Linpack Xtreme Released

[TesterMD]

@Regeneration Just curious if you will be updating this to use the most recent MKL binaries (2019.4.003)?

 

[er557]

this bench is pretty heavy on the ram as well, especially with the 8gb large benchmark, it's best to avoid it using the page file if at all possible, it would be also more accurate to detect multi socket systems as two cpus, but it probably has to do with Hyperthreading detection

 

[Regeneration]

@Regeneration Just curious if you will be updating this to use the most recent MKL binaries (2019.4.003)?

The latest MKL binaries are somehow buggy with thread binding.

 

[Regeneration]

There appears to be a problem with Zen 2 (Ryzen 3000 series) running Linpack with AVX support.

 

[EarthDog]

There appears to be a problem with Zen 2 (Ryzen 3000 series) running Linpack with AVX support.

............do tell.............

AIDA64 and P95 seem to do OK...

 

[Regeneration]

There are problems with other apps, Destiny 2, WHEA warnings in Windows.

hXXps://www.reddit.com/r/Amd/comments/cbls9g/the_final_word_on_idle_voltages_for_3rd_gen_ryzen/

 

[EarthDog]

That's pretty old news, the WHEA errors, but that was attributed to PCIe 4.0 devices like NVMe's and such last I read about it.

 

[er557]

whea is a bad sign for a hardware part, next up: bsod

 

[Mussels]

whea is a bad sign for a hardware part, next up: bsod

no, driver bug that caused errors in the event log and nothing else, with fixes already on the way.

 

[Arctucas]

Good, bad, otherwise...?

 

[EarthDog]

Good, bad, otherwise...?

This is a vomit of information in a post . Temps and voltages look good though if that is what you are asking.

How long did you run the test?

 

[Arctucas]

This is a vomit of information in a post . Temps and voltages look good though if that is what you are asking.

How long did you run the test?

One pass.

Just relidded, wanted to see how it is doing.

 

[Arctucas]

This is strange; I ran Linpack Extreme 1.1.1 again today, and it only used 8 threads. Linpack Extreme 1.0 only uses 8 as well.

Yes, HT is on. Linpack Extreme 0.9 uses all 16 threads.

Other apps such as LinX, RealBench , AIDA64, OCCT, Prime95 use all 16 threads.

WTF?

 

[Regeneration]

This is strange; I ran Linpack Extreme 1.1.1 again today, and it only used 8 threads. Linpack Extreme 1.0 only uses 8 as well.

Yes, HT is on. Linpack Extreme 0.9 uses all 16 threads.

Other apps such as LinX, RealBench , AIDA64, OCCT, Prime95 use all 16 threads.

WTF?

Linpack delivers better scores without HT/SMT (benchmarking).

HT/SMT should be used only for stress testing.

 

[EarthDog]

Not sure he asked about a score... just that for some reason the latest version for him isnt running all c/t.

 

[Mussels]

benchmark mode always limited the threads, stress test mode used them all

 

[redux]

Different residual values are normal on modern Intel CPUs.

Abnormal on AMD and legacy Intel CPUs (Sandy Bridge and older).

@Regeneration Do you still consider varying residuals normal? I'm asking because my 9900K calculates slightly different residuals on stock clocks, even if using memory at 2133 MHz. Most of the residuals are the same, but a small percentage of runs (like 10 %) give a residual that is between 0.5-1.5 x the normal residual.

I've had sandy bridge, ivy bridge and haswell-e systems. All of these have calculated the same residuals between runs.

 

[Regeneration]

@Regeneration Do you still consider varying residuals normal? I'm asking because my 9900K calculates slightly different residuals on stock clocks, even if using memory at 2133 MHz. Most of the residuals are the same, but a small percentage of runs (like 10 %) give a residual that is between 0.5-1.5 x the normal residual.

I've had sandy bridge, ivy bridge and haswell-e systems. All of these have calculated the same residuals between runs.

It is normal. You probably ran old version of Linpack on those systems.

 

[BoMbY]

There appears to be a problem with Zen 2 (Ryzen 3000 series) running Linpack with AVX support.

Yes, the problem is it isn't running at all. For some reason your Linpack Extreme and IBT AVX Linpack are both crashing pretty much instantly on Ryzen 3000.

 

[Tibor Hazafi]

Hi all,

I have an i7-6700K CPU @ 4.0GHz and 2*8GB 2133MHz DDR4 RAM. Which are the optimal settings for me to stress test both of them?

1. amount of RAM: I don't know
2. number of times to run: I don't know
3. all available thread: I think YES
4. disable sleep mode: I think YES
5. run CPUID HWMonitor in the backgrond: I think NO, since I run it manually

And what are the safe core voltage and temperature values, which I can still safely run the program without throttling or hardware damage with?

Thanks,
hazazs

 

[Octopuss]

So why hasn't this been updated since 2018? Must be using outdated libraries just as well.

 

[toyo]

Hi all,

I have an i7-6700K CPU @ 4.0GHz and 2*8GB 2133MHz DDR4 RAM. Which are the optimal settings for me to stress test both of them?

