New linpack from intel

Araxie said:
are you using custom problem size?.. check your SIZE and LDA are much lower than rest of test posted here O_O..
Click to expand...

You need to have >8GB for a problem size of 30000 or more. That's why mine stopped at 25000, I only have 8GB. ;)
 
livefastdieyoung said:
same here, always an error after every test (if you uncheck 'stop on error'). tested on multiple systems, some that aren't even overclocked.

probably a bug between linx and the latest version of linpack.
Click to expand...

If you notice, in the results from the people who run it directly, the value in residual and residual (norm) are different. I suspect that Linx sees that difference as an error, but I didn't check the older version to see for sure.

I dropped the new linpack files into IBT but it didn't run because the compiler file wasn't there - once I put that in it worked, but after it blue-screened the first time I went back to Linx, so I don't know if IBT works correctly with the new files or not.
 
HMB said:
If you want to run it properly on a 4 core cpu with HT enabled then change the contents of "runme_xeon64.bat" to the following:
Code: 
@echo off
SETLOCAL

rem Setting up affinity for better threading performance
set KMP_AFFINITY=nowarnings,compact,granularity=fine
rem Setting path to OpenMP library
set PATH=..\..\..\redist\intel64\compiler;%PATH%

echo Running linpack_xeon64.exe. Output can be found in win_xeon64.txt.
start /b /affinity 55 linpack_xeon64.exe lininput_xeon64 > win_xeon64.txt

echo.
echo When this window closes the calculation is done.

ENDLOCAL



Edit: I might aswell paste the numbers i got running it, i7-4770K (HT on)

Current date/time: Wed Jul 10 16:04:37 2013

CPU frequency: 4.194 GHz
Number of CPUs: -1
Number of cores: 1
Number of threads: 4
[/code]
Click to expand...

See you did all that and you only ran 4 threads. :D See. Easier to turn off HT in BIOS when running this.

HT is not needed or recommended to run this benchmark.
 
I ran with HT ON just to check how much worse can be the results without changing anything in the linpack setup.. well.. xD much worse than i thought, was exactly the half of performance with a max of 53 GFLOPS, most of the time i could see 2 cores running at 100% the others 2 in 1%. and thats is reflected in the temperatures even with a 2.5Chigher room temp.

 
SonDa5 said:
See you did all that and you only ran 4 threads. :D See. Easier to turn off HT in BIOS when running this.

HT is not needed or recommended to run this benchmark.
Click to expand...
You get better result running 4 threads, even on quadcores running HT. The thing I did earlier was assigning the threads to the physical cores, if not doing that then the first 4 cores (2 physical and 2 "HT"-cores) would get assigned and performance would be bad.

Why not modify the batch file a little so you other guys won't have to reboot and turn HT off/on?
 
HMB said:
You get better result running 4 threads, even on quadcores running HT. The thing I did earlier was assigning the threads to the physical cores, if not doing that then the first 4 cores (2 physical and 2 "HT"-cores) would get assigned and performance would be bad.

Why not modify the batch file a little so you other guys won't have to reboot and turn HT off/on?
Click to expand...



For me its easier to just boot into BIOS and turn off HT. Also if you turn off HT from the BIOS the entire system is going to be using less processing power from the virtual cores no longer being used.

Linpack is designed and optimized for 1 thread per core so for best performance its best to just turn of HT from the BIOS.
 
SonDa5 said:
For me its easier to just boot into BIOS and turn off HT. Also if you turn off HT from the BIOS the entire system is going to be using less processing power from the virtual cores no longer being used.

Linpack is designed and optimized for 1 thread per core so for best performance its best to just turn of HT from the BIOS.
Click to expand...
It's easier to go into bios, turn HT off, run benchmark, reboot into bios and enable HT compared to just changing the contents of the batch file (which takes a couple of seconds)?

I bought an i7 to get HT, not so I could turn it off to save a few watts :)
 
HMB said:
It's easier to go into bios, turn HT off, run benchmark, reboot into bios and enable HT compared to just changing the contents of the batch file (which takes a couple of seconds)?

I bought an i7 to get HT, not so I could turn it off to save a few watts :)
Click to expand...



For the Linpack benchmark you will stress the cpu more if you disable HT in the BIOS. This comes straight from Intel. Read the xhelp file.


