Linpack Xtreme Released

[RJARRRPCGP]

unless there is some background service that consume a lot of memory.

I know that I have with another RAM kit before, ran those other tests with no errors. Is Linpack silently failing to allocate RAM, because of NT 6/10's OOM'ing BS and causing truncated data?
In my case, it could be a service or the like, for all I know. For all I know, I may be changing tons of voltage settings for nothing.

I saw an issue like this with FX8350, as well! So I went to Prime95 and it always looked good in Prime95, when it's not related to core testing. Linpack is excellent for core testing, but don't know about anything else, as I also saw Linpack not reveal a bus instability before, that Prime95 blend did.

Note that I have a new 32 GB RAM kit that I didn't even try to use yet. It's G.Skill RipJaws V 3200.

 

[legitmaan]

Is it normal to have many variations in the residual values on a stock i5-13600K? I find that I have to disable ecores to get the residual values to match over 90% of the time. Do I have a defective CPU or ecores?

 

[CiTay]

I think for anything from Intel 12th gen onwards it can happen. The main thing is that there is always a PASS and not a FAIL, then you are good.

 

[Regeneration]

Is it normal to have many variations in the residual values on a stock i5-13600K? I find that I have to disable ecores to get the residual values to match over 90% of the time. Do I have a defective CPU or ecores?

It's normal for CPUs that support AVX 2 and above.

 

[legitmaan]

Thanks for the responses. Is it something that happens with Zen 4 as well?

 

[Regeneration]

Thanks for the responses. Is it something that happens with Zen 4 as well?

Nope.

 

[freeagent]

A little is normal I think.. mine fluctuates with temperature, when it is cold in here I can see very little change in output

 

[Octopuss]

So, will there be any updates? I mean, it's been 3 years according to first post...

 

[freeagent]

So, will there be any updates? I mean, it's been 3 years according to first post...

Does it need to be? Run it and see what happens.

 

[legitmaan]

I noticed my residuals match a lot more often with my ram at auto than with XMP. Like 10-30 of the same residuals in 50 run sets versus just 2.

I’m using my 13600K with Corsair Vengeance 32GB 6000 CL30 on an ASUS Strix Z790-A II.

Is having fewer matching residuals on XMP a sign of subtle memory instability or does it not matter? I would think that my system spec is strong enough for a 6000 CL30 kit. My system with XMP on passed dozens of hours of Karhu, TestMem 5, and HCI MemTest, for what it’s worth.

Edit: Oof, my PC shut down during a no AVX2/ecores off run after around 15 trials. It then finished 50 trials the next run. A fluke I hope.

 

[CiTay]

I remember even Buildzoid from "Actually Hardcore Overclocking" saying in one of his videos that he previously took non-matching residuals as a sign of hidden instability of sorts, but ever since 12th gen at the latest, he disregards it, because there it can happen even with a totally stable memory system.

Your system should have no problem with DDR5-6000, yes. Provided you are on a recent enough BIOS version and your IMC is not a total dud.

About the PC shutting down, have HWinfo64 Sensors open, with all the sensors expanded (the <-> symbol at the bottom) so you have three big columns of sensors at least, then monitor the power draw and temperatures of everything. A 13600K, depending on your cooler, can overwhelm your cooling in Linpack, or whatever else might be going on.

 

[dKenGuru]

I have a problem with this benchmark running on Win11 24H2, RTM build.
7950X3D.
OMP: Warning #80: OMP_NUM_THREADS="0": value too small.
OMP: Info #104: OMP_NUM_THREADS value "1" will be used.

Сorrectly detects CPU/Cores - 32. But can't create more then 1 thread.

SMT is on. No any problems in other benchmarks, programs or games.

 

[WhenMusicAttacks]

I have a problem with this benchmark running on Win11 24H2, RTM build.
7950X3D.
OMP: Warning #80: OMP_NUM_THREADS="0": value too small.
OMP: Info #104: OMP_NUM_THREADS value "1" will be used.

Сorrectly detects CPU/Cores - 32. But can't create more then 1 thread.

SMT is on. No any problems in other benchmarks, programs or games.

It relied on WMIC and that has gone on 24h2 apparently

WMI command line (WMIC) utility deprecation: Next steps | Windows IT Pro blog

Get ready for the next step in WMIC deprecation: this feature on demand will be disabled by default soon.

techcommunity.microsoft.com

maybe it's just possible to bring it back, i might try that

 

[sp00n]

WMI itself is still available according to that site, just the wmic command line utlity is gone. So you can still query WMI for the amount of processors (just not using wmic).

