinX.

Maybe the question is what does stable mean to you? Stable by LinX standards? It's all arbitrary.

But hey, do what makes you feel good.
 
RADEoN said:
Here's my edit in less words:

The creator of LinX admitted he's new to programming, and apologizes for any bugs.

Link



Doesn't seem like it would be too great at checking stability...
Click to expand...
So if you don't like LinX, use IntelBurnTest instead. It's based on Linpack as well. Also, these programs are just Linpack wrappers at heart, so when it comes to the actual stress testing, it's Intel's code that's doing the work.
 
Zero82z said:
So if you don't like LinX, use IntelBurnTest instead. It's based on Linpack as well. Also, these programs are just Linpack wrappers at heart, so when it comes to the actual stress testing, it's Intel's code that's doing the work.
Click to expand...

hehe if you remember from my firefox thread, there's nothing that can stress my computer to instability except nfl.com
 
So part of my job is working with Mathematica, crunching raw data.

Mathematica linear matrix manipulations wreck your machine HARDER than Linpack. You can see in my screenshot that over 200 loops I peaked out at 87C. With Mathematica I hit 94C.

It's quite simple to simulate. Just download mathematica, and multiply a 1000x1000 matrix with another 1000x1000 matrix.

So yes, people that actually USE their machines to the full potential of the hardware *need* Linpack stability. It SUCKS to have a calculation that takes HOURS crash your machine and then you have to start all over.

Oh, and yes. What Zero82z said. The actual STRESS CODE is not written by the LinX creator. All it does is take the output from Linpack, and compare to see if it's exactly equal to your previous output. If not, your computer couldn't do the math correctly and it's unstable. So there's no reason to doubt its effectiveness. Not to mention IBT is exactly the same thing, but with fewer options and a not as nice interface.
 
Wellll back on topic, I do 20 because thats default :D I'll try 50 when I go to sleep, but i hate not seeing BOiNC crunching away!
 
Zoson said:
So part of my job is working with Mathematica, crunching raw data.

Mathematica linear matrix manipulations wreck your machine HARDER than Linpack. You can see in my screenshot that over 200 loops I peaked out at 87C. With Mathematica I hit 94C.

It's quite simple to simulate. Just download mathematica, and multiply a 1000x1000 matrix with another 1000x1000 matrix.

So yes, people that actually USE their machines to the full potential of the hardware *need* Linpack stability. It SUCKS to have a calculation that takes HOURS crash your machine and then you have to start all over.

Oh, and yes. What Zero82z said. The actual STRESS CODE is not written by the LinX creator. All it does is take the output from Linpack, and compare to see if it's exactly equal to your previous output. If not, your computer couldn't do the math correctly and it's unstable. So there's no reason to doubt its effectiveness. Not to mention IBT is exactly the same thing, but with fewer options and a not as nice interface.
Click to expand...

Fail on your part for using an OC'd machine to crunch data in a work environment. Buy a faster processor and leave it stock. :rolleyes:
 
After finding an overclock point that's stable via a simple hour of LinX testing, I essentially test for as long as I use a machine at most during a day. My longest days on a machine doing intensive cpu work is generally 12 hours, so I test for 2 days at 12 hours each day, and if it passes, then I'm golden. You only need to test once, then you know your machine is relatively stable for the year or two that you use it.
 
HydroBudz said:
Fail on your part for using an OC'd machine to crunch data in a work environment. Buy a faster processor and leave it stock. :rolleyes:
Click to expand...

Then you noticed I have an E8600 and there is no core2 faster. Then you also remembered that these things are linear and highly clockspeed dependant, so i7 or c2q doesn't do squat for me.

http://reference.wolfram.com/mathematica/GUIKit/tutorial/Threads.html
the Mathematica kernel is not multithreaded, the only reason for attempting a multithreaded interface definition is to allow user interface updates to occur while a long calculation may be proceeding in Mathematica.
Click to expand...
 
Last edited:
Sucks for you then. But to say an i7 will not crunch data faster than a C2D, regardless of clock speed, is wrong. I'd put money on an i7 920 at 3.8 would crunch your mathematica data faster than your E8600 clocked to 4.4.

Personally I wouldn't rely on an OC'd machine to crunch important data though.
 
HydroBudz said:
Sucks for you then. But to say an i7 will not crunch data faster than a C2D, regardless of clock speed, is wrong. I'd put money on an i7 920 at 3.8 would crunch your mathematica data faster than your E8600 clocked to 4.4.

