WRF EMS Benchmark results and hardware

Looking for new hardware to run WRF? Intel or AMD? Check this forum.
theocarter2911
Posts: 61
Joined: Fri Mar 08, 2013 5:22 am

Re: WRF EMS Benchmark results and hardware

Post by theocarter2911 » Wed Oct 02, 2013 12:25 pm

Hi Meteoadriatic,
You were spot on right, and will be surprized by the results of the new benchmarks with the four channel ram!

So, comparing apples with apples, the large nested NMM bench done with two channel ram and otherwise stock settings on the benchmark (no decomp changes or anything else), the system came in at 34 minutes. After I added in extra ram to fill in the channels and force the machine into four channel mode - this is the result:
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
Basic System Information for localhost
System Date : Tue Oct 1 12:25:48 2013 UTC
System Hostname : localhost
System Address : 127.0.0.1
System OS : Linux
Linux Distribution : Scientific Linux release 6.4 (Carbon)
Kernel \r on an \m
OS Kernel : 2.6.32-358.18.1.el6.x86_64
Kernel Type : x86_64
Processor and Memory Information for localhost
CPU Name : Intel(R) Xeon(R) CPU E5-2690 0 @ 2.90GHz
CPU Instructions : sandybridge
CPU Type : 64-bit
CPU Speed : 2900.16 MHz
EMS Determined Processor Count
Physical CPUs : 2
Cores per CPU : 8
Total Processors : 16
Hyper-Threading : Off
System Memory : 62.8 Gbytes
EMS Release Information for localhost
EMS Release : 3.4.1.13.37
EMS Binaries : x64
Benchmark simulation length was 24 hours
Summary of nodes and processors used for benchmark simulation:
* 15 Processors on localhost
------------------------------
* 15 Total Processors
* 1 Tile per Processor
* 1 x 15 Domain Decomposition
EMS NMM core benchmark nested simulation completed in 16 minutes 15 seconds
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
So basically going from two channel to four channel (admittedly an extra 32Gig ram at four sticks of 8Gig each to have the same as what was in before and access four channel) the bench ran in 16 minutes instead of 34 minutes!
More than 50% improvement in run time. :-)
And now my model output comes in in time for the latest issues of landing forecasts, which is perfect.

Thank you for your help and suggestions, it paid off big time!
Theo

meteoadriatic
Posts: 1510
Joined: Wed Aug 19, 2009 10:05 am

Re: WRF EMS Benchmark results and hardware

Post by meteoadriatic » Wed Oct 02, 2013 2:11 pm

Fascinating!

From your findings it is perfectly clear how much WRF is restrained by memory bandwidth, not by CPU capability and this is number one feature that must be maximized in WRF computer.

I'm glad we solved the problem :)

theocarter2911
Posts: 61
Joined: Fri Mar 08, 2013 5:22 am

Re: WRF EMS Benchmark results and hardware

Post by theocarter2911 » Wed Oct 02, 2013 2:18 pm

You are absolutely right - there was a huge bottleneck on the ram, going from double to four channel opened it up. I was hoping for a 10-20% speed improvement, and got 50%. Really interesting. Thank you again!

nhampshire
Posts: 2
Joined: Thu Nov 14, 2013 10:35 pm

Re: WRF EMS Benchmark results and hardware

Post by nhampshire » Fri Nov 15, 2013 12:04 am

Ran the Large NMM Nested Benchmark. Does this seem reasonable?

Basic System Information for fwd-lw-wrf

System Date : Thu Nov 14 23:55:13 2013 UTC
System Hostname : fwd-lw-wrf
System Address : Not Resolved

System OS : Linux
Linux Distribution : Red Hat Enterprise Linux Workstation release 6.4 (Santiago)
OS Kernel : 2.6.32-358.23.2.el6.x86_64
Kernel Type : x86_64

Processor and Memory Information for fwd-lw-wrf

CPU Name : Intel(R) Xeon(R) CPU E5-2609 0 @ 2.40GHz
CPU Instructions : sandybridge
CPU Type : 64-bit
CPU Speed : 1200 MHz

EMS Determined Processor Count
Physical CPUs : 2
Cores per CPU : 4
Total Processors : 8

Hyper-Threading : Off

System Memory : 31.2 Gbytes

EMS Release Information for fwd-lw-wrf

EMS Release : 3.4.1.13.43
EMS Binaries : x64


EMS NMM core benchmark nested simulation completed in 55 minutes 39 seconds

theocarter2911
Posts: 61
Joined: Fri Mar 08, 2013 5:22 am

Re: WRF EMS Benchmark results and hardware

Post by theocarter2911 » Mon Nov 18, 2013 12:38 am

Hi nhampshire,

Would like to hear what meteoadriatic thinks too, but the following looks hinky to me - if you look at your cpu info "CPU E5-2609 0 @ 2.40GHz" and then look further down "CPU Speed : 1200 MHz" it looks like your cpu's are running at half their capability? If you look two posts back, at my benchmark for NMM large the cpu info speed and cpu speeds are the same. Not sure what could cause that - they are both xeons so should behave similarly. Maybe a motherboard or bios issue? Your hyperthreading is off, so that is not the problem.

