Haswell overclocking discussion.

[StinkyPinky]

This is my overclock. I originally had a 4770k that wouldn't go above 4.2Ghz and wasn't stable stock on Prime 95 with different sticks of RAM, memtest, everything. I RMA'd it and got one that now won't go above 4.5Ghz, but is completely stable, which I am OK with, though I really dislike odd numbers and wish it would hit 4.6Ghz at temperatures I could handle.

I haven't really been paying attention to over-clocking since I did my 2500k, but is 98c now an acceptable temp for a core to reach? That seems too high?

 

[fastamdman]

1.65v? that seems VERY high.

I meant to say 1.165, my bad. So used to other voltages lol.

 

[ehume]

[FONT=&quot]Feature---Set To------Reading[/FONT]
[FONT=&quot]BCLK-----100MHz----100+MHz[/FONT]
[FONT=&quot]CPU-------4.5GHz----4500+MHz[/FONT]
[FONT=&quot]RAM------1600MHz--1600+MHz[/FONT]
[FONT=&quot]Uncore----4.5GHz-----na[/FONT]
[FONT=&quot]Vcore-----1.250V-----1.260V[/FONT]
[FONT=&quot]Vccin-----1.650v------na (normal = 1.8v)
[/FONT]
[FONT=&quot]Vrin-------1.200v------1.644v (reads Vccin)
[/FONT]
[FONT=&quot]Vrin LLC---Extreme---na[/FONT]
[FONT=&quot]CPU LLC--Auto------na[/FONT]
[FONT=&quot]VAXG------Auto------0.12V[/FONT]
[FONT=&quot]Vdram------Auto-----1.488V[/FONT]
[FONT=&quot]PCH------1.090V-----na[/FONT]
[FONT=&quot]Sys Agent---Auto---na[/FONT]
[FONT=&quot]I/O Analog---Auto---na[/FONT]
[FONT=&quot]I/O Digital---Auto---na[/FONT]

Tested with LinX + AVX

The hottest core temp reflects the fact that this chip has not been delidded. I don't expect it ever to work this hard in normal use. With standard LinX its hottest core is 72c.

As best I can tell, this cpu can run at 4.7GHz. It certainly can run standard LinX at that OC. But LinX with AVX causes it to throttle, so 4.5 is the best I can get.

 

[24601]

[FONT=&quot]Feature---Set To------Reading[/FONT]
[FONT=&quot]BCLK-----100MHz----100+MHz[/FONT]
[FONT=&quot]CPU-------4.5GHz----4500+MHz[/FONT]
[FONT=&quot]RAM------1600MHz--1600+MHz[/FONT]
[FONT=&quot]Uncore----4.5GHz-----na[/FONT]
[FONT=&quot]Vcore-----1.250V-----1.260V[/FONT]
[FONT=&quot]Vccin-----1.650v------na (normal = 1.8v)
[/FONT]
[FONT=&quot]Vrin-------1.200v------1.644v (reads Vccin)
[/FONT]
[FONT=&quot]Vrin LLC---Extreme---na[/FONT]
[FONT=&quot]CPU LLC--Auto------na[/FONT]
[FONT=&quot]VAXG------Auto------0.12V[/FONT]
[FONT=&quot]Vdram------Auto-----1.488V[/FONT]
[FONT=&quot]PCH------1.090V-----na[/FONT]
[FONT=&quot]Sys Agent---Auto---na[/FONT]
[FONT=&quot]I/O Analog---Auto---na[/FONT]
[FONT=&quot]I/O Digital---Auto---na[/FONT]

Tested with LinX + AVX

The hottest core temp reflects the fact that this chip has not been delidded. I don't expect it ever to work this hard in normal use. With standard LinX its hottest core is 72c.

As best I can tell, this cpu can run at 4.7GHz. It certainly can run standard LinX at that OC. But LinX with AVX causes it to throttle, so 4.5 is the best I can get.