Aside from that, is there a way to directly start a stress test without going through the setup first? I've noticed that the main exe calls one of the binaries, e.g. linpack_amd64.exe linpack.ini, but I couldn't find any reference to the amount of threads that should be tested inside the ini file. And when I manually start the exe pointing to the ini file, it starts with all available threads.
If there's a way to set this (and read any error output), I might be able to add support for Linpack Xtreme to my CoreCycler script, that's intended to test single-core stability.


// Edit
Here's a one-liner that executes a powershell command from cmd.exe that returns the number of physical (installed) processors, the number of logical processors, and the number of physical cores:

powershell -command "$processor = Get-CimInstance -ClassName Win32_Processor; @($processor).Count; $($processor | Measure-Object -Property NumberOfCores -sum).Sum; $($processor | Measure-Object -Property NumberOfLogicalProcessors -sum).Sum"

 

[dKenGuru]

So, no update to this benchmark for actual Windows versions?

 

[sp00n]

So, no update to this benchmark for actual Windows versions?

The .exe file mostly seems to be a wrapper for a batch file (which is copied to the temp directory during runtime).
You could try to use this .bat file instead, I've changed it to use the powershell command instead of the wmic calls.

@shift /0
@echo off
SETLOCAL

:menu
if exist linpack.ini del /f /q linpack.ini
if exist results.txt del /f /q results.txt
if /i [%1] EQU [/residualcheck] set RESIDUALCHECK=1
echo %PROCESSOR_IDENTIFIER% | find /i "AMD64 Family 25" > nul
if %ERRORLEVEL% EQU 0 set MKL_DEBUG_CPU_TYPE=4
echo %PROCESSOR_IDENTIFIER% | find /i "AMD64 Family 23 Model 144" > nul
if %ERRORLEVEL% EQU 0 set MKL_DEBUG_CPU_TYPE=4
echo %PROCESSOR_IDENTIFIER% | find /i "AMD64 Family 23 Model 113" > nul
if %ERRORLEVEL% EQU 0 set MKL_DEBUG_CPU_TYPE=4
echo %PROCESSOR_IDENTIFIER% | find /i "AMD64 Family 23 Model 96" > nul
if %ERRORLEVEL% EQU 0 set MKL_DEBUG_CPU_TYPE=4
echo %PROCESSOR_IDENTIFIER% | find /i "AMD64 Family 23 Model 49" > nul
if %ERRORLEVEL% EQU 0 set MKL_DEBUG_CPU_TYPE=4
cls
echo ==============================================================
echo =-=     Linpack Xtreme v1.1.5 by Regeneration (64-bit)     =-=
echo =-=      Based on Intel Math Kernel Library Benchmark      =-=
echo =-=          Copyright (c) 2020 Intel Corporation          =-=
echo ==============================================================
echo.
echo 1. Benchmark
echo 2. Stress Test
echo 3. View Readme.txt
echo 4. Exit
echo.
choice /c:1234 /n /m "Please select an action to perform: "
if %ERRORLEVEL% EQU 4 goto end
if %ERRORLEVEL% EQU 3 goto readme
if %ERRORLEVEL% EQU 2 goto stresstest
if %ERRORLEVEL% EQU 1 goto benchmark

:benchmark
cls
echo ==============================================================
echo =-=     Linpack Xtreme v1.1.5 by Regeneration (64-bit)     =-=
echo =-=      Based on Intel Math Kernel Library Benchmark      =-=
echo =-=          Copyright (c) 2020 Intel Corporation          =-=
echo ==============================================================
echo.
echo 1. Quick 2GB Benchmark
echo 2. Standard 3GB Benchmark
echo 3. Extended 8GB Benchmark
echo.
echo 4. Go back
echo.
choice /c:1234 /n /m "Please select a profile: "
if %ERRORLEVEL% EQU 4 goto menu
echo Linpack data file > linpack.ini
echo Linpack Xtreme v1.1.5 by Regeneration (64-bit) >> linpack.ini
echo 1 >> linpack.ini
if %ERRORLEVEL% EQU 3 set PROBLEMSIZE=32000
if %ERRORLEVEL% EQU 2 set PROBLEMSIZE=20000
if %ERRORLEVEL% EQU 1 set PROBLEMSIZE=15000
echo %PROBLEMSIZE% >> linpack.ini
echo %PROBLEMSIZE% >> linpack.ini
goto runbenchmark