Theo

meteoadriatic
Posts: 1510
Joined: Wed Aug 19, 2009 10:05 am

Re: WRF EMS Benchmark results and hardware

Post by meteoadriatic » Mon Nov 18, 2013 7:43 am

It depends what was actual frequency when ems info script was executed. If computer were idle, frequency drops down to save power, and that will be printed here. If there was any significant load on the CPU, it's frequency will be maximized and then script will read maximum frequency.

Not to worrry much.

If you are still concerned, there is great tool to monitor what is going on with cpu. It is called i7z:
http://code.google.com/p/i7z/

Best regards

mehdii_kk
Posts: 3
Joined: Sun Mar 09, 2014 4:22 pm

Re: WRF EMS Benchmark results and hardware

Post by mehdii_kk » Sun Mar 09, 2014 5:13 pm

WTF :| i use 2 xeon 2660v2 20core @2100mhz and 4gb ramddr3 Ecc-non buffer @ 1333mhz
so plz tell why my time is so long????????? thankX
(i my cmd when top cpu is 51% load !!!)

Cpu(s): :!: :!: 51.5%us :!: :!: , 2.1%sy, 0.0%ni, 46.4%id, 0.0%wa, 0.0%hi, 0.0%si, 0.0%st
Mem: 3994684k total, 2899520k used, 1095164k free, 3700k buffers


Basic System Information for xxx

System Date : Sun Mar 9 16:44:40 2014 UTC
System Hostname : xxx
System Address : 127.0.1.1

System OS : Linux
Linux Distribution : wheezy/sid
OS Kernel : 3.5.0-41-generic
Kernel Type : x86_64

Processor and Memory Information for mazesta

CPU Name : Genuine Intel(R) CPU @ 2.10GHz
CPU Instructions : sandybridge
CPU Type : 64-bit
CPU Speed : 2101 MHz

EMS Determined Processor Count
Physical CPUs : 2
Cores per CPU : 10
Total Processors : 20

Hyper-Threading : On

Note: Attempting to use virtual "Hyper-threaded" CPUs while
running the EMS may result in a degradation in performance.

System Memory : 3.8 Gbytes

EMS Release Information for mazesta

EMS Release : 3.4.1.14.7
EMS Binaries : x64


EMS NMM core benchmark simulation completed in 5 minutes 47 seconds

meteoadriatic
Posts: 1510
Joined: Wed Aug 19, 2009 10:05 am

Re: WRF EMS Benchmark results and hardware

Post by meteoadriatic » Sun Mar 09, 2014 6:41 pm

Hello,

either:
a) turn off HyperThreading in BIOS and don't change configuration, or,
b) use 20 cores/CPU (total 40 cores)

I STRONGLY recommend option a.

mehdii_kk
Posts: 3
Joined: Sun Mar 09, 2014 4:22 pm

Re: WRF EMS Benchmark results and hardware

Post by mehdii_kk » Sun Mar 09, 2014 9:50 pm

Thanks meteoadriatic

wOOOOOOF i change my core to 40
REAL_NODECPUS = local:40
WRFM_NODECPUS = local:40

The WRF NMM core shall be run on the following systems and processors:
40 processors on mazesta (1 tile per processor)

Cpu(s): 99.4%us, 0.6%sy, 0.0%ni, 0.0%id, 0.0%wa, 0.0%hi, 0.0%si, 0.0%st
Mem: 3994684k total, 3887196k used, 107488k free, 17324k buffers


System Date : Sun Mar 9 21:41:42 2014 UTC
System Hostname :
System Address : 127.0.1.1

System OS : Linux
Linux Distribution : wheezy/sid
OS Kernel : 3.5.0-41-generic
Kernel Type : x86_64

Processor and Memory Information for

CPU Name : Genuine Intel(R) CPU @ 2.10GHz
CPU Instructions : sandybridge
CPU Type : 64-bit
CPU Speed : 2101 MHz

EMS Determined Processor Count
Physical CPUs : 2
Cores per CPU : 10
Total Processors : 20

Hyper-Threading : On

Note: Attempting to use virtual "Hyper-threaded" CPUs while
running the EMS may result in a degradation in performance.

System Memory : 3.8 Gbytes

EMS Release Information for

EMS Release : 3.4.1.14.7
EMS Binaries : x64


EMS NMM core benchmark simulation completed in 10 minutes 16 seconds F*****************K :O

theocarter2911
Posts: 61
Joined: Fri Mar 08, 2013 5:22 am

Re: WRF EMS Benchmark results and hardware

Post by theocarter2911 » Wed Mar 12, 2014 1:03 am

Interesting results! - I had on purpose gone with the octacores as the cpu speed was a bit higher - but it really looks like 10 cores beat 8 cores significantly - my best large NMM was 16 minutes and you clocked 10 minutes!
Theo

Post Reply

Who is online

Users browsing this forum: No registered users and 1 guest