LinX - A simple Linpack interface

[dnottis]

Hey Dua|ist,

Any thoughts on this particular issue?

http://www.xtremesystems.org/forums/...36&postcount=9

As soon as I have 2 GPUs in the motherboard LinX will crash at 21 or 38 seconds. If I pull it its stable for 20 passes. Does LinX stress the motherboard that much? I'm thinking its just a bad motherboard?

Any ideas would be appreciated. Thanks.

[Tekxpert]

Maybe you're right. Some people say Linpack is enough for both RAM and CPU testing, some prefer running additional, more specific tests like Memtest. If I pass Linpack at full or near full memory I'm usually stable in MemTest too, but I still prefer to run it just to be sure. This might be different on different rigs/processors/chipsets, etc., so it's better to decide for yourself.

It is of course good to have an all-in-one testing program. And even if Linpack isn't the best memory checking tool it is definitely not the worst one either.

Maybe we'll hear some other opinions here.

I agree with you totally. But personally I think Linpack running @ full stress eg. 1 value with the OS underneath in real time gives the best checking in my opinion. Memtest runs before OS loads so there is only a certain kind of I/O from MCH (at least on LGA775 boards) Its better to runs stress testing in as real world conditions as possible for most reliable results. In other words - when the OS is loaded and application(s) are running.

[Falkentyne]

The purpose of memtest86+ is to see if you can *load* the OS without instantly corrupting it. Without memtest available, there would be a LOT more corrupted OS's out there requiring reinstallation and overclocking would be nowhere as popular.

[jcool]

Funny thing is, for i7 this doesn't seem to apply. I can run memtest+ for hours at DDR3-2000 but I can't even load the OS properly with ram at 1600. It seems memtest focuses on the ram ICs and only fails if they are at fault, while the OS fails to load because the IMC isn't up to snuff.

[Dua|ist]

okorop
LinX is OK at memory testing. Not sure if it can be compared with MemTest, but it definitely does detect serious memory errors. And according to ReverendMaynard it even outperforms Memtest86.

Falkentyne
Well, MemTest is a nice tool, but the chance of corrupting the OS completely due to memory overclock is not that high. And, honestly, I don't think it had such a great impact on OCing's popularity.

jcool
i7 seem to be very different from previous CPU generations. No wonder that stress-testing apps act weird on them. I'm sure i7's specifics will be taken into account in newer stress-testing software.

[davidk21770]

So whats the verdict with regards to best memory stress tool, Linpack code, Prime 95 or Memtest??

I've been using non-gui Linpack for many months and never had an issue with my system OC. Don't see the point in running its competitors.

I can be 100 iterations stable in linpak, pass 300 trials of memtext-hcl and still fail prime95 blend. I now use 15 cycles of linpak as a quick test of CPU stability then move on to prime95 for the real testing.

[GFORCE100]

I can be 100 iterations stable in linpak, pass 300 trials of memtext-hcl and still fail prime95 blend. I now use 15 cycles of linpak as a quick test of CPU stability then move on to prime95 for the real testing.

What? Even with the "All" Memory to use option enabled in Linx?

[Tekxpert]

I can be 100 iterations stable in linpak, pass 300 trials of memtext-hcl and still fail prime95 blend. I now use 15 cycles of linpak as a quick test of CPU stability then move on to prime95 for the real testing.

Thats what I found with Linpack non-gui too. System can be error free for 15 runs but can't run Prime95 for more than 5 minutes without error. I know this debate about the best System OC stress testing tool has been regurgitated before many times on OC forums. But stability means different things to different users. At the end of the day if your system does'nt crash or lockup etc due to the applications you personally run when OC then its stable for you.

[okorop]

okorop
LinX is OK at memory testing. Not sure if it can be compared with MemTest, but it definitely does detect serious memory errors. And according to ReverendMaynard it even outperforms Memtest86.

Thank's a lot for the answer so if I tested the pc with linx I can be sure that also the rams are ok, great

[davidk21770]

What? Even with the "All" Memory to use option enabled in Linx?

Yes.

[MacClipper]

Funny thing is, for i7 this doesn't seem to apply. I can run memtest+ for hours at DDR3-2000 but I can't even load the OS properly with ram at 1600. It seems memtest focuses on the ram ICs and only fails if they are at fault, while the OS fails to load because the IMC isn't up to snuff.

Similar findings here, memtest86 used to be very useful in DOS testing o'c stability on earlier platforms but not i7.

[ChaosAD]

If i can run Prime without problems but when i run Linx i get a bsod on the first two minutes (using 64bit/full mem), what voltage shall i tweak? Max core temp is around 81C when it crash.

[Dua|ist]

If i can run Prime without problems but when i run Linx i get a bsod on the first two minutes (using 64bit/full mem), what voltage shall i tweak? Max core temp is around 81C when it crash.

Vcore, I'm afraid. But Nehalem owners might know better.

[Infa]

Could also be qpi/vtt voltage

but probably vcore

[omiez]

If i can run Prime without problems but when i run Linx i get a bsod on the first two minutes (using 64bit/full mem), what voltage shall i tweak? Max core temp is around 81C when it crash.

It is running too hot.

[Dua|ist]

Here you go, a new version: LinX 0.5.8

What's new:

- now it is not necessary to run a test for current settings to be saved
- added saving of the settings at the start of testing to prevent loosing them in case of BSOD or hang
- added an option to save Linpack's log during testing, after every completed test (decreases performance a bit when selected)
- graphs can now be shown during testing in real-time
- improvements and enhancements to graph windows, graph window positions and sizes are now saved on exit
- About and Information windows are now joined
- small fixes/improvements

Hope you'll like it.