Personally I wouldn't rely on an OC'd machine to crunch important data though.
Click to expand...

Actually, mathematica is one of the few things that the C2 beats the pants off the i7. 20% on average.
http://ixbtlabs.com/articles3/cpu/intel-ci7-920-p2.html scroll down to the scientific performance area. You'll notice it's directly clock dependant. IPC, thread count, and core count have little bearing.

And the reason why I stress to 200 loops is because this is a machine I do work on and it needs to be correct...
 
Last edited:
I test lightly while I am awake 5-20 loops and when I go to sleep I set it for 100. If it passes I consider it stable enough for anything I want to do.
 
Ok, so to summarize because you do professional work that exceeds the number crunching needs of 97% of users (my statistic) doing 200 loops makes sense for you.

What is dumbfounding to me is that you posted on the third post of the thread criticizing someone else for just running it 50 times, knowing they aren't your colleague at Mathematica.

The vast majority of users don't need to run it 200 loops. If you do, good for you!
 
I just wish there was an option that said "I don't use LinX because I don't give a shit, I'm satisfied with prime95 and windows tasks running completely fine, at least in vista, and games run good too!"

Not to mention the cheesy lines after each number.

If I had an issue where it kept messing up on the exact same loop, I'd blame software coding before hardware.
Posted via [H] Mobile Device[/
size]
 
Sojuuk said:
I test lightly while I am awake 5-20 loops and when I go to sleep I set it for 100. If it passes I consider it stable enough for anything I want to do.
Click to expand...

I do this a few times myself to setup overclocks. I used <20 loops to set the baseline oc, then I let it rip for a while when I am at work, sleeping, or gone for most of the day. then I back off the multiplier a bit to get a nice even numer and leave it there.
Posted via [H] Mobile Device
 
For the record:

I don't do F@H
I don't do it for bragging

I like my system stable 24/7 so I can go in and out of hibernation without weird problems but still have a system that will give me a boost in gaming.

I only run so many loops cause I work a 8-6 job and might as well test for stability then since my ADD prevents me from doing it in the evening. The plus side to this is that if there are problems, running for 3+ hrs usually finds it and my system ends up being very stable. The down side is the electric bill I just got. I about fell out of my chair when I opened it up.
 
RADEoN said:
I just wish there was an option that said "I don't use LinX because I don't give a shit, I'm satisfied with prime95 and windows tasks running completely fine, at least in vista, and games run good too!"

Not to mention the cheesy lines after each number.

If I had an issue where it kept messing up on the exact same loop, I'd blame software coding before hardware.
Posted via [H] Mobile Device[/
size]
Click to expand...


+1

The poll maker conveniently made the last option "...because my computer isn't stable". Fuck that. At the very least you could have included "I don't use it because my computer's stability meets my standards and stress testing is not crucial to my work/regular usage/self-esteem".
 
Two runs of 500, because I like to be as sure as possible, and LinX's GUI doesn't have and option for 1,000 yet.

I have had LinX fail between 200 and 400 runs many times, so I do not consider anything less than 500 proof of stability.
 
Zoson said:
Actually, mathematica is one of the few things that the C2 beats the pants off the i7. 20% on average.
http://ixbtlabs.com/articles3/cpu/intel-ci7-920-p2.html scroll down to the scientific performance area. You'll notice it's directly clock dependant. IPC, thread count, and core count have little bearing.

And the reason why I stress to 200 loops is because this is a machine I do work on and it needs to be correct...
Click to expand...

If you multiply large matrixs a lot, you might look at matlab and the CUDA plugin, it should make a significant difference in your times.


HydroBudz said:
Sucks for you then. But to say an i7 will not crunch data faster than a C2D, regardless of clock speed, is wrong. I'd put money on an i7 920 at 3.8 would crunch your mathematica data faster than your E8600 clocked to 4.4.

Personally I wouldn't rely on an OC'd machine to crunch important data though.
Click to expand...
The fact you would put more faith in an untested stock chip verses a statistically verified overclocked chip is amazing.
 
vengence said:
If you multiply large matrixs a lot, you might look at matlab and the CUDA plugin, it should make a significant difference in your times.
Click to expand...
Unfortunately we don't have a matlab license. :( matlab is also multithreaded... Really significantly superior to mathematica in that respect. I've been arguing to get a matlab license for two years now with no result...
[dream]
...Oh what I could do with a c2q, my gtx275's and matlab...
[/dream]

Oh, and I've decided to ignore the kiddies that would say Linpack has a programming problem, and they don't need to stress test with it. They don't understand the meaning of stable. They are not worth our time, cut and dry.
 
