Linpack Xtreme Released

[Solid]

The TS Bench is a 32 bit benchmark which only uses x86 code so it does not use any AVX instructions. Basically it is testing a completely different part of the CPU compared to where the 64 bit AVX instructions are running when Linpack testing.

Interesting that the TS Bench needs slightly more voltage to run reliably compared to Linpack testing. These are two completely different tests so when you think about it, slightly different voltage requirements should not be too surprising. A couple of other users have reported that some quick TS Bench testing has helped them get their CPU voltage in the ball park a lot quicker compared to the more popular AVX based tests.

The TS Bench does millions of calculations, stops, and then does the exact same calculations again. The CPU cores are usually a little hotter second time around. If it finds two different answers to the same question, it reports this as an error. Any mathematical calculation should return the exact same answer no matter how many times you run the test, no matter what core the calculation is run on or where in memory the answer is saved. A 100% stable computer should be able to pass any of these tests.

You might also want to try doing some stability testing with turbo boost disabled. The 8750H runs at only 2.20 GHz when Intel Turbo Boost is disabled so its offset voltage requirement at this speed could be significantly different compared to when it is running fully loaded at 3.90 GHz. An offset voltage that passes full load, full speed testing, might be inadequate when running fully loaded at only 2.20 GHz. Some single thread stability testing is also important. This will allow the 8750H to hits its maximum speed of 4.10 GHz. Watch the reported MHz in ThrottleStop closely since it is one of the few monitoring apps that reports single thread MHz testing like this correctly.


That is kind of an insult.
Your testing has shown that the TS Bench can detect errors quicker compared to Linpack testing. It also tests parts of your CPU that more popular testing software does not. That sounds like a pretty good test to me.

Sorry of course i didn't want to insult your work, i've just heard this but I think Throttlestop is the best app to do this, otherwise I would't have chosen it.
Anyway @unclewebb, i just wanted to ask to you what do you suggest to me:
TS BENCH say to me 0 errors at 1024M if I undervolt my DELL of around -105mV for core and cache (it seems to me too few for this cpu), but I can undervolt core of -130mV and cache of -105mV and I obtain the same 0 errors.
Every people with this notebook claim an undervolt of around -140mV for core and cache.
If I run Linpack I can pass the test for 3 hours using 10GB of RAM (35.000 of problem size) with an undervolting of -140mV for core and cache, but my pc tends to freeze if i try also at -105mV with 30 GB of RAM (but i'm not sure why).
Looking at this situation, can you give me some advice?? How many mVs should I go with??
Thank you!

Its a pretty heavy duty app, I cant say I am entirely surprised to see systems like Dells not being able to run it at full capacity. I was actually considering getting an XPS all done up it was such a good deal, especially with a 1070ti, even though you have to wait about 30 days for shipping vs 5 if you chose the 1060, 70, or 80. But I looked at the internal shots, and wasn't convinced.

I'm satisfied with this machine, it's beautiful but Dell should solve heat problems.

 

[EarthDog]

Every cpu is different. Just because one can undervolt a certain amount, doesnt mean yours will. They also may have used a different stress testing application (or none at all) which can get away with less voltage when running. Too many variables to worry about other results.

Pick your test and whatever it ends up at it does. It should be clear that this stress test isn't for you since you are getting all those throttling reasons...in particular the current limit and power limits.

 

[Regeneration]

Looking at this situation, can you give me some advice?? How many mVs should I go with??

Enough voltage for the computer to pass the test.

It is possible your CPU requires more voltage because you have 32GB of RAM.

More RAM puts more stress on the integrated memory controller.

 

[unclewebb]

Every people with this notebook claim an undervolt of around -140mV for core and cache.

The amount of voltage other people are using is nice to know and will give you some idea of where you can start testing but beyond that, it is not important. When it comes to how much voltage your CPU needs to be 100% stable, do lots of testing and give your CPU as much voltage as it needs. After initial break in, it might require a little more voltage. Any errors in any reputable benchmark are telling you that your CPU is not stable. Stability is important to me so I always give my CPUs a little bit of extra voltage so I can pass these tests.

The majority of users probably never run benchmarks or simply don't care if they are Linpack or Prime95 or TS Bench stable. As long as their computer doesn't crash too often while playing a game, they are happy. An undervolt of -140 mV might be fine for them and if this is how you mostly use your laptop, a less than 100% stable undervolt might be OK for you too.

It is totally up to you. If you are running the local nuclear power plant with your laptop, don't be cheap with the voltage. If you need to talk some more about voltage, you should start a new discussion thread.

 

[Solid]

thank you

 

[toyo]

@Regeneration:
Windows Defender deleted the x64 executable yesterday. Can anything be done to remove it from the databases/heuristics? I assume it's just a false detection.

