LinX - A simple Linpack interface

[janolle]

What you describe is probably my fault completely, but I'm not sure yet. I'd be grateful if you could try this version and report if you still get that strange behavior or not. Same goes to all having this issue. (Should be a lot of people )

Tried a few runs and it seems to work perfectly. Thx!

[GigaByte]

Linpack 10.2 is out on Intel's site if you find it worthy to update to as I have no idea what the differences are between 10.1 and .2.

[Dua|ist]

janolle, great news, thanks for testing!

GigaByte, I'll look into that, thanks for the news.

[stasio]

New in Intel MKL 10.2
This release of Intel Math Kernel Library (Intel MKL 10.2) provides
optimized multithreaded performance for the newest Intel®
processors, especially the Intel® Xeon® 5500 processor.
Performance Improvements
Performance improvements cover several key math routines including
LINPACK, Out-of-core PARDISO, BLAS, and FFT. In addition, Intel
AVX (Advanced Vector Extensions) support is included for advanced
vectorization being introduced in upcoming Intel® architecture
processors. This provides support for 256-bit vector operations, in
many cases doubling performance. These are provided earlier to test
and develop forward scaling in your applications.
C#/.Net support
Intel MKL 10.2 introduces C#/.Net support examples for calling Intel
MKL functions.
BLAS
Introduce better coverage of threading in level 1 BLAS routines, in
addition to the threading for level 2 and level 3 BLAS routines.

PDF
Download

[devsk]

Dua|ist, do you plan to do a native 64-bit GUI? I was trying to see if I could run it in wine in Linux and I can use 32-bit wine to run 32-bit Linx which is not really useful for me. I tried to run it with 64-bit wine and it failed because it didn't know what to do with 32-bit LinX.exe (Trying to load PE image for unsupported architecture (I386)). If the GUI becomes 64-bit native, then I don't need to write one for Linux.... I can run it under wine.

[Dua|ist]

Dua|ist, do you plan to do a native 64-bit GUI? I was trying to see if I could run it in wine in Linux and I can use 32-bit wine to run 32-bit Linx which is not really useful for me. I tried to run it with 64-bit wine and it failed because it didn't know what to do with 32-bit LinX.exe (Trying to load PE image for unsupported architecture (I386)). If the GUI becomes 64-bit native, then I don't need to write one for Linux.... I can run it under wine.

Sadly, Delphi has neither an x64 nor a Linux compiler, so making any of these versions will be almost equally hard. But x64 should be theoretically simplier.

[oc_junkY]

Congrats for the work man!
Especially the "remaining time" function is cool :-)

[monza1412]

Im having problems if I run it at real time priority, even at stock speeds, windows just freezes a couple of seconds after the start.
Anyone else experiencing this (running on win7 64b RC/ 4gb Ram)

[123456iambelial]

How to use this?
Do I have to look at the time? Lower is better?
or
The highest GFlops peaks? Higher is better?

[Dua|ist]

Congrats for the work man!
Especially the "remaining time" function is cool :-)

Thanks!

Im having problems if I run it at real time priority, even at stock speeds, windows just freezes a couple of seconds after the start.
Anyone else experiencing this (running on win7 64b RC/ 4gb Ram)

Yes, this is kind of normal behaviour. Consider reading the hint over the Priority setting.
I personally wouldn't set priority to anything over above normal, as it only makes the whole OS almost freeze (or completely freeze, as in your case).
In 0.6.1 (which is currently getting its last polish) LinX will have same priority as Linpack, so at least it won't freeze and you'll be able to stop it. For now stick with lower priorities.

How to use this?
Do I have to look at the time? Lower is better?
or
The highest GFlops peaks? Higher is better?

Well, you just test it, the more GFlops the faster is your system, on higher problem sizes/memory values you're likely to get higher GFlops and higher load and higher temps.
Linpack was designed as a benchmark, thus the time per test and GFlops values. If you're just stress testing/stability testing, you may simply ignore them.

[monza1412]

Thanks!


Yes, this is kind of normal behaviour. Consider reading the hint over the Priority setting.
I personally wouldn't set priority to anything over above normal, as it only makes the whole OS almost freeze (or completely freeze, as in your case).
In 0.6.1 (which is currently getting its last polish) LinX will have same priority as Linpack, so at least it won't freeze and you'll be able to stop it. For now stick with lower priorities.

roger that. Great and very useful program by the way

[Boogerlad]

hmm, I've got an error. It says "linpack has stopped unexpectedly, press the " " button for details." Where is that button? I can't find it. I'm sure my oc is unstable, but I can't find that button.