Not Stable. No idea of how many loops run or core temp.

17939d1211931574-dude-shits-himself.jpg


Until a program is documented that tests every cpu instruction, function and register and the memory and the video and the I/O all anyone can say is something along the lines of "The machine is stable doing xxxxxx under the enviromental conditions of yyyyyyy. And if thats good enough for the user then that is good enough for the user.

IMO the only absolutely 100% stable cpu I would trust would be sitting in a test fixture in Intels QC lab with "PASS" on the monitor.

But carry on, just keep it civil.
 
Last edited:
Zoson said:
Unfortunately we don't have a matlab license. :( matlab is also multithreaded... Really significantly superior to mathematica in that respect. I've been arguing to get a matlab license for two years now with no result...
[dream]
...Oh what I could do with a c2q, my gtx275's and matlab...
[/dream]

Oh, and I've decided to ignore the kiddies that would say Linpack has a programming problem, and they don't need to stress test with it. They don't understand the meaning of stable. They are not worth our time, cut and dry.
Click to expand...
Yeah, I mean who really cares if your computer adds 2+2 and gets 5 sometimes as long as it doesn't crash?
 
BillParrish said:
Not Stable. No idea of how many loops run or core temp.

Until a program is documented that tests every cpu instruction, function and register and the memory and the video and the I/O all anyone can say is something along the lines of "The machine is stable doing xxxxxx under the enviromental conditions of yyyyyyy. And if thats good enough for the user then that is good enough for the user.

IMO the only absolutely 100% stable cpu I would trust would be sitting in a test fixture in Intels QC lab with "PASS" on the monitor.

But carry on, just keep it civil.
Click to expand...

Hi Bill, while I can agree with you saying that my cpu cannot be considered 100% stable based on the criteria above. I think the basic criteria is flawed. The reason being that under this criteria - even the chip sitting in Intel QC would not meet the criteria.

I also think the idea that 'stable enough for me' is a load of crap... I can't believe for a single iota of a second that someone has a machine which cannot pass a LinX loop, but does not crash under moderately light workload.

So how do I define stability? You take the most intensive piece of testing software you can find(I don't think anyone that actually knows computer science will argue that Linpack is the best we have available right now), and test it beyond any time that you might load your cpu up to 100%

So if you do something that occupies your machine to 100% for 5 hours, you LinX test for 5 hours. If that's the worst YOU can do to it, your machine should be able to handle that, and then some - just in case. You never know when you're going to have an abnormally hot day.

This was the attitude I remember walking into here 10 years ago when I originally signed up. Your system wasn't stable until your eyes were bleeding from staring at prime95. Now it's just 'hey I can pass futuremark once, or this other synthetic benchmark once, it must be stable!' Which is just wrong and rather insulting to the entire community of overclocking. Whenever you hear someone giving overclocking a bad name, it's because they've had a run in with one of the noobs who thinks they don't need to stress test like a person with OCD.
 
BillParrish said:
Until a program is documented that tests every cpu instruction, function and register and the memory and the video and the I/O all anyone can say is something along the lines of "The machine is stable doing xxxxxx under the enviromental conditions of yyyyyyy. And if thats good enough for the user then that is good enough for the user.

IMO the only absolutely 100% stable cpu I would trust would be sitting in a test fixture in Intels QC lab with "PASS" on the monitor.

But carry on, just keep it civil.
Click to expand...
Well, considering IBT was designed by Intel......

Besides, you're talking (among other things) about things like ring back determining high-low voltages then it's a statistical thing. There is no non-empirical test and because it is an empirical test it means it is subject to statistics. Frankly, the 99.9999% is good enough for me.
 
vengence said:
Yeah, I mean who really cares if your computer adds 2+2 and gets 5 sometimes as long as it doesn't crash?
Click to expand...

funny story. I had a P4 2.4C running at 3.7ghz and it could not run calculator properly. EVERYTHING else was completely stable but calculator was useless.
 
Status
Not open for further replies.
Back
Top