 

[Regeneration]

Sent a false positive report to Microsoft.



Shows clean on their website with definition 1.283.1844.0.

 

[toyo]

Thanks. I rebooted and updated Defender, and downloaded the file again, and no detection this time.

 

[hat]

This thread needs a bump. Possibly a sticky? I thought of it again and the only way I found it was by using the Search function... of course, to do that, I already had to know what Linpack Xtreme was...

 

[KWSW]

Was searching around on information on linpack xtreme and chanced upon this thread. Agree with the above poster that a sticky would be good

Have a question, which problem size should I be choosing and how many runs to check for stability?

Thanks in advance!

 

[Regeneration]

80 percent of your RAM or maximum of 10GB. 1-4 hours for "basic" stability or 8-12 hours for "rock" stable. Just enter unlimited runs and stop it manually.

 

[phill]

Was searching around on information on linpack xtreme and chanced upon this thread. Agree with the above poster that a sticky would be good

Have a question, which problem size should I be choosing and how many runs to check for stability?

Thanks in advance!

Firstly, welcome to TPU

Secondly, depends on how well you wish to check your system to be honest. Too small a problem size wouldn't really push the limits of it and too few a runs wouldn't test it or heat it up for long enough. I'd go with a high problem size and then at least 10 to 15 runs of the test

 

[KWSW]

Thank you Regeneration and phill for the replies! So i don't need to run the 30GB option?

I am just aiming for mild overclocks for mostly gaming with a side of photo post processing. Probably 3.9ghz on my 1800x cos leaving it at stock is kind of a waste and maybe just seeing how much I can tighten the timings on my Corsair Vengeance LPX 32GB kit.

In this case, would 1 hour be good? And also would the test also check my rams at the same time? I do intent to still run a test with the google stress app test for safe checking.

 

[phill]

Have you ever tried Memtest @KWSW ? That might be a little better for just the RAM testing, when that's stable you could move on to the Linpack testing.. I'm not sure how much higher the 1800X's go but I'd do one thing at once with the overclocking and then you won't be pulling your hair out when it crashes out. Do you know if the Corsair kit is B die based RAM or not?

I find doing some basic tests first always help, CB 15/20 etc, things like that which will stress the system a little, then using Linpack could finish the overclock testing off

Feel free to post back if you get any issues, people here are always around to help

 

[KWSW]

Have you ever tried Memtest @KWSW ? That might be a little better for just the RAM testing, when that's stable you could move on to the Linpack testing.. I'm not sure how much higher the 1800X's go but I'd do one thing at once with the overclocking and then you won't be pulling your hair out when it crashes out. Do you know if the Corsair kit is B die based RAM or not?

I find doing some basic tests first always help, CB 15/20 etc, things like that which will stress the system a little, then using Linpack could finish the overclock testing off

Feel free to post back if you get any issues, people here are always around to help

would that be memtest86 or hci memtest? there is also y-cruncher and testmem5. which one in your experience would be the best to test stability?

I know stability is also subjective based on what I read on the internet, for me since my PC is not for "mission critical" work, as long as can play my games and do some light photo editing, do I really need to run these programs like linpack xtreme overnight or 1-2 hours is good enough for most everyday workload?

 

[Guest1]

Your testing has shown that the TS Bench can detect errors quicker compared to Linpack testing. It also tests parts of your CPU that more popular testing software does not. That sounds like a pretty good test to me.

Kind of having the opposite experience here, TS Bench is working fine while Linpack Xtreme Benchmark fails even with default CPU settings.

Size   LDA    Align. Time(s)    GFlops   Residual     Residual(norm) Check
20000  20000  4      25.338     210.5182 6.102235e-005 5.401815e+003   FAIL
20000  20000  4      25.430     209.7547 2.380633e-010 2.107382e-002   pass

Using a 8700K on Windows 1903, maybe a problem with the microcode update?

 

[toyo]

Kind of having the opposite experience here, TS Bench is working fine while Linpack Xtreme Benchmark fails even with default CPU settings.

Size   LDA    Align. Time(s)    GFlops   Residual     Residual(norm) Check
20000  20000  4      25.338     210.5182 6.102235e-005 5.401815e+003   FAIL
20000  20000  4      25.430     209.7547 2.380633e-010 2.107382e-002   pass

Using a 8700K on Windows 1903, maybe a problem with the microcode update?

I have the same(ish) config (new microcode 0xB4, new MDS mitigations enabled, 1903), I can try a quick test. Will edit in 10-20 mins.

