WRF EMS Benchmark results and hardware

Looking for new hardware to run WRF? Intel or AMD? Check this forum.
mehdii_kk
Posts: 3
Joined: Sun Mar 09, 2014 4:22 pm

Re: WRF EMS Benchmark results and hardware

Post by mehdii_kk » Wed Mar 12, 2014 7:46 pm

NO i use 2x xeon 2660v2 2x10core ! and my best time is 4min

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

Re: WRF EMS Benchmark results and hardware

Post by theocarter2911 » Sat Mar 15, 2014 6:04 pm

Hi mehdii_kk,
So it is two cpu's of 20 cores each for a total of 40 physical cores (or potentially 80 hyperthreaded logical cores?)
Or is it the i5's with two cpu's of 10 cores each for a total of 20 physical cores (potentially 40 hyperthreaded logical cores)?

My system has two cpu's of 8 cores each - 16 physical cores (potentially 32 hyperthreaded logical cores).

Can you confirm if the 4-5 minutes is on normal NMM benchmark or on "large" NMM benchmark?
Regards,
Theo

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

Re: WRF EMS Benchmark results and hardware

Post by theocarter2911 » Sun Mar 23, 2014 3:10 am

Hi meteoadriatic,

I am a little disappointed in my system speed - 2 8-core xeons at 2.9 Ghz versus mehdii_kk's 2 10-core xeons at 2.1 Ghz? Looking at things in a very basic way comparing raw power it is 46.4 Ghz on my system versus 42 Ghz on his one.

But he is getting 5 minutes versus my 16 minutes?

This is assuming he was doing "large" NMM benchmarks, he did not answer my question so I am unsure if he is really slow on the small benchmark or very quick on the large benchmark.

Am I running too slow here? What do you think? Since the last upgrade of Robert's software, I am running 5-10 minutes slower on getting operational output, which puts me only just after landing forecast time, and I would like to find a way to pull that back again.

Also, Robert is so busy that he has not updated the benchmarks link on the main website for a couple of years now - so it is difficult to do comparisons. Certainly not complaining, he is the guru. :-)

Theo

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

Re: WRF EMS Benchmark results and hardware

Post by meteoadriatic » Sun Mar 23, 2014 7:58 am

Hi,
theocarter2911 wrote:I am unsure if he is really slow on the small benchmark
I think this is the case, I think mehdii_kk was complaining about being slow, but neither I'm really sure from his post if I'm right. Let's wait for him if he comes back.

I don't think you're slow, but I have no means to verify. Probably Robert has better collection of benchmark results around world so he might can tell you is your performance reasonable enough.

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

Re: WRF EMS Benchmark results and hardware

Post by theocarter2911 » Sun Mar 23, 2014 8:20 am

Hi Meteoadriatic,

Thank you for your thoughts! I did manage to squeeze some minutes out of my operational run by minimizing the postprocessing - so I am right on TAF time again now. :-)

Plus I really do not want to bug Robert, he is busy enough as it is!

Cheers,
Theo

jackq7a
Posts: 10
Joined: Tue Oct 18, 2011 2:40 pm

Re: WRF EMS Benchmark results and hardware

Post by jackq7a » Sat Apr 05, 2014 4:58 am

I finally decided to get around to installing this and running the NMM Large benchmark on a newer system. Here are my results:

Basic System Information for jack-server

System Date : Sat Apr 5 04:47:03 2014 UTC
System Hostname : jack-server
System Address : 127.0.1.1

System OS : Linux
Linux Distribution : wheezy/sid
OS Kernel : 3.11.0-19-generic
Kernel Type : x86_64

Processor and Memory Information for jack-server

CPU Name : AMD FX(tm)-8320 Eight-Core Processor
CPU Instructions : bulldozer
CPU Type : 64-bit
CPU Speed : 1400 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 EMS may result in a degradation in performance.