Its just a benchmark. Virtual cores can be enjoyed in over all system processing.
 
SonDa5 said:
For the Linpack benchmark you will stress the cpu more if you disable HT in the BIOS. This comes straight from Intel. Read the xhelp file.


Its just a benchmark. Virtual cores can be enjoyed in over all system processing.
Click to expand...
I don't think you understand what I did with the batch file.

Here is with HT ON:
Code: 
Performance Summary (GFlops)

Size   LDA    Align.  Average  Maximal
1000   1000   4       118.0588 129.1411
2000   2000   4       126.8100 128.6219
3000   3000   4       139.3536 143.0048
4000   4000   4       154.8991 155.7175
5000   5000   4       161.4760 163.9433
10000  10000  4       187.9990 188.3203
15000  15000  4       197.8248 197.8607
20000  20000  4       204.4907 204.5550
25000  25000  4       204.5475 205.0333
30000  30000  4       209.1864 209.1864
35000  35000  4       210.3207 210.3207
40000  40000  4       210.8051 210.8051

Residual checks PASSED

And here is with HT OFF:
Code: 
Performance Summary (GFlops)

Size   LDA    Align.  Average  Maximal
1000   1000   4       128.6874 135.5502
2000   2000   4       128.0555 130.3411
3000   3000   4       142.3068 147.0210
4000   4000   4       154.5776 157.0803
5000   5000   4       164.9421 165.9811
10000  10000  4       189.6182 189.9491
15000  15000  4       198.9936 199.4964
20000  20000  4       205.9145 206.1138
25000  25000  4       206.6940 206.9440
30000  30000  4       210.6702 210.6702
35000  35000  4       211.6223 211.6223
40000  40000  4       211.4730 211.4730

Residual checks PASSED

The miniscule difference in results could be attributed to the fact that with HT ON I overclocked the processor from AI Suite, with HT OFF I overclocked from bios.
 
HMB said:
I don't think you understand what I did with the batch file.

Here is with HT ON:
Code: 
Performance Summary (GFlops)

Size   LDA    Align.  Average  Maximal
1000   1000   4       118.0588 129.1411
2000   2000   4       126.8100 128.6219
3000   3000   4       139.3536 143.0048
4000   4000   4       154.8991 155.7175
5000   5000   4       161.4760 163.9433
10000  10000  4       187.9990 188.3203
15000  15000  4       197.8248 197.8607
20000  20000  4       204.4907 204.5550
25000  25000  4       204.5475 205.0333
30000  30000  4       209.1864 209.1864
35000  35000  4       210.3207 210.3207
40000  40000  4       210.8051 210.8051

Residual checks PASSED

And here is with HT OFF:
Code: 
Performance Summary (GFlops)

Size   LDA    Align.  Average  Maximal
1000   1000   4       128.6874 135.5502
2000   2000   4       128.0555 130.3411
3000   3000   4       142.3068 147.0210
4000   4000   4       154.5776 157.0803
5000   5000   4       164.9421 165.9811
10000  10000  4       189.6182 189.9491
15000  15000  4       198.9936 199.4964
20000  20000  4       205.9145 206.1138
25000  25000  4       206.6940 206.9440
30000  30000  4       210.6702 210.6702
35000  35000  4       211.6223 211.6223
40000  40000  4       211.4730 211.4730

Residual checks PASSED

The miniscule difference in results could be attributed to the fact that with HT ON I overclocked the processor from AI Suite, with HT OFF I overclocked from bios.
Click to expand...

I understand and your results show that with the HT off your CPU is able to work better on the linpack batch file being processed.
 
SonDa5 said:
I understand and your results show that with the HT off your CPU is able to work better on the linpack batch file being processed.
Click to expand...
Like I said earlier, the 0,3% difference (40000 size) could be attributed to the different ways I overclocked the processor or even to different processes being run in the background.

Are you even reading what I've written?

