WRF EMS Benchmark results and hardware

Looking for new hardware to run WRF? Intel or AMD? Check this forum.
meteoadriatic
Posts: 1603
Joined: Wed Aug 19, 2009 10:05 am

Re: WRF EMS Benchmark results and hardware

Post by meteoadriatic » Sat Jul 07, 2012 8:16 am

pattim wrote:Wow - 20% slower with the slower memory. Did you try changing the Northbridge speed (or FSB or whatever they call that)?
No. I require stability and computer long life so I won't run components over specifications. Of course I can try for benchmark purporses but it doesn't have any practical sense if I won't push it in production time. BTW; 10%, not 20%.
pattim wrote:I wonder how Xeon's compare to i7's?
For same generation, number of cores, clock frequency and cache amount, they have the about same results. However they require different MBOs so... not easy to compare. Simple answer: nothing spectacular. Desktop CPUs have much better performance/price ratio.
pattim wrote:I wonder if it would be worthwhile to try this benchmark with and without hyperthreading? Hyperthreading isn't good for WRF (or CFD in general), apparently.
http://www.cpubenchmark.net/high_end_cpus.html
Did it now: NMM with HT, and memory clock 2133: 5:24, vs 5:07 when HT is off.
pattim wrote:What domain decomposition did you use? I found just today that the 1 x N decomposition can speed things up, at least on a nested domain.
Benchmark cases should not be changed/reconfigured in any way. If you do that, you lose their goal - you cannot directly compare different hardware anymore if different systems use differently configured benchmark domains! That's why I didn't changed anything in WRFEMS for benchmarking.

pattim
Posts: 199
Joined: Sun Jun 24, 2012 8:42 pm
Location: Los Angeles, CA, USA

Re: WRF EMS Benchmark results and hardware

Post by pattim » Wed Jul 18, 2012 11:31 pm

Running on 4 cores in a Linux guest virtual machine (WinXP x64 host); Phenom II 1090T
ARW benchmark

--domain 1:

Code: Select all

                System Date           : Wed Jul 18 13:58:32 2012 UTC
                System Hostname       : opensuse-vm-c
                System Address        : 8.15.7.117
            
                System OS             : Linux
                Linux Distribution    : openSUSE 12.1 (x86_64)
            VERSION = 12.1
            CODENAME = Asparagus
                OS Kernel             : 3.1.10-1.16-desktop
                Kernel Type           : x86_64
            
            Processor and Memory Information for opensuse-vm-c
            
                CPU Name              : AMD Phenom(tm) II X6 1090T Processor
                CPU Instructions      : barcelona
                CPU Type              : 64-bit
                CPU Speed             : 4026.84 MHz
            
                EMS Determined Processor Count
                    Physical CPUs     : 1
                    Cores per CPU     : 4
                    Total Processors  : 4
            
                Hyper-Threading       : Off
                  
                System Memory         : 5.7 Gbytes
            
            WRF EMS Release Information for opensuse-vm-c
            
                EMS Release           : 3.2.1.5.45.beta
                EMS Binaries          : x64


         WRF EMS ARW core benchmark simulation completed in 13 minutes 41 seconds
--domain 2

Code: Select all

                System Date           : Wed Jul 18 14:51:07 2012 UTC
                System Hostname       : opensuse-vm-c
                System Address        : 63.251.179.13
            
                System OS             : Linux
                Linux Distribution    : openSUSE 12.1 (x86_64)
            VERSION = 12.1
            CODENAME = Asparagus
                OS Kernel             : 3.1.10-1.16-desktop
                Kernel Type           : x86_64
            
            Processor and Memory Information for opensuse-vm-c
            
                CPU Name              : AMD Phenom(tm) II X6 1090T Processor
                CPU Instructions      : barcelona
                CPU Type              : 64-bit
                CPU Speed             : 4026.84 MHz
            
                EMS Determined Processor Count
                    Physical CPUs     : 1
                    Cores per CPU     : 4
                    Total Processors  : 4
            
                Hyper-Threading       : Off
                  
                System Memory         : 5.7 Gbytes
            
            WRF EMS Release Information for opensuse-vm-c
            
                EMS Release           : 3.2.1.5.45.beta
                EMS Binaries          : x64


         WRF EMS ARW core benchmark nested simulation completed in 51 minutes 9 seconds
More stats: (nmm this time)

2 cores i7-2640M; 2.7Ghz 8GB DDR3-1333 - - - - - - - - 9 min 56 sec

2 cores Phenom 1090T; 4GHz 16BG DDR3-1666 - - - - 12 min 58 sec
4 cores Phenom 1090T; 4GHz 16BG DDR3-1666 - - - - 8 min 41 sec
(same) but in a virtualbox guest - - - - - - - - - - - - - 11 min 52 sec