System Memory : 7.5 Gbytes

EMS Release Information for jack-server

EMS Release : 3.4.1.14.11
EMS Binaries : x64


EMS NMM core benchmark simulation completed in 24 minutes 51 seconds

Gippox
Posts: 59
Joined: Sat Mar 02, 2013 1:16 am

Re: WRF EMS Benchmark results and hardware

Post by Gippox » Sat Apr 05, 2014 12:25 pm

Hi jackq7a, with or without nesting ?

jackq7a
Posts: 10
Joined: Tue Oct 18, 2011 2:40 pm

Re: WRF EMS Benchmark results and hardware

Post by jackq7a » Sat Apr 05, 2014 2:51 pm

Without nesting, here it is with nesting:

Basic System Information for jack-server

System Date : Sat Apr 5 06:00:33 2014 UTC
System Hostname : jack-server
System Address : 127.0.1.1

System OS : Linux
Linux Distribution : wheezy/sid
OS Kernel : 3.11.0-19-generic
Kernel Type : x86_64

Processor and Memory Information for jack-server

CPU Name : AMD FX(tm)-8320 Eight-Core Processor
CPU Instructions : bulldozer
CPU Type : 64-bit
CPU Speed : 1400 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 EMS may result in a degradation in performance.

System Memory : 7.5 Gbytes

EMS Release Information for jack-server

EMS Release : 3.4.1.14.11
EMS Binaries : x64


EMS NMM core benchmark nested simulation completed in 56 minutes 5 seconds

jackq7a
Posts: 10
Joined: Tue Oct 18, 2011 2:40 pm

Re: WRF EMS Benchmark results and hardware

Post by jackq7a » Sat Apr 05, 2014 3:08 pm

Now a question: It says hyperthreading is turned on. I didn't think AMD did hyperthreading. Is there a way to turn it off? What is AMD's version of hyperthreading?

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

Re: WRF EMS Benchmark results and hardware

Post by meteoadriatic » Fri Apr 11, 2014 9:48 pm

Hello,

Let's see one Ivy-Bridge-E 6-cores system :)

Core i7-4930K @ 3.40GHz, 16GB RAM
Hyperthreading off
Unmodified EMS, version 3.4.1.14.15, x64 binaries

Notes:
1) latest EMS use total-1 number of cores as default setting, while it might be faster with 6 cores than with 5, I left that as is for benchmark comparisions with other systems.
2) As those new benchmark cases are a bit of mess (two nmm benchs, two arw benchs, all four can use single or nested domains... I run all benchmarks except ARW Large with nesting, didn't wanted to wait that much.


Results:

NMM small (with nesting) 10 minutes 1 seconds
NMM small (no nesting) 3 minutes 43 seconds

NMM large (with nesting) 32 minutes 38 seconds
NMM large (no nesting) 14 minutes 30 seconds

ARW small (with nesting) 17 minutes 51 seconds
ARW small (no nesting) 5 minutes 12 seconds

ARW large (no nesting) 32 minutes 16 seconds


I wanted to compare all those benchs with i5-4670 @ 3.40GHz, 16GB RAM, HT off.

So, difference here is basically 2 cores more for 4930K and 4-channel RAM with 4930K instead of 2-channel with 4670. i5-4670 has little advantage over i7-4930K because it is Haswell architecture, whereas 4930K is one generation older Ivy-Bridge.

NMM small (with nesting) 14 minutes 5 seconds
NMM small (no nesting) 5 minutes 12 seconds

NMM large (with nesting) 49 minutes 17 seconds
NMM large (no nesting) 23 minutes 7 seconds

ARW small (with nesting) 27 minutes 35 seconds
ARW small (no nesting) 8 minutes 8 seconds

ARW large (no nesting) 48 minutes 36 seconds

Around 50% difference. Is it worth around double the price? You need to decide that for your own application.

Post Reply

Who is online

Users browsing this forum: No registered users and 1 guest