Edit: For instance, while doing the HT test I had Asus AI Suite on, you don't think that this graphical interface steals a couple of CPU cycles?
software-oc.jpg


Edit:
Ok, I updated the HT ON results now with AI Suite 3 turned off:

HT OFF:
Code: 
Size   LDA    Align.  Average  Maximal
1000   1000   4       128.6874 135.5502
2000   2000   4       128.0555 130.3411
3000   3000   4       142.3068 147.0210
4000   4000   4       154.5776 157.0803
5000   5000   4       164.9421 165.9811
10000  10000  4       189.6182 189.9491
15000  15000  4       198.9936 199.4964
20000  20000  4       205.9145 206.1138
25000  25000  4       206.6940 206.9440
30000  30000  4       210.6702 210.6702
35000  35000  4       211.6223 211.6223
40000  40000  4       211.4730 211.4730

And HT ON:
Code: 
Size   LDA    Align.  Average  Maximal
1000   1000   4       101.6545 137.7243
2000   2000   4       128.3226 129.3443
3000   3000   4       145.0507 145.8724
4000   4000   4       155.8494 156.6988
5000   5000   4       164.4143 165.0972
10000  10000  4       190.2611 190.4261
15000  15000  4       200.2525 200.2718
20000  20000  4       207.1824 207.2691
25000  25000  4       205.4417 206.5686
30000  30000  4       211.9249 211.9249
35000  35000  4       213.1001 213.1001
40000  40000  4       213.9705 213.9705
 
Last edited:
HMB its simple. The benchmark is designed to be used with HT off from the BIOS. 1 thread per core. No virtual cores doing anything else. HT off.
 
SonDa5 said:
HMB its simple. The benchmark is designed to be used with HT off from the BIOS. 1 thread per core. No virtual cores doing anything else. HT off.
Click to expand...
The modified batch file assigns one thread to each of the physical cores, ergo, its exactly the same as running HT on as HT off, and my results proove that.

Why can't you understand anything I've written so far?
 
HMB said:
The modified batch file assigns one thread to each of the physical cores, ergo, its exactly the same as running HT on as HT off, and my results proove that.

Why can't you understand anything I've written so far?
Click to expand...


Its not the same because when you boot up into your OS the virtual cores are going to start to work. Those extra virtual cores working take up from process powers that could be focused on the benchmark.

I understand what you and I understand how Intel optimized this benchmark to run optimized with only "1 thread per core".
 
SonDa5 said:
Its not the same because when you boot up into your OS the virtual cores are going to start to work. Those extra virtual cores working take up from process powers that could be focused on the benchmark.

I understand what you and I understand how Intel optimized this benchmark to run optimized with only "1 thread per core".
Click to expand...
And you mean that simply because you're benchmarking, those processes won't affect a CPU running HT OFF? You think that the 4 physical cores running the benchmark aren't doing anything else in the meantime?

I'm sorry to say this but you have no idea what you're talking about here.

The benchmarks I have provided prove my point, don't bother replying unless you have some proof of what you're talking about. And no, the readme in the benchmark directory doesn't count.
 
Last edited:
dandragonrage said:
Yeah, well when everyone's recommending a stress test utility made in 1989 (slight exaggeration) of course it's not going to stress the chip enough compared to a custom modification of software (that's better to begin with) by Intel to specifically utilize as much of their chips as possible. Glad to hear this one works well for you also... I'll give it a shot on my home machine tonight as I've been having some stability issues that have been difficult to pin down.
Click to expand...

I still dont understand 1 thing.
WHY do you WANT your chips to reach 99 C in a non real world application that NO, NO software on the PLANET would EVER cause your chips to reach?

It's like DESTROYING GPUs' by running furmark.
 
Falkentyne said:
I still dont understand 1 thing.
WHY do you WANT your chips to reach 99 C in a non real world application that NO, NO software on the PLANET would EVER cause your chips to reach?

It's like DESTROYING GPUs' by running furmark.
Click to expand...

Kinda remember how my multiplayer Mafia game with heavy artillery was interrupted because one player's laptop reached 93 degrees. He used it as a stop gap because his desktop broke.

Also when I played Legends of Dawn fan on my card was as loud as when I ran Furmark on it... So perhaps you are not playing correct games.
 
Falkentyne said:
I still dont understand 1 thing.
WHY do you WANT your chips to reach 99 C in a non real world application that NO, NO software on the PLANET would EVER cause your chips to reach?

It's like DESTROYING GPUs' by running furmark.
Click to expand...

play crysis 3 and you will see in GPU and CPU same things like running IBT, LinX or P95.. or running Furmark in GPU.. :). lol even play GTA IV that still burn GPU in this times..
 