You may want to use Intel's newest compile of linpack or else you just aren't stressing the core anywhere near the maximum.

http://software.intel.com/en-us/articles/intel-math-kernel-library-linpack-download

I tested again with newest Intel Math Kernal Library Linpack (w_lpk_p_11.0.5.009) and I get:

Ambient Temp: 72F

Maximum Temps:
Core #0: 91C
Core #1: 100C
Core #2: 100C
Core #2: 92C
Package: 101C

3570k@4.6ghz@1.296v|Thermalright True Spirit 140
4x8gB Crucial Ballistix Sport LP 1.3v@2000mhz 9-9-9-24-2N@1.635v

I would guess that Haswell will get much hotter in this since it has single clock FMA + exactly the bandwidth to take advantage of it, along with the widening of the front end and various other increases in resources.

Intel(R) Optimized LINPACK Benchmark data

Current date/time: Fri Jul 05 01:53:34 2013

CPU frequency:    4.599 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      0.008      80.1420  1.029343e-012 3.510325e-002   pass
1000   1000   4      0.009      78.4243  1.029343e-012 3.510325e-002   pass
1000   1000   4      0.010      66.1790  1.029343e-012 3.510325e-002   pass
1000   1000   4      0.008      85.3792  1.029343e-012 3.510325e-002   pass
2000   2000   4      0.068      78.6838  4.298950e-012 3.739560e-002   pass
2000   2000   4      0.069      76.9367  4.298950e-012 3.739560e-002   pass
2000   2000   4      0.067      79.9057  4.298950e-012 3.739560e-002   pass
2000   2000   4      0.071      75.7190  4.298950e-012 3.739560e-002   pass
3000   3000   4      0.213      84.5168  8.755385e-012 3.371489e-002   pass
3000   3000   4      0.223      80.6831  8.755385e-012 3.371489e-002   pass
3000   3000   4      0.220      82.0125  8.755385e-012 3.371489e-002   pass
3000   3000   4      0.223      80.7084  8.755385e-012 3.371489e-002   pass
4000   4000   4      0.407      104.9645 1.896949e-011 4.134580e-002   pass
4000   4000   4      0.406      105.2312 1.896949e-011 4.134580e-002   pass
4000   4000   4      0.428      99.8058  1.896949e-011 4.134580e-002   pass
4000   4000   4      0.409      104.3529 1.896949e-011 4.134580e-002   pass
5000   5000   4      0.787      105.9639 2.581643e-011 3.599893e-002   pass
5000   5000   4      0.767      108.6443 2.581643e-011 3.599893e-002   pass
5000   5000   4      0.768      108.6290 2.581643e-011 3.599893e-002   pass
5000   5000   4      0.812      102.6540 2.581643e-011 3.599893e-002   pass
10000  10000  4      5.899      113.0406 9.603002e-011 3.386116e-002   pass
10000  10000  4      5.706      116.8768 9.603002e-011 3.386116e-002   pass
15000  15000  4      18.656     120.6298 2.042799e-010 3.217442e-002   pass
15000  15000  4      18.671     120.5335 2.042799e-010 3.217442e-002   pass
20000  20000  4      43.234     123.3770 4.097986e-010 3.627616e-002   pass
20000  20000  4      43.113     123.7243 4.097986e-010 3.627616e-002   pass
25000  25000  4      83.771     124.3612 6.089565e-010 3.462917e-002   pass
25000  25000  4      84.781     122.8808 6.089565e-010 3.462917e-002   pass
30000  30000  4      143.948    125.0578 8.421348e-010 3.319704e-002   pass

 

[Silverforce11]

Those temps...

Somehow I won't be surprised if anytime from now, these heavily OC Haswell suddenly blow up and die, forums filled with rage and Intel doing an pre-RMA armageddon recall.

