CPU benchmark - how many GFLOPs does your multicore machine do?

[calc]

System specs:

Intel Core i7 860 2.80 GHz (Lynnfield) oc to 169x21 3.549GHz
8GB DDR3 memory running at 1690 MHz cas 7

Input data or print help ? Type [data]/help :

Number of equations to solve (problem size): 30000
Leading dimension of array: 30000
Number of trials to run: 4
Data alignment value (in Kbytes): 4
Current date/time: Tue Oct 20 20:08:14 2009

CPU frequency:    3.549 GHz
Number of CPUs: 8
Number of threads: 8

Parameters are set to:

Number of tests                             : 1
Number of equations to solve (problem size) : 30000
Leading dimension of array                  : 30000
Number of trials to run                     : 4    
Data alignment value (in Kbytes)            : 4    

Maximum memory requested that can be used = 7200604096, at the size = 30000

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

Size   LDA    Align. Time(s)    GFlops   Residual     Residual(norm)
30000  30000  4      341.007    52.7902  8.446834e-10 3.329751e-02
30000  30000  4      340.978    52.7946  8.446834e-10 3.329751e-02
30000  30000  4      340.978    52.7945  8.446834e-10 3.329751e-02
30000  30000  4      340.987    52.7932  8.446834e-10 3.329751e-02

Performance Summary (GFlops)

Size   LDA    Align.  Average  Maximal
30000  30000  4       52.7931  52.7946

End of tests

Note that linpack gives much higher GFlops numbers under Linux than under Windows 7. The highest I got under Windows 7 was 46.74 which was ~ 13% slower than under Linux. Also I noticed that when stability testing under Windows 7 I found the voltage needed for it, which was stable for over 10 hours of linpack, but since it apparently wasn't stressing the CPU nearly as much the same voltage under Linux caused the system to immediately reboot when I ran linpack. I upped the Vtt core voltage another 0.020 and it became stable under Linux as well.

Also note that for Lynnfield the maximum safe Vtt voltage is 1.21 so that is why I limited my overclock to only 3.549 GHz as that was the highest I could go keeping the Vtt voltage within the guidelines in the chip documentation. The socket 1156 motherboards, at least my Gigabyte GA-P55-UD5, apparently still only show the limits based on Bloomfield which was 1.35 volts so be careful if you overclock a Lynnfield processor.

[Bachstelze]

Bow before my Celeron 220!

firas@iori linpack % ./xlinpack_xeon64
Input data or print help ? Type [data]/help :

Number of equations to solve (problem size): 13700
Leading dimension of array: 13700
Number of trials to run: 2
Data alignment value (in Kbytes): 2
Current date/time: Wed Oct 21 04:51:38 2009

CPU frequency:    1.200 GHz
Number of CPUs: 1
Number of threads: 1

Parameters are set to:

Number of tests                             : 1
Number of equations to solve (problem size) : 13700
Leading dimension of array                  : 13700
Number of trials to run                     : 2
Data alignment value (in Kbytes)            : 2

Maximum memory requested that can be used = 1501796048, at the size = 13700

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

Size   LDA    Align. Time(s)    GFlops   Residual     Residual(norm)
13700  13700  2      764.291    2.2434   1.921471e-10 3.624467e-02
13700  13700  2      883.600    1.9405   1.921471e-10 3.624467e-02

Performance Summary (GFlops)

Size   LDA    Align.  Average  Maximal
13700  13700  2       2.0919   2.2434

End of tests

[canduc17]

Great tool!
Do you know another one to obtain the bandwidth between CPU and main memory in Byte/seconds? I'm going crazy serching for it!
Thank you.

[wapko]

So i got a dualcore atom 330 board, the intel D945GCLF2D

this i my results. will be a nice little nas box when im done setting it up.

Number of equations to solve (problem size): 7000
Leading dimension of array: 7000
Number of trials to run: 2
Data alignment value (in Kbytes): 2
Current date/time: Fri Nov  6 09:02:00 2009

CPU frequency:    1.596 GHz
Number of CPUs: 4
Number of threads: 4

Parameters are set to:

Number of tests                             : 1
Number of equations to solve (problem size) : 7000
Leading dimension of array                  : 7000
Number of trials to run                     : 2
Data alignment value (in Kbytes)            : 2

Maximum memory requested that can be used = 392142048, at the size = 7000

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

Size   LDA    Align. Time(s)    GFlops   Residual     Residual(norm)
7000   7000   2      310.644    0.7364   5.086720e-11 3.650045e-02
7000   7000   2      310.633    0.7364   5.086720e-11 3.650045e-02

Performance Summary (GFlops)

Size   LDA    Align.  Average  Maximal
7000   7000   2       0.7364   0.7364

End of tests

disabling HT gives an avg. of 0.771

[drphilngood]

i7 Lynnfield 870 on 64-bit Jaunty [Average 58.6443, Maximal 58.6537
Linpack_58.6443_Score.png

[Uppsala9496]

Any way of getting Linpack to work on an AMD cpu? Version 10.2 from intel won't allow it to run on an AMD chip.