Falkentyne said:
I still dont understand 1 thing.
WHY do you WANT your chips to reach 99 C in a non real world application that NO, NO software on the PLANET would EVER cause your chips to reach?

It's like DESTROYING GPUs' by running furmark.
Click to expand...

You don't want it to reach 99C, but what you do want to do is put a more stressful than normal load on the system to accelerate the error finding process. I don't like running IBT or Prime95, but I also don't like playing games and getting random crashes because my supposedly stable overclock isn't actually stable. It's just the lesser of two evils.

We have pretty extensive evidence that a chip that is Prime stable is also going to be game or application stable, but you can't say that about lesser tests like Aida. Plenty of anecdotal evidence, including my own, that passing Aida for extensive periods does not provide game stability.
 
Falkentyne said:
I still dont understand 1 thing.
WHY do you WANT your chips to reach 99 C in a non real world application that NO, NO software on the PLANET would EVER cause your chips to reach?

It's like DESTROYING GPUs' by running furmark.
Click to expand...



I run furmark on my GPUs as well. :D


I only do it briefly and I do it to know what over clock is stable and to also see what I have to do to improve cooling.
 
Araxie said:
play crysis 3 and you will see in GPU and CPU same things like running IBT, LinX or P95.. or running Furmark in GPU.. :). lol even play GTA IV that still burn GPU in this times..
Click to expand...

Crysis 3 won't necessarily load all of your cores, so it wouldn't even be close to the same thing in certain cases.
 
Same after one pass it gives a stop error. This is running at 3500mhz and 1600mhz ram.
Running 64bit version.
 
Here is my 4670K @ 1.28v. Prime will crash it at some point though.

Intel(R) Optimized LINPACK Benchmark data

Current date/time: Fri Jul 19 20:03:13 2013

CPU frequency: 4.498 GHz
Number of CPUs: 1
Number of cores: 4
Number of threads: 4

Parameters are set to:

Number of tests: 12
Number of equations to solve (problem size) : 1000 2000 3000 4000 5000 10000 15000 20000 25000 30000 35000 40000
Leading dimension of array : 1000 2000 3000 4000 5000 10000 15000 20000 25000 30000 35000 40000
Number of trials to run : 4 4 4 4 4 2 2 2 2 1 1 1
Data alignment value (in Kbytes) : 4 4 4 4 4 4 4 4 4 4 4 4

Maximum memory requested that can be used=4210869504, at the size=40000

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

Size LDA Align. Time(s) GFlops Residual Residual(norm) Check
1000 1000 4 0.006 107.4953 1.194739e-012 4.074366e-002 pass
1000 1000 4 0.005 127.1921 1.194739e-012 4.074366e-002 pass
1000 1000 4 0.005 122.0351 1.194739e-012 4.074366e-002 pass
1000 1000 4 0.005 128.3316 1.194739e-012 4.074366e-002 pass
2000 2000 4 0.041 129.7615 4.536926e-012 3.946570e-002 pass
2000 2000 4 0.041 130.6524 4.536926e-012 3.946570e-002 pass
2000 2000 4 0.041 131.2825 4.536926e-012 3.946570e-002 pass
2000 2000 4 0.041 131.1338 4.536926e-012 3.946570e-002 pass
3000 3000 4 0.122 147.6352 8.334888e-012 3.209566e-002 pass
3000 3000 4 0.125 144.4000 8.334888e-012 3.209566e-002 pass
3000 3000 4 0.122 147.8019 8.334888e-012 3.209566e-002 pass
3000 3000 4 0.127 141.7619 8.334888e-012 3.209566e-002 pass
4000 4000 4 0.273 156.2542 1.519912e-011 3.312792e-002 pass
4000 4000 4 0.267 160.0726 1.519912e-011 3.312792e-002 pass
4000 4000 4 0.286 149.4218 1.519912e-011 3.312792e-002 pass
4000 4000 4 0.304 140.2918 1.519912e-011 3.312792e-002 pass
5000 5000 4 0.503 165.6539 2.471656e-011 3.446525e-002 pass
5000 5000 4 0.499 166.9604 2.471656e-011 3.446525e-002 pass
5000 5000 4 0.498 167.3805 2.471656e-011 3.446525e-002 pass
5000 5000 4 0.495 168.6157 2.471656e-011 3.446525e-002 pass
10000 10000 4 3.810 175.0433 9.436774e-011 3.327502e-002 pass
10000 10000 4 3.945 169.0364 9.436774e-011 3.327502e-002 pass
15000 15000 4 12.754 176.4468 2.169435e-010 3.416896e-002 pass
15000 15000 4 13.293 169.2911 2.169435e-010 3.416896e-002 pass
20000 20000 4 30.707 173.7109 3.504283e-010 3.102058e-002 pass
20000 20000 4 31.723 168.1452 3.504283e-010 3.102058e-002 pass
25000 25000 4 60.302 172.7616 5.194889e-010 2.954147e-002 pass
25000 25000 4 60.529 172.1140 5.194889e-010 2.954147e-002 pass
30000 30000 4 104.493 172.2778 7.133177e-010 2.811906e-002 pass
35000 35000 4 165.483 172.7412 1.085449e-009 3.150893e-002 pass
40000 40000 4 246.586 173.0427 1.338275e-009 2.976370e-002 pass

