Linpack Xtreme Released

[AlwaysHope]

701.6337 GFlops average
View attachment 230395

Interesting, cpu package power max = 322w! what kind of cooling?

 

[Det0x]

Interesting, cpu package power max = 322w! what kind of cooling?

Cooling is TechN AM4 block with liquid metal on custom watercooling.
Thinking about doing a new run with cold air in a few days..

 

[AlwaysHope]

Wondering how accurate the GFLOPS measurement is in this app? I did some testing today with MSI's unique feature for applying auto volts to CPU with their "CPU lite mode" option in bios.

As you can see from the results, around runs 7-8 it drops off. This is with 10GB ram option.

 

[Regeneration]

Wondering how accurate the GFLOPS measurement is in this app? I did some testing today with MSI's unique feature for applying auto volts to CPU with their "CPU lite mode" option in bios.
View attachment 234569
As you can see from the results, around runs 7-8 it drops off. This is with 10GB ram option.

Accurate when you use benchmark mode.

 

[Octopuss]

Are there going to be any updates?

 

[Regeneration]

Are there going to be any updates?

When i get more free time on my hands. I have a girlfriend, wrist problems, and neighbor wars.

 

[Athlonite]

When i get more free time on my hands. I have a girlfriend, wrist problems, and neighbor wars.

Well atleast now that you have a GF the wrist problem should be alleviated significantly /Jk

 

[AlwaysHope]

When i get more free time on my hands. I have a girlfriend, wrist problems, and neighbor wars.

It would be greatly appreciated if you could take out of experimental, memory capacities higher than 10GB.
I think its fair to ask this considering most systems today have 16GB or more.

 

[freeagent]

It would be greatly appreciated if you could take out of experimental, memory capacities higher than 10GB.
I think its fair to ask this considering most systems today have 16GB or more.

How much do you hate your CPU?

 

[AlwaysHope]

How much do you hate your CPU?

Nah, if it can pass 10 runs with whatever component OC I throw at it, it'll pass NEARLY anything. Was OC testing couple days back & it passed LinpackXtreme but BSOD in CB! So throw every bit of code variety you can at an OC test.... just to be sure!

 

[freeagent]

Nah, if it can pass 10 runs with whatever component OC I throw at it, it'll pass NEARLY anything. Was OC testing couple days back & it passed LinpackXtreme but BSOD in CB! So throw every bit of code variety you can at an OC test.... just to be sure!

I usually set 8-10GB for 25 passes, I do hate my fans

 

[AlwaysHope]

I usually set 8-10GB for 25 passes, I do hate my fans

Fans?? tell me about them ..... Should hear mine! but it's only during extreme AVX testing that this app does, don't think you'll get that kind of intensity in most apps today unless your running very specific heavy AVX, especially AVX512.
Y-Cruncher is good for AVX512.

 

[freeagent]

Fans?? tell me about them ..... Should hear mine! but it's only during extreme AVX testing that this app does, don't think you'll get that kind of intensity in most apps today unless your running very specific heavy AVX, especially AVX512.
Y-Cruncher is good for AVX512.

NF-A14 iPPC 3K x2, NF-F12 iPPC 3K, TY-143

I don’t even hear the 143 anymore, it’s the quietest fan in my system

 

[Arctucas]

NF-A14 iPPC 3K x2, NF-F12 iPPC 3K, TY-143

I don’t even hear the 143 anymore, it’s the quietest fan in my system

I know what you mean, running 4 @ iPPC 140mm 3K, 7 @ iPPC 140mm 2K, 1 @ iPPC 120mm 2K.

All are connected directly to PSU, 100% all the time.

I rather enjoy the white noise.

 

[freeagent]

I have mine on a controller because I am soft

 

[Det0x]

730 GFLOPS average

 

[Solid State Brain]

Does anybody have a source for the statement according to which it's "normal" for Linpack residuals to occasionally mismatch on modern Intel processors?
I'm not convinced that this is accurate or representative of fully stable operating conditions.

EDIT: it looks like I get lots of residual mismatches, although the tests pass, if I enable AVX in BIOS and the problem size is relatively small (10000). If I disable AVX, all residuals match.

I tried increasing CPU voltage or decreasing RAM speed but this didn't affect the results, so it could be due to some quirk with AVX instructions. The system appears to be stable otherwise.


EDIT2: below are results with AVX disabled.

