Linpack Xtreme Released

[Regeneration]

Hi there. My 9900K@5.0 only got around 230 GFLOPS, which is ridiculous. I've seen people getting at least 500 GFLOPS or more.
Also, how come my 9900K only has 8 threads? And why is my maximum CPU usage only 52%?

Linpack benchmark runs faster and more effective only on real cores (not virtual ones).

Your score is the result of throttling or slow memory.

 

[EarthDog]

Linpack benchmark runs faster and more effective only on real cores (not virtual ones).

Your score is the result of throttling or slow memory.

This doesnt use HT to stress test??

...or the benchmark only doesnt?

 

[toyo]

Hi there. My 9900K@5.0 only got around 230 GFLOPS, which is ridiculous. I've seen people getting at least 500 GFLOPS or more.
Also, how come my 9900K only has 8 threads? And why is my maximum CPU usage only 52%?

You need to see what is causing your chip to throttle under the stresstest, could be VRMs heating up, CPU thermals, power limits in BIOS. You should have also finished the Windows update process before stresstesting.
You can see the throttling reasons in HWinfo64.

 

[liujohn6571]

Linpack benchmark runs faster and more effective only on real cores (not virtual ones).

Your score is the result of throttling or slow memory.

But there's no throttling since the highest temperature of my CPU is only 80°c. And I'm using DDR4 4400 17-17-17-37 Dual Channel, which is not slow at all.

You need to see what is causing your chip to throttle under the stresstest, could be VRMs heating up, CPU thermals, power limits in BIOS. You should have also finished the Windows update process before stresstesting.
You can see the throttling reasons in HWinfo64.

I'm using the ASUS ROG Maximus XI APEX, which shouldn't have any VRMs overheat issue. Both of the Tjmax and CPU power limit have been adjusted to their maximum value.

 

[Regeneration]

This doesnt use HT to stress test??

...or the benchmark only doesnt?

Only the benchmark doesn't. Better scores that way.

 

[EarthDog]

Only the benchmark doesn't. Better scores that way.

That doesnt make sense to me why you would make the bench not use HT/SMT...but ok. Who cares about faster... you care about how it performs with all cores and threads...since most dont shut off HT/SMT.

 

[toyo]

But there's no throttling since the highest temperature of my CPU is only 80°c. And I'm using DDR4 4400 17-17-17-37 Dual Channel, which is not slow at all.
I'm using the ASUS ROG Maximus XI APEX, which shouldn't have any VRMs overheat issue. Both of the Tjmax and CPU power limit have been adjusted to their maximum value.

I think I know what the "issue" is. You're likely comparing the results from the benchmark to those from stresstesting, which uses all the threads. I'm getting 232 Gflops from the bench, but 340 Gflops on the stresstest with 8GB RAM used that runs on all threads.

 

[borandi]

Is there a list of benchmark flags so I can call the benchmark from a batch file without having to input keys ? Also a flag to output the results file? I can't see any FAQ/readme on any of the command line flags

 

[Regeneration]

Is there a list of benchmark flags so I can call the benchmark from a batch file without having to input keys ? Also a flag to output the results file? I can't see any FAQ/readme on any of the command line flags

Batch run is supported on Linux, but not on Windows. I'll consider adding it in the next version.

There is only a single command line for Windows: /residualcheck

Forces check of residual values for processors without AVX.

 

[borandi]

Thanks for the answer!
Just to clarify, by batch file I don't mean batch run. I just want to automate it as a series of tests and get the GFLOPs data out. None of the current LinX derivatives seem to work on Zen2 CPUs.

As it stands with the request for user input, that doesn't allow for automation.
You could make it so if it already detects the ini file present, it doesn't present options, it just runs what's in the ini file.

 

[Pop000100]

So does using /residualcheck on a AVX capable CPU cause false positives?

 

[Regeneration]

So does using /residualcheck on a AVX capable CPU cause false positives?

Yes.

 

[buildzoid]

Hi I've been seeing some conflicting info on the residuals having to all be the same and so I was wondering if you could provide a link to an article/documentation on why AVX float math isn't always consistent.

 

[Mezilkree]

Hi I've been seeing some conflicting info on the residuals having to all be the same and so I was wondering if you could provide a link to an article/documentation on why AVX float math isn't always consistent.

Introduction to Conditional Numerical Reproducibility (CNR)

Starting with 11.0 release, Intel® MKL introduces a feature called Conditional Numerical Reproducibility (CNR) which provide

software.intel.com

 

[mrthanhnguyen]

When it shows pass is pass or all residuals need to be the same.

 

[pipes]

after testing a stress test with the bootable version, again from that version I launched a test with the linpack libraries and the system crashed

I think is more good an update with last linpack library intel for find problem accuracy

 

[Athlonite]

When it shows pass is pass or all residuals need to be the same.

All residuals should technically be the same if they're not then there's a problem with one or more cores depending on how many have different residuals but not bad enough to cause a crash so it gets a pass

 

[pipes]

I need password root for launch mprime on live usb distro

 

[sp00n]

Is it broken again on Ryzen 5000? I'm receiving that hardware failure error every time, which was mentioned for Ryzen 3000.

I'm running a Ryzen 5900x on an MSI MAG X570 Tomahawk with the latest BIOS (2020-11-16).

 

[Regeneration]

Is it broken again on Ryzen 5000? I'm receiving that hardware failure error every time, which was mentioned for Ryzen 3000.

I'm running a Ryzen 5900x on an MSI MAG X570 Tomahawk with the latest BIOS (2020-11-16).

Fixed in the latest version.

 

[Octopuss]

Can you update the bootable one too?

You should also update the TPU download links, they still have the old version.

 

[sp00n]

I'm a bit confused. Back on my Ryzen 1700x Linpack Xtreme v1.1.1 used to generate the most heat and draw the most power of all the stress tests. Now on my new Ryzen 5900x the new version 1.1.4 doesn't come anywhere near Prime 95 Small FFT regarding CPU Package Power, the temperature is also around 15°K lower. It also doesn't seem to detect errors as fast.

 

[RJARRRPCGP]

Do the Gflops seem unusually low? If so, it could be the same issue that I had with another Linpack test (LinX, IIRC)
(Where when I was using LinX, IIRC, it would randomly fail to fully load the CPU and I would get ridiculously low Gflops, until I restarted the test.)

 

[sp00n]

Do the Gflops seem unusually low? If so, it could be the same issue that I had with another Linpack test (LinX, IIRC)
(Where when I was using LinX, IIRC, it would randomly fail to fully load the CPU and I would get ridiculously low Gflops, until I restarted the test.)

I don't know if the Gflops are low, I have nothing to compare to. But it happened across multiple runs with different RAM settings (2GB, 8GB, and max). Also in the task manager all cores appear as 100% used.

 

[Regeneration]

I'm a bit confused. Back on my Ryzen 1700x Linpack Xtreme v1.1.1 used to generate the most heat and draw the most power of all the stress tests. Now on my new Ryzen 5900x the new version 1.1.4 doesn't come anywhere near Prime 95 Small FFT regarding CPU Package Power, the temperature is also around 15°K lower. It also doesn't seem to detect errors as fast.

I mostly use Intel lately with no access to the latest Ryzen processors. I'll send you a beta build later to test if interested.