 

[BallaTheFeared]

You may want to use Intel's newest compile of linpack or else you just aren't stressing the core anywhere near the maximum.

http://software.intel.com/en-us/articles/intel-math-kernel-library-linpack-download

I tested again with newest Intel Math Kernal Library Linpack (w_lpk_p_11.0.5.009) and I get:

Ambient Temp: 72F

Maximum Temps:
Core #0: 91C
Core #1: 100C
Core #2: 100C
Core #2: 92C
Package: 101C

3570k@4.6ghz@1.296v|Thermalright True Spirit 140
4x8gB Crucial Ballistix Sport LP 1.3v@2000mhz 9-9-9-24-2N@1.635v

I would guess that Haswell will get much hotter in this since it has single clock FMA + exactly the bandwidth to take advantage of it, along with the widening of the front end and various other increases in resources.

Intel(R) Optimized LINPACK Benchmark data

Current date/time: Fri Jul 05 01:53:34 2013

CPU frequency:    4.599 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      0.008      80.1420  1.029343e-012 3.510325e-002   pass
1000   1000   4      0.009      78.4243  1.029343e-012 3.510325e-002   pass
1000   1000   4      0.010      66.1790  1.029343e-012 3.510325e-002   pass
1000   1000   4      0.008      85.3792  1.029343e-012 3.510325e-002   pass
2000   2000   4      0.068      78.6838  4.298950e-012 3.739560e-002   pass
2000   2000   4      0.069      76.9367  4.298950e-012 3.739560e-002   pass
2000   2000   4      0.067      79.9057  4.298950e-012 3.739560e-002   pass
2000   2000   4      0.071      75.7190  4.298950e-012 3.739560e-002   pass
3000   3000   4      0.213      84.5168  8.755385e-012 3.371489e-002   pass
3000   3000   4      0.223      80.6831  8.755385e-012 3.371489e-002   pass
3000   3000   4      0.220      82.0125  8.755385e-012 3.371489e-002   pass
3000   3000   4      0.223      80.7084  8.755385e-012 3.371489e-002   pass
4000   4000   4      0.407      104.9645 1.896949e-011 4.134580e-002   pass
4000   4000   4      0.406      105.2312 1.896949e-011 4.134580e-002   pass
4000   4000   4      0.428      99.8058  1.896949e-011 4.134580e-002   pass
4000   4000   4      0.409      104.3529 1.896949e-011 4.134580e-002   pass
5000   5000   4      0.787      105.9639 2.581643e-011 3.599893e-002   pass
5000   5000   4      0.767      108.6443 2.581643e-011 3.599893e-002   pass
5000   5000   4      0.768      108.6290 2.581643e-011 3.599893e-002   pass
5000   5000   4      0.812      102.6540 2.581643e-011 3.599893e-002   pass
10000  10000  4      5.899      113.0406 9.603002e-011 3.386116e-002   pass
10000  10000  4      5.706      116.8768 9.603002e-011 3.386116e-002   pass
15000  15000  4      18.656     120.6298 2.042799e-010 3.217442e-002   pass
15000  15000  4      18.671     120.5335 2.042799e-010 3.217442e-002   pass
20000  20000  4      43.234     123.3770 4.097986e-010 3.627616e-002   pass
20000  20000  4      43.113     123.7243 4.097986e-010 3.627616e-002   pass
25000  25000  4      83.771     124.3612 6.089565e-010 3.462917e-002   pass
25000  25000  4      84.781     122.8808 6.089565e-010 3.462917e-002   pass
30000  30000  4      143.948    125.0578 8.421348e-010 3.319704e-002   pass

Your flops are crazy low!