:runbenchmark
cls
echo ==============================================================
echo =-=     Linpack Xtreme v1.1.5 by Regeneration (64-bit)     =-=
echo =-=      Based on Intel Math Kernel Library Benchmark      =-=
echo =-=          Copyright (c) 2020 Intel Corporation          =-=
echo ==============================================================
echo.
set /p RUNS="Enter number of times to run Linpack (1-5): "
if %RUNS% LSS 1 goto runbenchmark
if %RUNS% GTR 5 goto runbenchmark
echo %RUNS% >> linpack.ini
echo 4 >> linpack.ini
FOR /F "usebackq tokens=1,2,3 delims=," %%G IN (`powershell.exe -Command "$processor = Get-CimInstance -ClassName Win32_Processor; @(@($processor).Count; $($processor | Measure-Object -Property NumberOfCores -sum).Sum; $($processor | Measure-Object -Property NumberOfLogicalProcessors -sum).Sum) -Join ','"`) DO (
    set "CPUS=%%G"
    set "CORES=%%H"
    set "THREADS=%%I"
)
set CORES=%CORES: =%
set CPUS=%CPUS: =%
set THREADS=%THREADS: =%
set /a OMP_NUM_THREADS="CORES * CPUS"
echo.
echo %CPUS% CPU(s) %CORES% core(s) %THREADS% thread(s) detected.
echo.
echo Ready to execute Linpack binary.
echo.
echo Press CTRL+BREAK anytime to abort.
echo.
pause
set OMP_PLACES=CORES
set OMP_PROC_BIND=TRUE
set MKL_DYNAMIC=FALSE
set PATH=.\binaries\x64;%PATH%
cls
linpack_amd64.exe linpack.ini
pause
goto menu

:stresstest
cls
echo ==============================================================
echo =-=     Linpack Xtreme v1.1.5 by Regeneration (64-bit)     =-=
echo =-=      Based on Intel Math Kernel Library Benchmark      =-=
echo =-=          Copyright (c) 2020 Intel Corporation          =-=
echo ==============================================================
echo.
echo 1. 2GB
echo 2. 4GB
echo 3. 6GB
echo 4. 8GB
echo 5. 10GB
echo 6. 14GB (experimental)
echo 7. 30GB (experimental)
echo.
echo 8. Go back
echo.
choice /c:12345678 /n /m "Please select amount of RAM to be used: "
if %ERRORLEVEL% EQU 8 goto menu
echo Linpack data file > linpack.ini
echo Linpack Xtreme v1.1.5 by Regeneration (64-bit) >> linpack.ini
echo 1 >> linpack.ini
if %ERRORLEVEL% EQU 7 set PROBLEMSIZE=62897
if %ERRORLEVEL% EQU 6 set PROBLEMSIZE=42789
if %ERRORLEVEL% EQU 5 set PROBLEMSIZE=35000
if %ERRORLEVEL% EQU 4 set PROBLEMSIZE=32209
if %ERRORLEVEL% EQU 3 set PROBLEMSIZE=27818
if %ERRORLEVEL% EQU 2 set PROBLEMSIZE=22611
if %ERRORLEVEL% EQU 1 set PROBLEMSIZE=15825
echo %PROBLEMSIZE% >> linpack.ini
echo %PROBLEMSIZE% >> linpack.ini
goto runstresstest

:runstresstest
cls
echo ==============================================================
echo =-=     Linpack Xtreme v1.1.5 by Regeneration (64-bit)     =-=
echo =-=      Based on Intel Math Kernel Library Benchmark      =-=
echo =-=          Copyright (c) 2020 Intel Corporation          =-=
echo ==============================================================
echo.
set /p RUNS="Enter number of times to run Linpack (1-10000): "
if %RUNS% LSS 1 goto runstresstest
if %RUNS% GTR 10000 goto runstresstest
echo %RUNS% >> linpack.ini
echo 4 >> linpack.ini
FOR /F "usebackq tokens=1,2,3 delims=," %%G IN (`powershell.exe -Command "$processor = Get-CimInstance -ClassName Win32_Processor; @(@($processor).Count; $($processor | Measure-Object -Property NumberOfCores -sum).Sum; $($processor | Measure-Object -Property NumberOfLogicalProcessors -sum).Sum) -Join ','"`) DO (
    set "CPUS=%%G"
    set "CORES=%%H"
    set "THREADS=%%I"
)
set CORES=%CORES: =%
set CPUS=%CPUS: =%
set THREADS=%THREADS: =%
set /a MAXTHREADS="THREADS * CPUS"
echo.
echo %CPUS% CPU(s) %CORES% core(s) %THREADS% thread(s) detected.
echo.
choice /c:yn /n /m "Do you wish to use all available threads? (Y/N)"
if %ERRORLEVEL% EQU 2 goto specifythreads
set OMP_NUM_THREADS=%MAXTHREADS%
goto sleepmode

