Mini cluster Information: would 1x i7 7700 + 1x i7 8700 be OK?

Astan · December 23, 2018, 15:44

Hello everyone, I kindly wanted to ask you for an opinion.

I'm doing a direct RANS simulation on a (simplified) car, mesh of about 30 or 40 million cells, with OpenFOAM
Currently I am using a workstation with an i7 7700 processor clocked 3.60 - 4.20 Ghz and quad-core and 40GB ram ddr4 2400 mhz.
I thought of building a simple 2-node cluster by purchasing a second workstation with an i7 8700 processor with six cores clocked 3.20 - 4.60 Ghz and 32Gb ram ddr4 2666 mhz.

In your opinion, could the calculation time be halved?

Thanks for the time taken to read the message.

Astan

pibil1 · December 24, 2018, 11:29

I don't think so, benchmarks evaluates them as 60% faster, and the downside is that they both are dual channel. I would rather look for i7 5820k quad channel CPU build.

Reference:

https://cpu.userbenchmark.com/Compar...700/3940vs3887

Astan · December 24, 2018, 18:36

Hi Pibil1 Merry Christmas, thanks you for you answer.

how can i deduce from the link you sendt me that if I join those two pc speed is increased by 60%?

flotus1 · December 25, 2018, 06:12

I don't think pibil1 necessarily meant connecting these two machines.
An I7-5820k can be twice as fast compared to an I7-7700 all by itself.

Did you investigate if the cases you run fit into memory? 40 million cells with 40GB of RAM could be quite close.

naffrancois · December 25, 2018, 17:31

If you can have a second workstation for a few bucks and it time is cheap then it may be a good option. This is what I do with unused workstations. Otherwise, as stated by the other forumers it may be cheaper to go for a more powerful setup.

I agree with the remark of flotus 40milion cells for 40gb ram, you may be on the edge

Astan · December 26, 2018, 09:58

hi naffrancois and flotus1
Thank you very much for your answers and happy holidays.

Yes, on a single node I'm at the limit of ram.

But I thought that on 2 nodes, reaching 72 GB of ram that would no longer be a problem.
However, i'm not sure about the improvement of the calculation speed.

Based on a study of "speed up" and parallel efficiency, when in your opinion would be better to increase the number of cores?

Astan · January 2, 2019, 06:07

hello guys, happy new year!

These days I have done some simulations on the cavity and motorbike test cases.
I performed simulation with 1 core and 2 cores with i7 7700.

I calculated the speed up: clocktime (1 core) / clocktime (2 cores) and I get 1.6 and not 2.

What can this problem depend on?

Astan

Simbelmynë · January 2, 2019, 06:20

Most likely because you are not bandwidth limited on 1 core in combination with the fact that 1 core turbo is higher than 2 core turbo (i.e. the CPU is running at a higher frequency on 1 core compared to 2 cores).

Astan · January 2, 2019, 09:09

Hi Simbelmynë thank you for your answer, i have tried to turn off the speed boost and done the simulation again. But the speed up is still less than 2.

What could i try to solve the problem?

Astan

Simbelmynë · January 2, 2019, 12:14

If you have made sure that the frequency is not dynamically altered then I would guess that your memory is limiting the results. You could try to under-clock all cores to test this if you are accustomed to this kind of tinkering.

Also, a speed-up of 1.6 is not all that bad either since most i7 CPUs seem to be in the 1.7-1.8 span when going from 1 to 2 cores (with memory @ 3200 MHz).

Simbelmynë · January 2, 2019, 12:30

*double post - please delete at convencience*

December 23, 2018, 15:44	Mini cluster Information: would 1x i7 7700 + 1x i7 8700 be OK?	#1
Astan Member Andrew Join Date: Mar 2018 Posts: 82 Rep Power: 8	Hello everyone, I kindly wanted to ask you for an opinion. I'm doing a direct RANS simulation on a (simplified) car, mesh of about 30 or 40 million cells, with OpenFOAM Currently I am using a workstation with an i7 7700 processor clocked 3.60 - 4.20 Ghz and quad-core and 40GB ram ddr4 2400 mhz. I thought of building a simple 2-node cluster by purchasing a second workstation with an i7 8700 processor with six cores clocked 3.20 - 4.60 Ghz and 32Gb ram ddr4 2666 mhz. In your opinion, could the calculation time be halved? Thanks for the time taken to read the message. Astan