Intel(R) Optimized LINPACK Benchmark data

Current date/time: Mon Jun 24 19:14:37 2013

CPU frequency: 4.597 GHz
Number of CPUs: 1
Number of cores: 4
Number of threads: 4

Parameters are set to:

Number of tests: 9
Number of equations to solve (problem size) : 1000 2000 3000 4000 5000 10000 15000 20000 25000
Leading dimension of array : 1000 2000 3000 4000 5000 10000 15000 20000 25000
Number of trials to run : 4 4 4 4 4 2 2 2 2
Data alignment value (in Kbytes) : 4 4 4 4 4 4 4 4 4

Maximum memory requested that can be used=705536800, at the size=25000

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

Size LDA Align. Time(s) GFlops Residual Residual(norm) Check
1000 1000 4 0.007 100.0239 1.002198e-012 3.417754e-002 pass
1000 1000 4 0.007 98.9791 1.002198e-012 3.417754e-002 pass
1000 1000 4 0.007 94.5137 1.002198e-012 3.417754e-002 pass
1000 1000 4 0.007 101.0553 1.002198e-012 3.417754e-002 pass
2000 2000 4 0.035 152.2219 4.327261e-012 3.764187e-002 pass
2000 2000 4 0.036 146.6798 4.327261e-012 3.764187e-002 pass
2000 2000 4 0.036 146.4302 4.327261e-012 3.764187e-002 pass
2000 2000 4 0.037 143.0490 4.327261e-012 3.764187e-002 pass
3000 3000 4 0.114 157.6002 9.653528e-012 3.717342e-002 pass
3000 3000 4 0.112 160.2734 9.653528e-012 3.717342e-002 pass
3000 3000 4 0.108 166.9275 9.653528e-012 3.717342e-002 pass
3000 3000 4 0.110 163.3613 9.653528e-012 3.717342e-002 pass
4000 4000 4 0.238 179.5631 1.565487e-011 3.412126e-002 pass
4000 4000 4 0.246 173.8406 1.565487e-011 3.412126e-002 pass
4000 4000 4 0.237 179.9607 1.565487e-011 3.412126e-002 pass
4000 4000 4 0.237 180.3108 1.565487e-011 3.412126e-002 pass
5000 5000 4 0.450 185.1852 2.512907e-011 3.504046e-002 pass
5000 5000 4 0.451 184.7307 2.512907e-011 3.504046e-002 pass
5000 5000 4 0.449 185.6883 2.512907e-011 3.504046e-002 pass
5000 5000 4 0.451 185.0557 2.512907e-011 3.504046e-002 pass
10000 10000 4 3.358 198.6099 1.012627e-010 3.570624e-002 pass
10000 10000 4 3.278 203.4404 1.012627e-010 3.570624e-002 pass
15000 15000 4 12.120 185.6833 1.979849e-010 3.118294e-002 pass
15000 15000 4 12.153 185.1834 1.979849e-010 3.118294e-002 pass
20000 20000 4 24.397 218.6350 3.564028e-010 3.154945e-002 pass
20000 20000 4 24.390 218.6981 3.564028e-010 3.154945e-002 pass
25000 25000 4 48.645 214.1629 7.503647e-009 4.267055e-001 pass
25000 25000 4 48.662 214.0885 6.397731e-010 3.638160e-002 pass

Performance Summary (GFlops)

Size LDA Align. Average Maximal
1000 1000 4 98.6430 101.0553
2000 2000 4 147.0953 152.2219
3000 3000 4 162.0406 166.9275
4000 4000 4 178.4188 180.3108
5000 5000 4 185.1650 185.6883
10000 10000 4 201.0252 203.4404
15000 15000 4 185.4334 185.6833
20000 20000 4 218.6666 218.6981
25000 25000 4 214.1257 214.1629

Residual checks PASSED

End of tests

Mon 06/24/2013
07:19 PM

 

[24601]

Your flops are crazy low!