:specifythreads
echo.
set /p CUSTOMTHREADS="Specify amount of threads (1-%MAXTHREADS%): "
if %CUSTOMTHREADS% LSS 1 goto specifythreads
if %CUSTOMTHREADS% GTR %MAXTHREADS% goto specifythreads
set OMP_NUM_THREADS=%CUSTOMTHREADS%
goto sleepmode

:sleepmode
echo.
choice /c:yn /n /m "Do you wish to disable sleep mode? (Y/N)"
if %ERRORLEVEL% EQU 2 goto hwmonitor
powercfg.exe -change -standby-timeout-ac 0
powercfg.exe -change -hibernate-timeout-ac 0
goto hwmonitor

:hwmonitor
echo.
choice /c:yn /n /m "Do you wish to run CPUID HWMonitor in the background? (Y/N)"
echo.
echo Ready to execute Linpack binary.
echo.
echo Press CTRL+BREAK anytime to abort.
echo.
pause
if exist results.txt goto menu
set OMP_PLACES=CORES
set OMP_PROC_BIND=SPREAD
set MKL_DYNAMIC=FALSE
set PATH=.\binaries\x64;%PATH%
if %ERRORLEVEL% EQU 1 start /min .\hwmonitor\HWMonitor_x64.exe
echo %PROCESSOR_IDENTIFIER% | find /i "AMD" > nul
if %ERRORLEVEL% EQU 0 set RESIDUALCHECK=1
if %ERRORLEVEL% EQU 0 start /b linpack_amd64.exe linpack.ini > results.txt
if %ERRORLEVEL% EQU 1 start /b linpack_intel64.exe linpack.ini > results.txt
echo.
echo Stress test will begin in 5 seconds...
timeout 5 /nobreak > nul
goto checkprocess

:checkprocess
tasklist.exe | find /i "linpack_" > nul
if %ERRORLEVEL% EQU 0 goto checkstatus
if exist results.txt goto checkresults
goto menu

:checkstatus
cls
type results.txt
find /i "fail" results.txt > nul
if %ERRORLEVEL% EQU 0 goto fail
if defined RESIDUALCHECK goto checkresidual
goto noerrors

:checkresidual
for /f "tokens=6 skip=22" %%G IN ('type results.txt') DO set RESIDUAL=%%G
if not defined RESIDUAL goto noerrors
for /f "tokens=6 skip=22" %%H IN ('type results.txt') DO if /i %RESIDUAL% NEQ %%H goto fail
goto noerrors

:noerrors
echo.
echo No errors detected so far. [%date% - %time%]
timeout 10 /nobreak > nul
goto checkprocess

:checkresults
cls
type results.txt
find /i "passed" results.txt > nul
if %ERRORLEVEL% EQU 0 goto success
goto fail

:fail
tasklist.exe | find /i "linpack_" > nul
if %ERRORLEVEL% EQU 0 taskkill.exe /f /im linpack_* > nul
echo 
echo Hardware failure detected. System failed to maintain stability.
echo 
pause
goto menu

:success
pause
goto menu

:readme
if exist readme.txt (
    cls
    type readme.txt|more
) else (
    echo.
    echo Error! the file is missing.
    echo.
)
pause
goto menu

:end
if exist linpack.ini del /f /q linpack.ini
if exist results.txt del /f /q results.txt
ENDLOCAL
exit

 

[dKenGuru]

Now works fine. Thx!
But some mismatch in cpu detection/information.

 

[sp00n]

I think that's normal. When I looked through the thread earlier, it was mentioned that for a benchmark run Hyperthreading / SMT is not used, so it's only using half of the available "virtual" threads.
During a stress test run, all of them are being used (and displayed as "number of threads").

 

[sp00n]

While looking through the batch file and reading up on how Linpack / Intel's MKL library handles things, I noticed that Linpack Xtreme uses only AVX instructions for AMD processors, and not AVX2.
Intel processors are always using AVX2, but since Intel didn't really care about AMD support, you need to set an undocumented environment variable to enable AVX or AVX2 support for AMD processors (MKL_DEBUG_CPU_TYPE=4 resp. MKL_DEBUG_CPU_TYPE=5, see here for the original thread).

In Linpack Xtreme this variable is set to 4 for AMD (see the batch file code above), which means that the instructions to be used are set to AVX, and not to AVX2.

And interestingly, setting it to 4 does increase performance / GFlops during an all-core test.
However already for the benchmark with only half of the available virtual CPUs / threads, and even more so when using a low thread count like 1 or 2, using MKL_DEBUG_CPU_TYPE=5 over 4 to enable AVX2 does indeed increase performance. At least on my Ryzen 5900x processor.