Took me a few tries to realize you're running the 3GB bench, not the stability tests. Here's my stock 8700K on that. I think the microcode is fine, no crashes, no stability issues, no BSODs.

PS: not only your PC is unstable, but you also seem to have some sort of performance issue, there's shouldn't be 45-ish Gflops between the same stock CPUs. Could be that you have power capped in BIOS.

 

[Guest1]

PS: not only your PC is unstable, but you also seem to have some sort of performance issue, there's shouldn't be 45-ish Gflops between the same stock CPUs. Could be that you have power capped in BIOS.

That's the AVX offset reducing the clock, but the thing is that TS Bench and Prime95 works fine for me, only Linpack causes problems. (Currently testing OCCT Linpack with AVX).
Not sure if this is a hardware or software issue.

 

[toyo]

That's the AVX offset reducing the clock, but the thing is that TS Bench and Prime95 works fine for me, only Linpack causes problems.
Not sure if this is a hardware or software issue.

You shouldn't need to use an AVX offset for stock operation. That for high OCs where the CPU starts to draw insane amounts of power and it becomes almost impossible to cool. You can cool a 8700K under stress testing, with a very cheap cooler (30$), up to about 150W drawn. That's about 4.5-4-7GHz under prime95, depending on your silicon lottery luck. At stock it should run 4.3GHz when under stress. If you see the clocks going down, and you haven't manually set the AVX offset, you need to take a look at the power limits, those can be set at 95W (as it should as per Intel spec, yet some manufacturers remove the limits). If that's the case, under very heavy load the clocks will drop to around 3.8-4.0GHz so the power limit is respected.

As for what's causing instability, could be anything from the typical unstable OCs or undervolts, to badly implemented XMPs. Poorly implemented XMP with stupid high voltages tormented me for almost a year, and I wasted hundreds of hours on stresstesting until I got it stable. Linpack itself, I'd exclude it from being the cause. Probably prime95 (with the correct settings to create the same amount of stress as linpack) would fail too.

 

[Regeneration]

That's the AVX offset reducing the clock, but the thing is that TS Bench and Prime95 works fine for me, only Linpack causes problems. (Currently testing OCCT Linpack with AVX).
Not sure if this is a hardware or software issue.

Sounds like a memory / IMC problem.

 

[ehume]

First of all, I'd like to thank OP for putting in the work to do this. I had to pass on OCCT because their stress test with AVX ran too cool -- about 10c cooler than LinX 0.6.5. LinX 0.6.5 advertised itself as running AVX2, so I thought that was the best you could get.

I have looked through all 10 pages of this thread, and OP not once mentions what AVX this app uses, AVX2 or AVX-512. I would like to know.

 

[freeagent]

Kind of having the opposite experience here, TS Bench is working fine while Linpack Xtreme Benchmark fails even with default CPU settings.

Could be a bad setting for ram, might not like a certain divider you are using. Linpack is sensitive to these things. But a fail is a fail, you need to fix it, as its not linpack, its your system. Could be more vcore, imc juice, mem juice, timings, who knows. All of these mitigations do a number to your rig. And you probably wouldn't know until you experience instability, like I did.

This is my favorite stress testing app, It hammers your rig like no other. Things that I passed before (like IBT, Linx 0.6.5) wasn't passing with this. Personally, I find if I can pass this, everything else is no problem. (so far)

Also, prime 95 can suck it. That app needs a trip to the glue factory.

 

[Regeneration]

First of all, I'd like to thank OP for putting in the work to do this. I had to pass on OCCT because their stress test with AVX ran too cool -- about 10c cooler than LinX 0.6.5. LinX 0.6.5 advertised itself as running AVX2, so I thought that was the best you could get.

I have looked through all 10 pages of this thread, and OP not once mentions what AVX this app uses, AVX2 or AVX-512. I would like to know.

AVX-512.

 

[rubiconlexicon]

Sometimes after running for a while, the console window goes black and I can't see anything. Test seems to continue running according to CPU usage, but without any information I have to just close and restart. Any fix?

Also, if I have 16GB of RAM, should I use the 14GB option or stick to 10GB?

 

[toyo]

Sometimes after running for a while, the console window goes black and I can't see anything. Test seems to continue running according to CPU usage, but without any information I have to just close and restart. Any fix?

Also, if I have 16GB of RAM, should I use the 14GB option or stick to 10GB?

If you use 14GB on 16GB RAM it will spill in the pagefile. Windows and your software occupy some RAM too, quite a bit of it. If you're browsing or doing anything during the test, that will increase further.
As for the black window, that can be cause by running out of actual RAM, I've seen it on other testing apps.

By the way, even 10GB might prove too much if you have browser with a few tabs open. Monitor the pagefile.