January 2, 2019, 12:30		#11
Simbelmynë Senior Member Join Date: May 2012 Posts: 546 Rep Power: 15	double post - please delete at convencience Last edited by Simbelmynë; January 2, 2019 at 12:36. Reason: double post

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
ERROR #001100279 has occurred in subroutine ErrAction.	smnaryal	CFX	11	December 20, 2017 16:32
[OpenFOAM.org] OpenFOAM Cluster Setup for Beginners	Ruli	OpenFOAM Installation	7	July 22, 2016 04:14
libz.so.1: no version information available	dmaz	OpenFOAM Running, Solving & CFD	3	January 4, 2015 16:54
Information / Hardware / Computer / Cluster	Tobi	Hardware	1	October 16, 2013 15:08
Terrible Mistake In Fluid Dynamics History	Abhi	Main CFD Forum	12	July 8, 2002 09:11

December 24, 2018, 11:29		#2
pibil1 New Member Join Date: Aug 2017 Posts: 28 Rep Power: 8	I don't think so, benchmarks evaluates them as 60% faster, and the downside is that they both are dual channel. I would rather look for i7 5820k quad channel CPU build. Reference: https://cpu.userbenchmark.com/Compar...700/3940vs3887

December 24, 2018, 18:36		#3
Astan Member Andrew Join Date: Mar 2018 Posts: 82 Rep Power: 8	Hi Pibil1 Merry Christmas, thanks you for you answer. how can i deduce from the link you sendt me that if I join those two pc speed is increased by 60%?

December 25, 2018, 06:12		#4
flotus1 Super Moderator Alex Join Date: Jun 2012 Location: Germany Posts: 3,399 Rep Power: 46	I don't think pibil1 necessarily meant connecting these two machines. An I7-5820k can be twice as fast compared to an I7-7700 all by itself. Did you investigate if the cases you run fit into memory? 40 million cells with 40GB of RAM could be quite close.

December 25, 2018, 17:31		#5
naffrancois Senior Member Join Date: Oct 2011 Posts: 239 Rep Power: 16	If you can have a second workstation for a few bucks and it time is cheap then it may be a good option. This is what I do with unused workstations. Otherwise, as stated by the other forumers it may be cheaper to go for a more powerful setup. I agree with the remark of flotus 40milion cells for 40gb ram, you may be on the edge

December 26, 2018, 09:58		#6
Astan Member Andrew Join Date: Mar 2018 Posts: 82 Rep Power: 8	hi naffrancois and flotus1 Thank you very much for your answers and happy holidays. Yes, on a single node I'm at the limit of ram. But I thought that on 2 nodes, reaching 72 GB of ram that would no longer be a problem. However, i'm not sure about the improvement of the calculation speed. Based on a study of "speed up" and parallel efficiency, when in your opinion would be better to increase the number of cores?

January 2, 2019, 06:07		#7
Astan Member Andrew Join Date: Mar 2018 Posts: 82 Rep Power: 8	hello guys, happy new year! These days I have done some simulations on the cavity and motorbike test cases. I performed simulation with 1 core and 2 cores with i7 7700. I calculated the speed up: clocktime (1 core) / clocktime (2 cores) and I get 1.6 and not 2. What can this problem depend on? Astan

January 2, 2019, 06:20		#8
Simbelmynë Senior Member Join Date: May 2012 Posts: 546 Rep Power: 15	Most likely because you are not bandwidth limited on 1 core in combination with the fact that 1 core turbo is higher than 2 core turbo (i.e. the CPU is running at a higher frequency on 1 core compared to 2 cores).

January 2, 2019, 09:09		#9
Astan Member Andrew Join Date: Mar 2018 Posts: 82 Rep Power: 8	Hi Simbelmynë thank you for your answer, i have tried to turn off the speed boost and done the simulation again. But the speed up is still less than 2. What could i try to solve the problem? Astan

January 2, 2019, 12:14		#10
Simbelmynë Senior Member Join Date: May 2012 Posts: 546 Rep Power: 15	If you have made sure that the frequency is not dynamically altered then I would guess that your memory is limiting the results. You could try to under-clock all cores to test this if you are accustomed to this kind of tinkering. Also, a speed-up of 1.6 is not all that bad either since most i7 CPUs seem to be in the 1.7-1.8 span when going from 1 to 2 cores (with memory @ 3200 MHz).