[w0mbatharcos]

Specs:
2 x Xeon X5365
16 GB FB-Dimm DDR2 RAM

w0mbatharcos@w0mbatharcos-desktop:~/Letöltések/linpack_10.2.3/benchmarks/linpack
$ ./xlinpack_xeon64
Input data or print help ? Type [data]/help :

Number of equations to solve (problem size): 40000
Leading dimension of array: 40000
Number of trials to run: 4
Data alignment value (in Kbytes): 4
Current date/time: Wed Feb 24 17:43:18 2010

CPU frequency:    2.992 GHz
Number of CPUs: 8
Number of threads: 8

Parameters are set to:

Number of tests                             : 1
Number of equations to solve (problem size) : 40000
Leading dimension of array                  : 40000
Number of trials to run                     : 4    
Data alignment value (in Kbytes)            : 4    

Maximum memory requested that can be used = 12800804096, at the size = 40000

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

Size   LDA    Align. Time(s)    GFlops   Residual     Residual(norm)
40000  40000  4      729.120    58.5224  1.435151e-09 3.191825e-02
40000  40000  4      721.039    59.1783  1.435151e-09 3.191825e-02
40000  40000  4      724.087    58.9292  1.435151e-09 3.191825e-02
40000  40000  4      726.529    58.7311  1.435151e-09 3.191825e-02

Performance Summary (GFlops)

Size   LDA    Align.  Average  Maximal
40000  40000  4       58.8402  59.1783 

End of tests

[calc]

i7 Lynnfield 870 on 64-bit Jaunty [Average 58.6443, Maximal 58.6537
Linpack_58.6443_Score.png

Sorry I didn't see this earlier, I don't know if you care or not but your system isn't stable. The Residual(norm) numbers should all match if your system is working properly. When overclocking my machine I let the test run for about 24 hours (few hundred times) and made sure all the numbers matched before doing anything critical on the system.

[drphilngood]

Sorry I didn't see this earlier, I don't know if you care or not but your system isn't stable. The Residual(norm) numbers should all match if your system is working properly. When overclocking my machine I let the test run for about 24 hours (few hundred times) and made sure all the numbers matched before doing anything critical on the system.

That's strange; my rig has ran eight instances of Prime95 for 19 hours, along with various other stability testing apps, without any errors. In addition, I have not had ANY problems with data corruption nor seen any other behavior that would indicate instability. I suppose that Linpack could be a more strenuous test but it didn't seem to be and my rig didn't get as hot running it as it did when running Prime95. *scratches head* However, as consequence of your notification, I will now do further testing and let you know if I get any errors.
Thanks for the 411, as well as your concern, my friend.
Have a good one!! (^_^)

addendum:
I have no explanation for this. Attached are two more tests that were ran at higher overclocks - both with and without hyper-threading - and the Residual(norm) numbers do, indeed, match; albeit, at somewhat lower scores. Thus, I suppose I should do further runs and see if I can duplicate the errors found in the aforementioned flawed run.
Thanks again. =)
Additional Linpack Results 3.884GHz.txt

[calc]

That's strange; my rig has ran eight instances of Prime95 for 19 hours, along with various other stability testing apps, without any errors. In addition, I have not had ANY problems with data corruption nor seen any other behavior that would indicate instability. I suppose that Linpack could be a more strenuous test but it didn't seem to be and my rig didn't get as hot running it as it did when running Prime95. *scratches head* However, as consequence of your notification, I will now do further testing and let you know if I get any errors.
Thanks for the 411, as well as your concern, my friend.
Have a good one!! (^_^)

addendum:
I have no explanation for this. Attached are two more tests that were ran at higher overclocks - both with and without hyper-threading - and the Residual(norm) numbers do, indeed, match; albeit, at somewhat lower scores. Thus, I suppose I should do further runs and see if I can duplicate the errors found in the aforementioned flawed run.
Thanks again. =)
Additional Linpack Results 3.884GHz.txt

The highest GFlops you can get it to do combined with a stable set of residual numbers should determine a stable system. If you aren't reaching your highest GFlops number then you aren't stressing the CPU as much and so you are more likely to get a stable residual number. As I mentioned back when I first posted my numbers when I was running under Windows it tested stable (over many hours of linpack) but the GFlops were lower than under Linux and so when I ran Linpack on Linux with the same bios settings the GFlops were higher and caused a near immediate reboot of my system. I think this was due to other processes stealing time on the CPU and keeping the system from reaching peak load. So when I did my testing under Linux I booted into single user mode so that I could be sure nothing else (or very little) would be stealing CPU time and keeping the system from reaching peak load. Linpack does stress the CPU a huge amount more than Prime95 does, I don't remember the exact numbers at the moment but I had a Kill A Watt meter hooked up to my pc when I was doing burn in testing and from what I recall my system with the dinky Radeon 4350 and a 80Plus power supply was pulling over 230 Watts when under load. I used Prime95 to do initial testing to find a close range of what was stable and used Linpack to finish dialing it in. Also for long term stability you may want to keep your Vtt below 1.21 as I mentioned in my previous post. I was able to overclock higher than 3.55GHz with my CPU but it was pushing Vtt over the maximum safe limits from the Intel spec sheet. The Bloomfield LGA 1366 i7 CPUs didn't have the lower Vtt limit in their spec sheet so probably can run safe at higher Vtt voltage and thus higher overclocks. I tend to err on the safe side since I use my system for work and want it to last at least 6 years, 3 for myself and then 3 for general home use.