Intel(R) Optimized LINPACK Benchmark data

Current date/time: Mon Jun 24 19:14:37 2013

CPU frequency: 4.597 GHz
Number of CPUs: 1
Number of cores: 4
Number of threads: 4

Parameters are set to:

Number of tests: 9
Number of equations to solve (problem size) : 1000 2000 3000 4000 5000 10000 15000 20000 25000
Leading dimension of array : 1000 2000 3000 4000 5000 10000 15000 20000 25000
Number of trials to run : 4 4 4 4 4 2 2 2 2
Data alignment value (in Kbytes) : 4 4 4 4 4 4 4 4 4

Maximum memory requested that can be used=705536800, at the size=25000

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

Size LDA Align. Time(s) GFlops Residual Residual(norm) Check
1000 1000 4 0.007 100.0239 1.002198e-012 3.417754e-002 pass
1000 1000 4 0.007 98.9791 1.002198e-012 3.417754e-002 pass
1000 1000 4 0.007 94.5137 1.002198e-012 3.417754e-002 pass
1000 1000 4 0.007 101.0553 1.002198e-012 3.417754e-002 pass
2000 2000 4 0.035 152.2219 4.327261e-012 3.764187e-002 pass
2000 2000 4 0.036 146.6798 4.327261e-012 3.764187e-002 pass
2000 2000 4 0.036 146.4302 4.327261e-012 3.764187e-002 pass
2000 2000 4 0.037 143.0490 4.327261e-012 3.764187e-002 pass
3000 3000 4 0.114 157.6002 9.653528e-012 3.717342e-002 pass
3000 3000 4 0.112 160.2734 9.653528e-012 3.717342e-002 pass
3000 3000 4 0.108 166.9275 9.653528e-012 3.717342e-002 pass
3000 3000 4 0.110 163.3613 9.653528e-012 3.717342e-002 pass
4000 4000 4 0.238 179.5631 1.565487e-011 3.412126e-002 pass
4000 4000 4 0.246 173.8406 1.565487e-011 3.412126e-002 pass
4000 4000 4 0.237 179.9607 1.565487e-011 3.412126e-002 pass
4000 4000 4 0.237 180.3108 1.565487e-011 3.412126e-002 pass
5000 5000 4 0.450 185.1852 2.512907e-011 3.504046e-002 pass
5000 5000 4 0.451 184.7307 2.512907e-011 3.504046e-002 pass
5000 5000 4 0.449 185.6883 2.512907e-011 3.504046e-002 pass
5000 5000 4 0.451 185.0557 2.512907e-011 3.504046e-002 pass
10000 10000 4 3.358 198.6099 1.012627e-010 3.570624e-002 pass
10000 10000 4 3.278 203.4404 1.012627e-010 3.570624e-002 pass
15000 15000 4 12.120 185.6833 1.979849e-010 3.118294e-002 pass
15000 15000 4 12.153 185.1834 1.979849e-010 3.118294e-002 pass
20000 20000 4 24.397 218.6350 3.564028e-010 3.154945e-002 pass
20000 20000 4 24.390 218.6981 3.564028e-010 3.154945e-002 pass
25000 25000 4 48.645 214.1629 7.503647e-009 4.267055e-001 pass
25000 25000 4 48.662 214.0885 6.397731e-010 3.638160e-002 pass

Performance Summary (GFlops)

Size LDA Align. Average Maximal
1000 1000 4 98.6430 101.0553
2000 2000 4 147.0953 152.2219
3000 3000 4 162.0406 166.9275
4000 4000 4 178.4188 180.3108
5000 5000 4 185.1650 185.6883
10000 10000 4 201.0252 203.4404
15000 15000 4 185.4334 185.6833
20000 20000 4 218.6666 218.6981
25000 25000 4 214.1257 214.1629

Residual checks PASSED

End of tests

Mon 06/24/2013
07:19 PM

Thats what the 1 clock FMA and double bandwidth is for

 

[BallaTheFeared]

Doh, you're running Ivy!

 

[Idontcare]

Your flops are crazy low!

Indeed. They look to be right near 50% of what they should be for that clockspeed which makes me think something isn't right in regards to AVX2 compatibility (be it OS, the linx/IBT binaries, or something else).

edit: derp, didn't refresh the screen before posting, yeah IB vs HW does make a difference

 

[ehume]

Thats what the 1 clock FMA and double bandwidth is for

What is that?

Also, How do you get Linpack 11+ to run on Haswell? I'm running 10.3.10.017, which is the latest I can run without crashing.