So it seems clear that, from what I read, Intel QuickPath has much better effective bandwidth than AMD HyperTransport, due to better utilization algorithms that keep the bus clear of unnecessary "snoops." It has also been reported that the AMD be made faster by careful utilization of CPU nodelocking (and other technologies I don't understand).
Last edited by pattim on Sat Jul 21, 2012 6:30 pm, edited 6 times in total.

pattim
Posts: 199
Joined: Sun Jun 24, 2012 8:42 pm
Location: Los Angeles, CA, USA

Re: WRF EMS Benchmark results and hardware

Post by pattim » Fri Jul 20, 2012 7:30 pm

meteoadriatic wrote:
pattim wrote:What domain decomposition did you use? I found just today that the 1 x N decomposition can speed things up, at least on a nested domain.
Benchmark cases should not be changed/reconfigured in any way. If you do that, you lose their goal - you cannot directly compare different hardware anymore if different systems use differently configured benchmark domains! That's why I didn't changed anything in WRFEMS for benchmarking.
:oops:

ceterus
Posts: 14
Joined: Sat Mar 06, 2010 12:28 am

Re: WRF EMS Benchmark results and hardware

Post by ceterus » Sat Jul 28, 2012 5:44 pm

Code: Select all

*  Hey, hey! Your WRF model simulation appears to have been successful.

         Mining system information from the benchmark simulation

         *  Gathering information for localhost blackdeath

            Basic System Information for blackdeath
            
                System Date           : Sun Jul 29 17:35:47 2012 UTC
                System Hostname       : blackdeath
                System Address        : 127.0.0.1
            
                System OS             : Linux
                Linux Distribution    : Fedora release 17 (Beefy Miracle)
                OS Kernel             : 3.3.4-5.fc17.x86_64
                Kernel Type           : x86_64
            
            Processor and Memory Information for blackdeath
            
                CPU Name              : AMD FX(tm)-8150 Eight-Core Processor 
                CPU Instructions      : shanghai
                CPU Type              : 64-bit
                CPU Speed             : 3300 MHz
            
                EMS Determined Processor Count
                    Physical CPUs     : 0.5
                    Cores per CPU     : 8
                    Total Processors  : 4
            
                Hyper-Threading       : On
                  
            Note: Attempting to use virtual "Hyper-threaded" CPUs while
            running the WRF EMS may result in a degradation in performance.
            
                System Memory         : 15.6 Gbytes
            
            WRF EMS Release Information for blackdeath
            
                EMS Release           : 3.2.1.5.45.beta
                EMS Binaries          : x64


         WRF EMS NMM core benchmark simulation completed in 9 minutes 15 seconds
Not as fast as I imagined it would be, anyone know what it means when it says:
" EMS Determined Processor Count
Physical CPUs : 0.5
Cores per CPU : 8
Total Processors : 4"

-Tim

meteoadriatic
Posts: 1603
Joined: Wed Aug 19, 2009 10:05 am

Re: WRF EMS Benchmark results and hardware

Post by meteoadriatic » Sat Jul 28, 2012 7:58 pm

Hi, this means that install script has failed correctly guessing your CPU configuration.

You need to fix that in EMS.cshrc file, lines
setenv NCPUS
setenv CORES

must be correct. Also, check that you have correct number of threads in runs/domainname/conf/ems_run/run_ncpus.conf

ceterus
Posts: 14
Joined: Sat Mar 06, 2010 12:28 am

Re: WRF EMS Benchmark results and hardware

Post by ceterus » Sun Jul 29, 2012 4:22 pm

Thanks for the reply meteoadriatic...

In both the files you mentioned the number of CPUs (8) is correct.

Perhaps it is a fluke?

-Tim

meteoadriatic
Posts: 1603
Joined: Wed Aug 19, 2009 10:05 am

Re: WRF EMS Benchmark results and hardware

Post by meteoadriatic » Mon Jul 30, 2012 8:34 pm

Try to turn off hyperthreading in BIOS. It should improve performance slightly and it might fix this "error".

pattim
Posts: 199
Joined: Sun Jun 24, 2012 8:42 pm
Location: Los Angeles, CA, USA

Re: WRF EMS Benchmark results and hardware

Post by pattim » Sat Sep 22, 2012 1:26 am

I reinstalled 3.2 beta and am running on AMD Phenom II 1090T black overclocked to 3.9 GHz. I ran with the memory controllers ganged and unganged, and with the memory clocked down from 1600 -> 1333, and also with the CPU clocked down to near-native 3.4 GHz - just to see what really matters with these CPUs...