1. amount of RAM: 10GB. If you use the 14GB experimental option, it will leak into the SSD/HDD pagefile, unless you have a very clean lightweight Windows install.
2. number of times to run: 20-30 times looks enough for a rough test. For actual stability testing, leave over night, 10-12 hrs, even 24 hrs.
3. all available thread: yes
4. disable sleep mode: not needed if you use a power plan like High Performance that disables Sleep, but if you leave it over the night you could use that to make sure it stays awake.
5. run CPUID HWMonitor in the backgrond: nah, not needed, but you should monitor with HWinfo64 while stresstesting, it looks to be the most extensive

And what are the safe core voltage and temperature values, which I can still safely run the program without throttling or hardware damage with?

Thanks,
hazazs

Added text to your questions.

Modern CPUs don't get damaged so easily. You'll have to test what's best for you step by step, incrementally. Also check a guide for Skylake, so you get a rough idea about how much it can OC on average. From what I remember 4.5-4.6 GHz were somewhat common. If it throttles, you'll need to either get better cooling or reduce voltage/frequency. I assume Skylakes have the horrible toothpaste TIM under the heatsink, which means you can probably delid them for better thermals. You'll have to have a decent VRM on the motherboard too if you start drawing a lot of power during the tests.

All in all, you'll have to test.

EDIT: noticed you seem to have a B250 chipset, which does not allow overclocking, I believe? So you will pretty much just stress-test the default config. I don't think it will even use XMP for the RAM above 2400 MHz. Depending on what's in the BIOS, you might try to increase the bus to OC slightly, since the multiplier is not there. You could also undervolt, which has pretty good results IMO, I undervolted my 8700K as well and reduced power consumption from approx. 135W to 118W (peak, under AVX2 stresstesting), it can have decent result on temps.

All in all, with this motherboard, if everything is stable, there's not much you can do unless you feel like playing with it and trying to undervolt or do that small bus OC. You could also tighten the timings on the RAM I guess to make up for the lower frequency and lack of XMP above 2.4GHz, but I'm unsure if it's really worth your time. Up to you, of course.

 

[Tibor Hazafi]

EDIT: noticed you seem to have a B250 chipset, which does not allow overclocking, I believe? So you will pretty much just stress-test the default config. I don't think it will even use XMP for the RAM above 2400 MHz. Depending on what's in the BIOS, you might try to increase the bus to OC slightly, since the multiplier is not there. You could also undervolt, which has pretty good results IMO, I undervolted my 8700K as well and reduced power consumption from approx. 135W to 118W (peak, under AVX2 stresstesting), it can have decent result on temps.

All in all, with this motherboard, if everything is stable, there's not much you can do unless you feel like playing with it and trying to undervolt or do that small bus OC. You could also tighten the timings on the RAM I guess to make up for the lower frequency and lack of XMP above 2.4GHz, but I'm unsure if it's really worth your time. Up to you, of course.

Yes, I have a B250 chipset instead of Z270, because I don't want to OC, I just simply want to stress test my system at base clocks. The XMP profile settings for the RAM are:
2133MHz (highest frequency which are supported by the motherboard with a Skylake CPU)
1,2V
CL13
My result with 10GB RAM and 30 times to run >>LINK<<
Voltage, temperature and clock values under the test according to HWMonitor >>LINK<<

Again, I don't want to OC, but what I definiately want:
Keep all core at 4GHz minimum with the possibility of the 4,2GHz Turbo on one core if needed. For this I have set the minimum processor state to 100% by enabling the "High performance" power profile in Windows. Most of the time the CPU runs at 4GHz on each core, but monitoring the clocks for 3 hours this was the result according to CPUID HWMonitor:

For some reason each core drops rarily to 800MHz, even if I set EIST to disabled. Core #0 drops to 800MHz even under the stress test. Why? It should keep the 4GHz all the time, isn't it? Or what setting did I miss?

 

[phill]

Yes, I have a B250 chipset instead of Z270, because I don't want to OC, I just simply want to stress test my system at base clocks. The XMP profile settings for the RAM are:
2133MHz (highest frequency which are supported by the motherboard with a Skylake CPU)
1,2V
CL13
My result with 10GB RAM and 30 times to run >>LINK<<
Voltage, temperature and clock values under the test according to HWMonitor >>LINK<<

Again, I don't want to OC, but what I definiately want:
Keep all core at 4GHz minimum with the possibility of the 4,2GHz Turbo on one core if needed. For this I have set the minimum processor state to 100% by enabling the "High performance" power profile in Windows. Most of the time the CPU runs at 4GHz on each core, but monitoring the clocks for 3 hours this was the result according to CPUID HWMonitor:

For some reason each core drops rarily to 800MHz, even if I set EIST to disabled. Core #0 drops to 800MHz even under the stress test. Why? It should keep the 4GHz all the time, isn't it? Or what setting did I miss?

Just a quick check at this point but what are you're temps at the time of running the tests? It might drop speeds when changing the tests very breifly I think? Been a while since I used the software to be honest... There could be other factors why your getting a drop but temps would be my first port of call

 

[R-T-B]

So why hasn't this been updated since 2018? Must be using outdated libraries just as well.

The latest MKL binaries are somehow buggy with thread binding.