Performance Summary (GFlops)

Size LDA Align. Average Maximal
1000 1000 4 121.2635 128.3316
2000 2000 4 130.7075 131.2825
3000 3000 4 145.3998 147.8019
4000 4000 4 151.5101 160.0726
5000 5000 4 167.1526 168.6157
10000 10000 4 172.0398 175.0433
15000 15000 4 172.8689 176.4468
20000 20000 4 170.9281 173.7109
25000 25000 4 172.4378 172.7616
30000 30000 4 172.2778 172.2778
35000 35000 4 172.7412 172.7412
40000 40000 4 173.0427 173.0427

Residual checks PASSED

End of tests

Fri 07/19/2013
08:19 PM
Click to expand...
 
Finally got the command line version working.
Ran it for 4 hours. reached 99c, but no throttling or crashes.

4300mhz, 1.22v, uncore (cache ratio) 3400. It took a while for it to hit 99c, maybe a few hours not sure.
 
I thought there was something seriously wrong with my system when I compared my GFLOPS to some of the ones posted here, until I realized I'm looking at my IB numbers compared to Haswell numbers with AVX2. Are there any programs/games that make use of AVX2 yet?
 
I can't really use this new linpack for testing. I noticed if I up my ram speed 2133 which is the speed of my ram the test heats up to 99 nearly instantly.

I'd have to lower the overclock to around 4000 probably to get the heat under control. But that be useless since I don't know of any real world application that will heat your cpu to 99c and throttle it in under 1min. And that's with a room temp of 70F.
 
That's like saying I can't use a program that brings out the weaknesses for testing. Isn't that kind of the point of stress testing? What good is testing if you're going to limit it to applications that won't stress your setup?
 
Cause linpak is overkill way overkill. Even delidders are reaching close to 99c with this. Basically its a program to heat your chip up way beyond any real program will.
 
Not even sure if I should give this a try considering how I'm already 24hr Prime stable and don't crash at all in games.
 
timmylol said:
Not even sure if I should give this a try considering how I'm already 24hr Prime stable and don't crash at all in games.
Click to expand...

IF you are running the 3770K in your sig, then there is no point in running in, since our cpu's only support AVX, not AVX2..
 
SonDa5 said:
This new avx2 linpack makes prime95 whimpy.
Click to expand...

Except that Prime95 is still more useful for normal stability. Linpack goes to extremes that nothing other than Linpack will ever stress so much.
 
Forceman said:
Until applications that use AVX2 come out, that is.
Click to expand...

Almost nothing ever gets a 3770K as hot as linpack/intel burn test either. By the time something actually uses AVX2, new CPU's will probably be out again.

I'm just saying, most people get a 4770k, or especially a 4670k, to play games, and there is absolutely no point to stress test a CPU this hard just for gaming.
 
My cores topped out at 93/98/97/89 C. 4770K @ 4.3 GHz / 1.218V, HT off (for running the test).

I ran the Aida64 FPU-test before starting the linpack test, also with HT off, and got a max of 83/84C, and that test is already considered brutal :eek: Even with HT On that "only" goes to about 87C max.
 
You must log in or register to reply here.
Back
Top