ARW solver, fresh install of EMS 3.2 beta: (G)= memory controllers ganged, (UG)= memory controllers unganged; (UG+NB) = same as UG, but also 20% increase in NorthBridge clock speed (2000 -> 2400)
MOBO: GigabyteGA-880GM

3.9 GHz + 1600 DDR3
................sec..............sec..............sec
Dom 1.........807 (G)........592 (UG)........564 (UG+NB)
Dom 2.......2822 (G).......2206 (UG).......2140 (UG+NB)

3.9 GHz + 1333 DDR3
.....................sec.
Dom 1............ 628 (UG)
Dom 2............ 2336 (UG)

3.4 GHz + 1600 DDR3
.....................sec.
Dom 1............ 649 (UG)
Dom 2............ 2390 (UG)

So the biggest speed up (25%) is to ungang the memory controllers on the Phenom II. Memory, CPU, and NorthBridge clocks all help to a lesser extent, and the biggest of those is probably the NorthBridge.

I guess clocking things this fast can shorten lifetimes of components unless you are VERY careful to have extra cooling for everything, and actually measure temperatures. Under linux (I am using Opensuse 12.1) you can install the sensors package, run the sensors-detect script (as superuser), and then use GKrellm to view CPU temperatures.
http://forums.opensuse.org/archives/sf- ... speed.html
My CPU is about 58C while I'm running the above benchmark a maximum stress level, below the max continuous of 61C. I also have a fan cooling both my memory and chipset heat sinks.

vonjack
Posts: 16
Joined: Thu Sep 27, 2012 7:14 pm

Re: WRF EMS Benchmark results and hardware

Post by vonjack » Sun Feb 03, 2013 12:55 pm

My configuration:
CPU i5 3450 3.1 GHz
8 GB RAM DDR3 1600Mhz

Code: Select all

           Basic System Information for localhost
           
               System Date           : Sun Feb  3 09:47:08 2013 UTC
               System Hostname       : localhost
               System Address        : 127.0.0.1
           
               System OS             : Linux
               Linux Distribution    : CentOS release 6.3 (Final)
               OS Kernel             : 2.6.32-279.19.1.el6.x86_64
               Kernel Type           : x86_64
           
           Processor and Memory Information for localhost
           
               CPU Name              : Intel(R) Core(TM) i5-3450 CPU @ 3.10GHz
               CPU Instructions      : nehalem
               CPU Type              : 64-bit
               CPU Speed             : 1600 MHz
           
               EMS Determined Processor Count
                   Physical CPUs     : 1
                   Cores per CPU     : 4
                   Total Processors  : 4
           
               Hyper-Threading       : Off
                 
               System Memory         : 5.6 Gbytes
           
           WRF EMS Release Information for localhost
           
               EMS Release           : 3.2.1.5.45.beta
               EMS Binaries          : x64


           Benchmark simulation length was 24 hours

           Summary of nodes and processors used for benchmark simulation:

             *  4  Processors on localhost    
             ------------------------------
             *  4  Total Processors
             *  4  Tiles per Processor

             *  1 x 4 Domain Decomposition

           WRF EMS ARW core benchmark simulation completed in 7 minutes 52 seconds


capitanlibeccio
Posts: 18
Joined: Sat Feb 13, 2010 9:36 am

Re: WRF EMS Benchmark results and hardware

Post by capitanlibeccio » Mon Feb 25, 2013 4:41 pm

Dual x5690-Evga sr2. WRF 3.1.1.5.1

System Information for new-host

System Date : Mon Feb 25 16:38:06 2013 UTC
Machine Address : 127.0.0.1
Alternate Hostname : None
Alternate Address : None
Machine OS : Linux
Kernel : 2.6.32-279.22.1.el6.x86_64
Linux Distribution : CentOS release 6.3 (Final)
CPU Name : Intel(R) Xeon(R) CPU X5690 @ 3.47GHz
CPU Type : x86_64
CPU Instructions : nehalem
CPU Speed (MHz) : 1600.000
System Memory : 24022 Mb
Hyper-Threading : On

EMS.cshrc Defined Processor Count:
Physical CPUs : 2
Cores per CPU : 6
Total Processors : 12

EMS Determined Processor Count:
Physical CPUs : 2
Cores per CPU : 6
Total Processors : 12

Binaries : STRC Compiled x64



WRF EMS NMM core benchmark simulation completed in 3 minutes 1 second

Post Reply