Note that I'm testing under Linux with an i7-12700K and limiting the test on certain cores with taskset (this is mostly for consistency with other tests I was doing to check whether some core was giving bad results):

user@linux:~/bin/LinpackXtreme-1.1.5/linpack-xtreme-1.1.5-amd64> taskset -c 0-8 ./run_stress_test
Linpack Xtreme v1.1.5 by Regeneration

Current date/time: Tue May 17 16:30:13 2022

CPU frequency:    4.987 GHz
Number of CPUs: 1
Number of cores: 12
Number of threads: 20

Parameters are set to:

Number of tests: 1
Number of equations to solve (problem size) : 10000
Leading dimension of array                  : 10000
Number of trials to run                     : 30
Data alignment value (in Kbytes)            : 4

Maximum memory requested that can be used=800204096, at the size=10000

=================== Timing linear equation system solver ===================

Size   LDA    Align. Time(s)    GFlops   Residual     Residual(norm) Check
10000  10000  4      5.773      115.5059 1.012665e-10 3.570760e-02   pass
10000  10000  4      9.378      71.1131  1.012665e-10 3.570760e-02   pass
10000  10000  4      6.873      97.0332  1.012665e-10 3.570760e-02   pass
10000  10000  4      8.773      76.0137  1.012665e-10 3.570760e-02   pass
10000  10000  4      8.950      74.5140  1.012665e-10 3.570760e-02   pass
10000  10000  4      9.465      70.4527  1.012665e-10 3.570760e-02   pass
10000  10000  4      9.311      71.6240  1.012665e-10 3.570760e-02   pass
10000  10000  4      9.066      73.5599  1.012665e-10 3.570760e-02   pass
10000  10000  4      9.207      72.4315  1.012665e-10 3.570760e-02   pass
10000  10000  4      8.854      75.3170  1.012665e-10 3.570760e-02   pass
10000  10000  4      9.003      74.0711  1.012665e-10 3.570760e-02   pass
10000  10000  4      9.471      70.4097  1.012665e-10 3.570760e-02   pass
10000  10000  4      8.897      74.9565  1.012665e-10 3.570760e-02   pass
10000  10000  4      8.588      77.6499  1.012665e-10 3.570760e-02   pass
10000  10000  4      9.467      70.4429  1.012665e-10 3.570760e-02   pass
10000  10000  4      9.414      70.8402  1.012665e-10 3.570760e-02   pass
10000  10000  4      9.259      72.0257  1.012665e-10 3.570760e-02   pass
10000  10000  4      9.220      72.3274  1.012665e-10 3.570760e-02   pass
10000  10000  4      8.072      82.6146  1.012665e-10 3.570760e-02   pass
10000  10000  4      9.346      71.3538  1.012665e-10 3.570760e-02   pass
10000  10000  4      9.228      72.2691  1.012665e-10 3.570760e-02   pass
10000  10000  4      9.230      72.2502  1.012665e-10 3.570760e-02   pass
10000  10000  4      7.142      93.3674  1.012665e-10 3.570760e-02   pass
10000  10000  4      8.868      75.1982  1.012665e-10 3.570760e-02   pass
10000  10000  4      5.569      119.7422 1.012665e-10 3.570760e-02   pass
10000  10000  4      9.056      73.6392  1.012665e-10 3.570760e-02   pass
10000  10000  4      9.178      72.6626  1.012665e-10 3.570760e-02   pass
10000  10000  4      9.259      72.0251  1.012665e-10 3.570760e-02   pass
10000  10000  4      8.932      74.6572  1.012665e-10 3.570760e-02   pass
10000  10000  4      8.590      77.6374  1.012665e-10 3.570760e-02   pass

Performance Summary (GFlops)

Size   LDA    Align.  Average  Maximal
10000  10000  4       77.9235  119.7422

Residual checks PASSED

End of tests

Press [Enter] key to continue...

This is with AVX enabled. I get lower GFLOPS, higher times and residuals all mismatch (although they pass).

user@localhost:~/bin/LinpackXtreme-1.1.5/linpack-xtreme-1.1.5-amd64> taskset -c 0-8 ./run_stress_test
Linpack Xtreme v1.1.5 by Regeneration

Current date/time: Tue May 17 13:55:54 2022

CPU frequency:    4.987 GHz
Number of CPUs: 1
Number of cores: 12
Number of threads: 20

Parameters are set to:

Number of tests: 1
Number of equations to solve (problem size) : 10000
Leading dimension of array                  : 10000
Number of trials to run                     : 30  
Data alignment value (in Kbytes)            : 4   