But when I ran LinX/Linpack 11+, I was getting IIRC 166 Gflops, but the system would crash on second rep. But that was when the cpu was at 4.7 GHz and it was throttling from the heat on the first rep.

This early Haswell will go up to 5.0 GHz in a trophy run, but it needs 1.25v to OC to 4.5 GHz -- no low-Voltage overclocking for me. And I'm too chicken to delid: if I break the chip and need a new one the wife will kill me.

Any thoughts on Linpack 11?

 

[BallaTheFeared]

Try lowering your cpuNB, also if you're freezing towards the top you might be running out of ram. You can set the number of tests to coincide with how much ram you end up using.

 

[24601]

What is that?

Also, How do you get Linpack 11+ to run on Haswell? I'm running 10.3.10.017, which is the latest I can run without crashing.

But when I ran LinX/Linpack 11+, I was getting IIRC 166 Gflops, but the system would crash on second rep. But that was when the cpu was at 4.7 GHz and it was throttling from the heat on the first rep.

This early Haswell will go up to 5.0 GHz in a trophy run, but it needs 1.25v to OC to 4.5 GHz -- no low-Voltage overclocking for me. And I'm too chicken to delid: if I break the chip and need a new one the wife will kill me.

Any thoughts on Linpack 11?

1 clock FMA is:
Single Clock cycle fused multiply/add operation.
This is the main big boost from Haswell.
Since this can potentially double your throughput, Haswell also doubled L1 and L2 bandwidth to deal with doubled throughput.

If you are crashing in latest Intel Math Kernel Linpack, that just plain means your settings are not stable for your chip.

Offtopic: This is the main reason why Bulldozer's (Faildozer) cache is so high latency. It had to be high enough bandwidth to accommodate Bulldozer's FMA.

Offtopic2: This is also the main reason Tahiti scores so high on some of the GPGPU loads compared to Kepler.

Offtopic3: This is also the main reason AMD's GPUs are so good at SHA256 and Scrypt mining compared to Nvidia's cards.

 

[ehume]

OK. I'm out of my depth here. I reset my cpu to "optimized defaults." LinX 0.6.4 (with the .exe files from the most recent Linpack substituted in) crashes after 1 rep. With that one rep, done twice, the results were 156 and 159 Gflops. Clearly, LinX needs updating to work with Linpack 11.

If I run Linpack 11+ by itself, it starts, and uses about 51% of the CPU resources.

Can I just run the sample script? What are the grammar and the parameters if I should be modifying the script?

How long should I expect Linpack to run? At 4.5 GHz LinX with AVX (Linpack 10) takes 59 minutes to run. How long will this script (AVX2, Linpack 11) run?

Reading the script, I should expect it to say "Done" when it is done. Does it do that?

 

[BallaTheFeared]

F:\Users\Balla\Downloads\w_lpk_p_11.0.3.008\linpack_11.0.3\benchmarks\linpack

Double click: runme_xeon64


Edit: Runtime is about 3 minutes for me, 9 tests.

 

[ehume]

F:\Users\Balla\Downloads\w_lpk_p_11.0.3.008\linpack_11.0.3\benchmarks\linpack

Double click: runme_xeon64

That's what I do. The file's full name is "runme_xeon64.bat" after all.

I take it that I don't need to amend the batch file. But about how long will it run?

 

[BallaTheFeared]

How much system ram do you have?

If you have 8GB or so edit the file lininput_xeon64 with notepad and place this in there, it will stop you at 25000 problem size.

Sample Intel(R) Optimized LINPACK Benchmark data file (lininput_xeon64)
Intel(R) Optimized LINPACK Benchmark data
9                        # number of tests
1000 2000 3000 4000 5000 10000 15000 20000 25000 30000 35000 40000 # problem sizes
1000 2000 3000 4000 5000 10000 15000 20000 25000 30000 35000 40000 # leading dimensions
4 4 4 4 4 2 2 2 2 1 1 1   # times to run a test
4 4 4 4 4 4 4 4 4 4 4 4   # alignment values (in KBytes)

 

[ehume]

I indeed have 8GB RAM. How long will it take?

