Thread replies: 6
Thread images: 6
Thread images: 6
File: 73812main_Altix_front.jpg (303KB, 2964x1656px) Image search:
[Google]
303KB, 2964x1656px
Anyone in the mood for some dick waving?
STREAM measures sustained memory bandwidth, which is a pretty big deal in HPC/Supercomputing especially where problems often rely on huge amounts of data that don't fit in cache (and that expensive hardware needs to be well utilized)
I'm fucking around with compiling a big list of results in addition to those already submitted to the website, particularly interested in the Raspi 3 and Ryzen chips if anyone's bored enough to enable my autism.
Source is here: http://www.cs.virginia.edu/stream/FTP/Code/
Some building tips for the C version in the main .c file:* For a simple single-core version, try compiling with:
* cc -O stream.c -o stream
* This is known to work on many, many systems....
*
* To use multiple cores, you need to tell the compiler to obey the OpenMP
* directives in the code. This varies by compiler, but a common example is
* gcc -O -fopenmp stream.c -o stream_omp
* The environment variable OMP_NUM_THREADS allows runtime control of the
* number of threads/cores used when the resulting "stream_omp" program
* is executed.
*
* To run with single-precision variables and arithmetic, simply add
* -DSTREAM_TYPE=float
* to the compile line.
* Note that this changes the minimum array sizes required --- see (1) above.
There's prebuilt binaries too, but they're probably old and shitty and use arrays that are too small. Those with more than 20MB of cache will probably need to increase the array size as well.
Please blog your results with your CPU and motherboard, and anything else you want to blog with it.
>>
Some old supercomputer results on the site with their dates submitted:
16-processor Cray C90: 103.8 Gb/s (1991)
512-processor Cray T3E-1200: 284.9 Gb/s (1995)
8-processor Cray Y-MP: 26.8 Gb/s (1991)
32-processor Cray SV1ex: 47.8 Gb/s (2002)
256-processor SGI Origin 3800/500 MHz: 99.6 Gb/s (2001)
32-processor SGI Origin 300/500 MHz: 11.8 Gb/s (2002)
512-processor SGI Altix 3000: 1 Tb/s (2003)
2048-processor SGI Altix UV 2000: 7.1 Tb/s (2012)
Some badass workstations and servers with five-figure price tags:
SGI Power Indigo2: 82.1 Mb/s (1994)
SGI Indigo2/195 MHz: 109.1 Mb/s (1996)
SGI Octane 2x300 MHz: 493.6 Mb/s (1999)
Sun Ultra Enterprise 6000 (16CPU): 2.4 Gb/s (1996)
Sun Ultra 60 2x360 MHz: 374.9 Mb/s (1998)
Sun Fire V210: 905 Mb/s (2004)
Some regular stuff:
Raspberry Pi 2B: 498 Mb/s (2015)
Raspberry Pi 1: 248.8 Mb/s (2012)
i7 2600: 13.8 Gb/s (2011)
Q6600: 5.3 Gb/s (2007)
Dual Athlon 64 FX-72: 15.4 Gb/s (2007)
Athlon 64 X2 5200+: 4.8 Gb/s (2006)
Dual Opteron 244: 4.5 Gb/s (2003)
Pentium 4 2.8: 2.7 Gb/s (2003)
Athlon 1200: 1 Gb/s (2001)
Pentium 4 1400: 1.5 Gb/s (2000)
Pentium III 933 (i840): 634 Mb/s (2000)
Athlon 600: 417 Mb/s (2000)
Pentium 90: 69.9 Mb/s (1994)
>>
File: image010.gif (51KB, 402x415px) Image search:
[Google]
51KB, 402x415px
Ryzen 1700X @ 3.9GHz allcores
16GB Corsair LPX-3000 running @ DDR4-2166 (ASUS, fix your BIOS)
CFLAGS:-O3 -march=nativeCopy: 12800.3 0.050500 0.049999 0.052003
Scale: 23276.0 0.028001 0.027496 0.028500
Add: 24306.1 0.039777 0.039496 0.041000
Triad: 24000.6 0.040334 0.039999 0.040501
>>
File: 1280px-Blue_Mountain_Supercomputer.jpg (200KB, 1280x663px) Image search:
[Google]
200KB, 1280x663px
>>59395407
Scores a little better than a loaded 9-rack Origin 2K cluster that probably cost a couple million or more not including software in 1998.
Good shit.
>>
File: Screenshot_2017-03-14_03-57-58.png (167KB, 1172x516px) Image search:
[Google]
167KB, 1172x516px
>>59394593
Here you go champ-------------------------------------------------------------
STREAM version $Revision: 5.10 $
-------------------------------------------------------------
This system uses 8 bytes per array element.
-------------------------------------------------------------
Array size = 10000000 (elements), Offset = 0 (elements)
Memory per array = 76.3 MiB (= 0.1 GiB).
Total memory required = 228.9 MiB (= 0.2 GiB).
Each kernel will be executed 10 times.
The *best* time for each kernel (excluding the first iteration)
will be used to compute the reported bandwidth.
-------------------------------------------------------------
Number of Threads requested = 4
Number of Threads counted = 4
-------------------------------------------------------------
Your clock granularity/precision appears to be 1 microseconds.
Each test below will take on the order of 7286 microseconds.
(= 7286 clock ticks)
Increase the size of the arrays if this shows that
you are not getting at least 20 clock ticks per test.
-------------------------------------------------------------
WARNING -- The above is only a rough guideline.
For best results, please be sure you know the
precision of your system timer.
-------------------------------------------------------------
Function Best Rate MB/s Avg time Min time Max time
Copy: 15071.8 0.010652 0.010616 0.010770
Scale: 15002.4 0.010679 0.010665 0.010695
Add: 16913.4 0.014211 0.014190 0.014245
Triad: 16928.7 0.014281 0.014177 0.014972
-------------------------------------------------------------
Solution Validates: avg error less than 1.000000e-13 on all three arrays
-------------------------------------------------------------
>>
File: r96r1384.jpg (355KB, 2207x2496px) Image search:
[Google]
355KB, 2207x2496px
>>59395572
Took a good $2.5 million and some big air conditioners for those kinds of results 20 years ago. Nice.
Thread posts: 6
Thread images: 6
Thread images: 6