Maximum memory requested that can be used=800204096, at the size=10000

=================== Timing linear equation system solver ===================

Size   LDA    Align. Time(s)    GFlops   Residual     Residual(norm) Check
10000  10000  4      10.198     65.3891  8.936940e-11 3.151256e-02   pass
10000  10000  4      11.546     57.7549  1.207682e-10 4.258408e-02   pass
10000  10000  4      10.163     65.6191  9.546952e-11 3.366352e-02   pass
10000  10000  4      10.173     65.5510  8.508394e-11 3.000146e-02   pass
10000  10000  4      11.553     57.7213  1.075682e-10 3.792964e-02   pass
10000  10000  4      11.564     57.6674  9.443812e-11 3.329984e-02   pass
10000  10000  4      11.539     57.7935  9.973072e-11 3.516606e-02   pass
10000  10000  4      11.556     57.7051  8.705343e-11 3.069592e-02   pass
10000  10000  4      11.573     57.6211  9.485529e-11 3.344694e-02   pass
10000  10000  4      11.590     57.5406  1.019788e-10 3.595876e-02   pass
10000  10000  4      11.724     56.8805  1.007526e-10 3.552638e-02   pass
10000  10000  4      11.565     57.6623  1.101995e-10 3.885746e-02   pass
10000  10000  4      11.583     57.5753  9.029810e-11 3.184002e-02   pass
10000  10000  4      11.581     57.5821  9.781812e-11 3.449166e-02   pass
10000  10000  4      11.530     57.8359  9.533982e-11 3.361779e-02   pass
10000  10000  4      11.601     57.4835  8.729871e-11 3.078241e-02   pass
10000  10000  4      11.651     57.2361  9.254220e-11 3.263132e-02   pass
10000  10000  4      11.636     57.3083  9.643886e-11 3.400532e-02   pass
10000  10000  4      11.515     57.9115  1.078766e-10 3.803838e-02   pass
10000  10000  4      10.155     65.6679  9.228884e-11 3.254198e-02   pass
10000  10000  4      11.601     57.4824  9.546830e-11 3.366309e-02   pass
10000  10000  4      11.633     57.3234  9.761916e-11 3.442150e-02   pass
10000  10000  4      11.584     57.5661  1.053115e-10 3.713390e-02   pass
10000  10000  4      11.592     57.5285  9.667123e-11 3.408725e-02   pass
10000  10000  4      11.605     57.4642  9.761894e-11 3.442143e-02   pass
10000  10000  4      11.619     57.3963  9.895956e-11 3.489414e-02   pass
10000  10000  4      11.593     57.5237  9.235229e-11 3.256435e-02   pass
10000  10000  4      11.611     57.4319  9.445195e-11 3.330471e-02   pass
10000  10000  4      11.586     57.5567  1.007096e-10 3.551123e-02   pass
10000  10000  4      10.151     65.6946  9.080495e-11 3.201874e-02   pass

Performance Summary (GFlops)

Size   LDA    Align.  Average  Maximal
10000  10000  4       58.8825  65.6946

Residual checks PASSED

End of tests

Press [Enter] key to continue...

 

[CiTay]

To me, this is not normal, and i would consider it a kind of instability. Not as severe as a "fail" on a pass, but it would certainly raise an eyebrow and prompt me to do a bunch of further tinkering and testing.

Do you have any inkling of problems in things like Prime95 or so?

 

[Regeneration]

Does anybody have a source for the statement according to which it's "normal" for Linpack residuals to occasionally mismatch on modern Intel processors?
I'm not convinced that this is accurate or representative of fully stable operating conditions.

On AVX2+ hardware, it is normal to get different residuals since Linpack switches between the AVX modes and extensions.

Run it on stock clocks and see for yourself.

 

[CiTay]

That's good to know.

 

[Solid State Brain]

To me, this is not normal, and i would consider it a kind of instability. Not as severe as a "fail" on a pass, but it would certainly raise an eyebrow and prompt me to do a bunch of further tinkering and testing.

Do you have any inkling of problems in things like Prime95 or so?

I rarely observed random crashes in other programs which I thought could have potentially been due to the slight overclock I'm running, but it turned out that what I'm observing with Linpack Xtreme does not depend on the overclock, just its use of AVX instructions.

On AVX2+ hardware, it is normal to get different residuals since Linpack switches between the AVX modes and extensions.

Run it on stock clocks and see for yourself.

After more tests (as I added in an edit to the previous message), I did narrow down this oddity to AVX instructions. Disabling AVX in BIOS makes all residuals equal.

CPU or memory clocks did not seem to make a difference. However, running Linpack Xtreme with a small problem size (< 1GB memory) made this much clearer. With a large problem size, residual mismatches do not occur often.

EDIT: FYI, I tried the latest version of Linpack from the Intel website, and using the same test settings I don't get the same issue with it. It also has slightly higher performance and generates somewhat more heat:

Intel® Math Kernel Library (Intel® MKL) Benchmarks Suite

Intel® Math Kernel Library (Intel® MKL) Benchmarks package includes Intel® Distribution for LINPACK* Benchmark, Intel® Distr

www.intel.com

 

[cyh]

Any recommendations on how many times to run linpack extreme stress test after overclocking?

 

[Regeneration]

Any recommendations on how many times to run linpack extreme stress test after overclocking?

Enough times for it to run overnight (8-12 hours) if you want 1000% stable system.

 

[Blaeza]

With how demanding your program is Mr Regeneration, I'm a bit scared to run it! It's not going to damage anything is it?

 

[Regeneration]

With how demanding your program is Mr Regeneration, I'm a bit scared to run it! It's not going to damage anything is it?

Not as long as the CPU cores temperatures are below Tjunction by at least 10c/50f.