And shouldn't Linpack use nearly 100% of CPU resources?

 

[BallaTheFeared]

Shouldn't take more than 3 1/3 minutes.

Yes, it should use 100% at all times no drop.

What OS are you using?

 

[ehume]

Thanks for your help.

I'm using Win7-64.

I'm noticing Linpack using bits of 100%, then longer bits, but with big drops in between. I suspect that if I went into the input file you pointed out, I could delete the smaller tests, then set the reps for 20 and I'd replicate LinX.

Or, I could put 9 in the first line, and for the fourth line, where it says "times to run a test", I could put:

0 0 0 0 0 0 0 0 20 0 0 0

to get 20 reps of 25000.

Yes? No?

 

[BallaTheFeared]

I haven't tried that, maybe it will.

Also you did upgrade to SP1 right, and your OS is fully updated beyond that correct?

Are you getting the 120~ GFLOPs in IBT/LinX?

 

[ehume]

Ran the test. Got 98 Gflops :-(
Also, the result text said the cpu was running 2.98GHz, which should not be correct. I saw it on Real Temp represented as 3700 MHz.

With Linx, got 156-159.

Edit: is there an IBT with Linpack 11?

OS updated SP1.

 

[BallaTheFeared]

Try mine: https://mega.co.nz/#!6YZhwSoZ!H4FB5hglPXd0gRdNyCxcgWP_9q5LAk5Lapa6_7TfJ2M

 

[ehume]

Intel(R) Optimized LINPACK Benchmark data

Current date/time: Fri Jul 05 22:56:51 2013

CPU frequency:    3.889 GHz
Number of CPUs: 1
Number of cores: 4
Number of threads: 8

Parameters are set to:

Number of tests: 1
Number of equations to solve (problem size) : 25000
Leading dimension of array                  : 25000
Number of trials to run                     : 1    
Data alignment value (in Kbytes)            : 4    

Maximum memory requested that can be used=705536800, at the size=25000

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

Size   LDA    Align. Time(s)    GFlops   Residual     Residual(norm) Check
25000  25000  4      105.553    98.6985  5.194889e-010 2.954147e-002   pass

Performance Summary (GFlops)

Size   LDA    Align.  Average  Maximal
25000  25000  4       98.6985  98.6985 

Residual checks PASSED

End of tests

Fri 07/05/2013 
10:59 PM

I discovered I got errors with 0 reps. The minimum is 1 rep.

So I deleted all but the 25000 test, both rows (lines 2 & 3). In line 1 I put "1" to represent the sole remaining test. In line 4 I put "1" for the number of reps I wanted to do. If I put "20" there, I should replicate LinX.

OTOH, I did not replicate the Gflops.

Thanks for the Linpack, BTW. How does one put that together with the IBT front end? Or is there an up-to-date IBT for Haswell?

Edit:

Extended help in xhelp.lpk. Good stuff.

The CPU speed is too fast for 4 cores/8 threads. And LinX is again giving me 159 Gflops . . and one completed rep.

 

[24601]

Intel(R) Optimized LINPACK Benchmark data

Current date/time: Fri Jul 05 22:56:51 2013

CPU frequency:    3.889 GHz
Number of CPUs: 1
Number of cores: 4
Number of threads: 8

Parameters are set to:

Number of tests: 1
Number of equations to solve (problem size) : 25000
Leading dimension of array                  : 25000
Number of trials to run                     : 1    
Data alignment value (in Kbytes)            : 4    

Maximum memory requested that can be used=705536800, at the size=25000

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

Size   LDA    Align. Time(s)    GFlops   Residual     Residual(norm) Check
25000  25000  4      105.553    98.6985  5.194889e-010 2.954147e-002   pass

Performance Summary (GFlops)

Size   LDA    Align.  Average  Maximal
25000  25000  4       98.6985  98.6985 

Residual checks PASSED

End of tests

Fri 07/05/2013 
10:59 PM

I discovered I got errors with 0 reps. The minimum is 1 rep.