P. S. There is also another mirror now, just in case: http://dl.getdropbox.com/u/663563/LinX/en/LinX.7z

[WaterFlex]

Well done, bro Awesome program !

[omiez]

Very nice prog. thank you.


The only question i have.

To test stability of the System must i set something once started the program or i just start it and only set the number of tests to run ?

It will be great if you explain if there are some settings to perform.

Thank you.

[Dua|ist]

Very nice prog. thank you.


The only question i have.

To test stability of the System must i set something once started the program or i just start it and only set the number of tests to run ?

It will be great if you explain if there are some settings to perform.

Thank you.

The most common settings that you might want to set manually are in the main window : Problem size and corresponding amount of memory and number of runs.

All other Linpack settings can usually be left as they are, LinX automatically enables 64-bit Linpack on 64-bit OSes and sets number of threads equal to the number of logical cores. Data Alignment and Leading Dimensions settings don't have much (if any) impact on testing/performance anyway, but by default they're also set to Intel's recommended values.

Advanced settings are there just to tweak consumed memory formula in case Linpack fails to start with no memory message, no need to tweak these if everything works ok.

Everything else is just LinX own settings. Their default values are also suitable for most cases.

There are hints to almost every setting, but if something is unclear, feel free to ask.

[slim142]

Hey Dualist

So basically now on 0.5.8 you can save the log (similar to what prime95 does) or is it something different? How bad is the impact? Do you mean less resources to use the PC for other things (surfing, listening music) while testing?

[omiez]

The most common settings that you might want to set manually are in the main window : Problem size and corresponding amount of memory and number of runs.

All other Linpack settings can usually be left as they are, LinX automatically enables 64-bit Linpack on 64-bit OSes and sets number of threads equal to the number of logical cores. Data Alignment and Leading Dimensions settings don't have much (if any) impact on testing/performance anyway, but by default they're also set to Intel's recommended values.

Advanced settings are there just to tweak consumed memory formula in case Linpack fails to start with no memory message, no need to tweak these if everything works ok.

Everything else is just LinX own settings. Their default values are also suitable for most cases.

There are hints to almost every setting, but if something is unclear, feel free to ask.

Thank You (Спасибо) !

Ok everything works fine, no changes made since first start.

What does Problem Size affect ?
What happens if i select Bigger or Smaller Problem Size ?

[Dua|ist]

Hey Dualist

So basically now on 0.5.8 you can save the log (similar to what prime95 does) or is it something different? How bad is the impact? Do you mean less resources to use the PC for other things (surfing, listening music) while testing?

It is just like in some previous versions, Linpack's text log can be saved during testing, so one can see how far it went in case everything hangs or throws a blue screen. I think it was davidk21770 who missed this feature.

No, the impact only affects GFlops numbers slightly. Shouldn't be noticeable at all (including error detection and heating effectiveness), but I've probably become too paranoid about those numbers during LinX' development while trying to minimize its impact on Linpack's performance.

What does Problem Size affect ?
What happens if i select Bigger or Smaller Problem Size ?

This is probably the most important Linpack's setting. It determines the dimensions of a matrix (or a linear equations system, whichever you like best) that is to be solved. The higher the Problem Size number the more CPU stress, more memory consumption, longer time to solve each test.

I really can not suggest you an universal value for it. Basically it is the higher the better, but in some cases you might want to run LinX with lower values of it, for example during initial OC attempts to avoid BSODs, etc. The default value of 10000 is, let's say, a recommended minimum. Maybe you should play with it a little and determine your own testing methodology.

[Slyr7.62]

Thanks.
Now overclock it and the difference in temps will be more evident.

It's OC'd, just not in that pic(was doing a lil stock clock undervolting ). When OC'd, LinX hits 48°C peak(usually 44°C). I just ran IBT for 30min which got me a max temp of 54°C. That's exactly 10°C higher than the 44°C OCCT gets my cpu to. I'm currently running OCCT at this moment, auto-mix test. OC'd to 3.6 (9*400) w/ 1.225V, 4x1GB KHX @ 1066 5-5-5-15 PL7(for now ) w/ 1.9V. Might drop the Voltage on 1 or both a notch or 2. Overclock hax ftw . . I have a good E8400 C0(and my KHX Micron D9GKX might help. ). My E8400 before this one was OEM(and used) off Ebay. That one could do 3.6 but needed maybe 1.235V (Also had a Ep35-DS3L w/ that proc so ).
OCCT Done!.
P.S. Within the week when I get time I'll do a 20+ run of LinX and SS it. (latest version, 0.5.8).

[Smartidiot89]

Again awesome tool

Using it on my Phenom II right now. I stopped using Prime95 a long time ago since that program is just a joke for stability, OCCT is ok but LinX really does it.

I don't consider a clock 100% stable unless it can pass hours of LinX at max RAM usage.

Keep it up I want to see this program becoming THE stability tester

[slim142]

It is just like in some previous versions, Linpack's text log can be saved during testing, so one can see how far it went in case everything hangs or throws a blue screen. I think it was davidk21770 who missed this feature.

No, the impact only affects GFlops numbers slightly. Shouldn't be noticeable at all (including error detection and heating effectiveness), but I've probably become too paranoid about those numbers during LinX' development while trying to minimize its impact on Linpack's performance.

I didnt know you removed that feature in recent versions. I also like it
Anyways, thanks for keeping it.