Here are some examples:

Stress test, all threads, 2GB
MKL_DEBUG_CPU_TYPE=4: 405 GFlops
MKL_DEBUG_CPU_TYPE=5: 332 GFlops

Benchmark, 12 threads (automatic), 2GB
MKL_DEBUG_CPU_TYPE=4: 417 GFlops
MKL_DEBUG_CPU_TYPE=5: 423 GFlops

Stress test, 1 thread, 2GB
MKL_DEBUG_CPU_TYPE=4: 45 GFlops
MKL_DEBUG_CPU_TYPE=5: 66 GFlops

 

[Det0x]

Would be really nice if we could get a update to this benchmark, i kinda use it alot already on vanilla Zen4
(but lately i have to fire up LinX if i want to see those ~1900GFLOPS numbers)

Defaulting to AVX2 / AVX512 for supported CPU's would also leveling the playing field

 

[sp00n]

I guess that would require a larger overhaul. Intel's Math Kernel Library has been changed in 2020 (also see the original thread) and doesn't support that debug flag anymore.
At the same time they have added some improvements for AMD, but it being an Intel library designed for Intel chips, they probably don't put a lot of effort into it.

As a reaction, AMD has developed its AOCL package (AMD Optimizing CPU Libraries), which - if updating MKL (which is now also a part of Intel's oneAPI package) - should be used as an alternative for AMD processors.

It's all a bit complicated, and I'm not very deep into the issue either.

 

[WhenMusicAttacks]

So, no update to this benchmark for actual Windows versions?

Add or Remove WMIC command Feature in Windows 11

This tutorial will show you how to add or remove the WMIC optional feature for all users in Windows 11. The WMI command-line (WMIC) utility provides a command-line interface for Windows Management Instrumentation (WMI). WMIC is compatible with existing shells and utility commands. The WMIC...

www.elevenforum.com

this worked for me

 

[dKenGuru]

But not for me.

More better if someone create AOCL based test/benchmark.

 

[sp00n]

So I did make some tests with the latest oneMKL release (2024.2.1.0). I patched the linpack_xeon64.exe to allow it to run on AMD CPUs (replace the hex values E8 38 4F with B8 01 00 at offset 0x533).

This version runs faster (more GFlops) on one thread than the version from Linpack Xtreme, when it's running with the default MKL_DEBUG_CPU_TYPE=4 settings. But it's still slower than if you set it to MKL_DEBUG_CPU_TYPE=5.
As mentioned, since 2020.something this environment variable isn't available anymore for Intel's MKL library, so it at least runs with decent speed now by default.
I didn't test with all threads or benchmark, but these are the results for me:
Linpack Xtreme 1 thread MKL_DEBUG_CPU_TYPE=4: 45 GFlops
Linpack Xtreme 1 thread MKL_DEBUG_CPU_TYPE=5: 65 GFlops
Linpack 1 thread 2024.2.1.0: 59 GFlops

And while looking at the binaries, I wonder why the linpack_amd64.exe is so different in size than the linpack_intel64.exe. Normally the patch I mentioned above doesn't change the file size, so I wonder if some additional changes have been made to the binary? Is it maybe not even Intel's MKL linpack version but actually one that was already compiled using AMD's libraries? @Regeneration

The latest HPL version ("High Performance Linpack"), the original code Intel's binary is based on, is from 2018, so also not exactly very recent.
Trying to compile this HPL code with AMD's libraries would be the next step I guess, although I have no experience with these kind of things, so I'm not sure I'm willing to invest the time to learn how to do it.


// Edit
I got ahold of a pristine version of Intel's MKL linpack 2018.0.3.1, (the included linpack_xeon64.exe perfectly matches Linpack Xtreme's linpack_intel64.exe), and when patching it to allow AMD, it shows the same performance as Linpack Xtreme's linpack_amd64.exe file. So there don't seem to be any additional optimizations, which makes me wonder even more why the file size is so different.

 

[sp00n]

Would be really nice if we could get a update to this benchmark, i kinda use it alot already on vanilla Zen4
(but lately i have to fire up LinX if i want to see those ~1900GFLOPS numbers)

Defaulting to AVX2 / AVX512 for supported CPU's would also leveling the playing field

I have no way to test this myself, but you could try to add this to the batch file:

set "MKL_DEBUG_CPU_TYPE=7"
set "MKL_ENABLE_INSTRUCTIONS=AVX512"

This supposedly activates resp. forces the AVX512 pathway. And crashes on my non-AVX512 capable Zen 3.
"5" is AVX2, no idea what "6" would do.