So I deleted all but the 25000 test, both rows (lines 2 & 3). In line 1 I put "1" to represent the sole remaining test. In line 4 I put "1" for the number of reps I wanted to do. If I put "20" there, I should replicate LinX.

OTOH, I did not replicate the Gflops.

Thanks for the Linpack, BTW. How does one put that together with the IBT front end? Or is there an up-to-date IBT for Haswell?

Edit:

Extended help in xhelp.lpk. Good stuff.

The CPU speed is too fast for 4 cores/8 threads. And LinX is again giving me 159 Gflops . . and one completed rep.

I don't think you are supposed to use the extra 4 logical threads in Linpack.

The code is highly optimized. Only extremely unoptimized code gets you extra performance with hyper-threading.

 

[ehume]

So I learned that to test this beast I must disable HT. Then, to keep it from getting too hot, I had to reduce Vcore. I ended up with a CPU speed of 4.3GHz on a Vcore of 1.19v. The hottest core peaks at 94c doing Linpack. Results:

Intel(R) Optimized LINPACK Benchmark data

Current date/time: Sun Jul 07 18:16:55 2013

CPU frequency:    4.298 GHz
Number of CPUs: 1
Number of cores: 4
Number of threads: 4

Parameters are set to:

Number of tests: 1
Number of equations to solve (problem size) : 25000
Leading dimension of array                  : 25000
Number of trials to run                     : 20   
Data alignment value (in Kbytes)            : 4    

Maximum memory requested that can be used=705536800, at the size=25000

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

Size   LDA    Align. Time(s)    GFlops   Residual     Residual(norm) Check
25000  25000  4      48.878     213.1411 5.194889e-010 2.954147e-002   pass
25000  25000  4      48.851     213.2601 5.194889e-010 2.954147e-002   pass
25000  25000  4      48.762     213.6494 5.194889e-010 2.954147e-002   pass
25000  25000  4      48.752     213.6900 5.194889e-010 2.954147e-002   pass
25000  25000  4      48.732     213.7779 5.194889e-010 2.954147e-002   pass
25000  25000  4      48.680     214.0070 5.194889e-010 2.954147e-002   pass
25000  25000  4      48.740     213.7456 5.194889e-010 2.954147e-002   pass
25000  25000  4      48.692     213.9560 5.194889e-010 2.954147e-002   pass
25000  25000  4      48.771     213.6081 5.194889e-010 2.954147e-002   pass
25000  25000  4      48.710     213.8784 5.194889e-010 2.954147e-002   pass
25000  25000  4      48.688     213.9729 5.194889e-010 2.954147e-002   pass
25000  25000  4      48.725     213.8086 5.194889e-010 2.954147e-002   pass
25000  25000  4      48.681     214.0058 5.194889e-010 2.954147e-002   pass
25000  25000  4      48.715     213.8526 5.194889e-010 2.954147e-002   pass
25000  25000  4      48.687     213.9778 5.194889e-010 2.954147e-002   pass
25000  25000  4      48.707     213.8887 5.194889e-010 2.954147e-002   pass
25000  25000  4      48.705     213.8974 5.194889e-010 2.954147e-002   pass
25000  25000  4      48.735     213.7686 5.194889e-010 2.954147e-002   pass
25000  25000  4      48.713     213.8645 5.194889e-010 2.954147e-002   pass
25000  25000  4      48.724     213.8150 5.194889e-010 2.954147e-002   pass

Performance Summary (GFlops)

Size   LDA    Align.  Average  Maximal
25000  25000  4       213.7783 214.0070

Residual checks PASSED

End of tests

Sun 07/07/2013 
06:40 PM

I'm sure that further tweaking will get a slightly better OC, but I would have to delid this chip to get adequate cooling. Both of my heatsinks have barely gotten warm.