[SimpleTECH]

hmm, I've got an error. It says "linpack has stopped unexpectedly, press the " " button for details." Where is that button? I can't find it. I'm sure my oc is unstable, but I can't find that button.

Bottom right-hand corner.

[Boogerlad]

Thanks.

[Dr.Paneas]

I am using Vista 64 Bit. Which version of Linx should I run ?
I want to test my CPU for stability purposes only. Are there any settings that I need to change ? ( such as Problem Size, Priority from TaskManager etc).

Thank you

[Dua|ist]

hmm, I've got an error. It says "linpack has stopped unexpectedly, press the " " button for details." Where is that button? I can't find it. I'm sure my oc is unstable, but I can't find that button.

That button has a long story of renaming. This'll be fixed, but it is the one in the right part of the statusbar, as already mentioned.

I am using Vista 64 Bit. Which version of Linx should I run ?
I want to test my CPU for stability purposes only. Are there any settings that I need to change ? ( such as Problem Size, Priority from TaskManager etc).

Thank you

Run Linpack 64-bit, it is somewhat more stressful and it's default on LinX launched on a 64-bit OS. I'd say, keep priority at Normal, unless you're using your computer during testing, then you might lower it to reduce the impact on the other apps you're running.
As for Problem Size/Memory, well, larger values are better, but they also need more time to complete. Play around with them, find the ones that suit you best or just use the all mem option for maximum stress (if it's gonna work for you ).

[Brama]

I found a strange issue with last LinX.

Using Win7 x64 7100 official.

Processor i7 975 8 virtual cores, 6 Gb RAM.

Set 4 cores used in LinX and 5300 Mb of RAM.
It starts correctly, I see with task manager that occupies correctly all the memory and that the average load of the 8 cpu is 50%.
After some minutes I see that every cpu go near 0% load and I expect to have the first result on LinX.
No way, the cpu remain near 0% of load, the memory taken by LinX stays at 5300 MB and nothing happens. I waited for 10 minutes and after I shut down the program that is not blocked as it accept every command.

I tested same setup with last IBT and it worked as expected.

[003]

Will LinX be updated to the newer linpack libraries?

[Fritz_the_germ]

Does anyone have a link for the lastest linpack version?
The 6.0. has a Vista64 bug.
There should be a .6.0.2 out but I cannot find a site

[Dua|ist]

Alright. First of all, let me introduce a newer version: LinX 0.6.1

Main changes:

- Linpack files updated to version 10.2
- main algorithm rewritten, might improve performance a bit
- fixed an issue with inconsistent GFlops values in some cases
- hotkeys for Start and Stop buttons added: Enter and Escape keys respectively

Will update the 1st page & main mirror later.

The new 10.2 binaries were reported to have lower GFlops numbers, but good thing is they don't require external dlls. If you're concerned about GFlops, take 32-bit and/or 64-bit folders from previous LinX releases and put them into LinX folder (don't forget to delete the newer 10.2 Linpack files). But I'm sure this didn't affect the testing efficiency in any way.

Brama,
are you talking about the 0.6.1 version? Anyway, after the load of all cores went to 0% (obviously 1st pass finished as you mentioned), did it rise again (2nd pass started)? Does this issue happen with all 8 cores/some other problem size/memory values?

This new LinX uses 2 threads now with 2nd thread being idle most of the time to further lower the impact on the running Linpack. This is (imo) a big step forward, but might introduce new bugs.

If anyone else is having issues, please report them.

[El Greco]

The 6.0. has a Vista64 bug.
There should be a .6.0.2 out but I cannot find a site

LInx 0.6 worked fine with my Vista 64 business.. 20 to 25 pass @ maximum stress

[Fritz_the_germ]

Soory El Greco I cannot agree. Myself and many other friends of mine had a bluesreen after the second loop using an 64 bit OS.

The version 6.1 works just fine:



I know 3 loops are about nothing but I will run 25 loops today.

@Dual|ist,
thanks for your link

[El Greco]

Wait a minite...you say a bluesreen after the second loop ?? Even @ default speed???

[Fritz_the_germ]

Nop, by running my oc speed. But look at this now:

[Fritz_the_germ]

Yoh guys,
using LinX0.6.1 makes me wonder why I am having different GFlops results by using the same settings.
Normally I select "All" (memory) on Vista 64 Bit and I get about 52,xxx GFlops in hardly a minute. But sometimes when I select "All" I receive about 29,xxx GFlops in a minute and about 30 seconds )on the very same OS).



Why do I have so large differences using the same settings and the same OS?