CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   Hardware (http://www.cfd-online.com/Forums/hardware/)
-   -   Which is better for CFD 4 core i7-2600 or AMD 8 core FX-8150? (http://www.cfd-online.com/Forums/hardware/95930-better-cfd-4-core-i7-2600-amd-8-core-fx-8150-a.html)

GregShaffer January 8, 2012 10:56

Which is better for CFD 4 core i7-2600 or AMD 8 core FX-8150?
 
Which is better for CFD? A faster 4 core cpu or does AMD having 8 cores make up for it being slower?

Has anyone done any benchmarks for CFD?

GTCo8 January 8, 2012 16:41

Look here:

http://www.cfd-online.com/Forums/har...re-6-core.html

scipy January 13, 2012 07:43

I've actually done a bit more testing over more iterations and with i7's HT turned on vs the FX8150, here are the results:

First batch of test for the i7-2600k:

1. HyperThreading = off (only physical cores used) @ stock speed (3.4 GHz)

on 4 cores = 31.6 seconds/iteration (1st order disc.)
on 4 cores = 32.9 seconds/iteration (2nd order disc.)
on 3 cores = 35.2 seconds/iteration (1st order disc.)
on 3 cores = 36.8 seconds/iteration (2nd order disc.)

2. HyperThreading = off @ overclocked 4.5 GHz

on 4 cores = 30.5 seconds/iteration (1st order disc.)
on 4 cores = 32.6 seconds/iteration (2nd order disc.)
on 3 cores = 34.0 sec/iter (1st order disc.)
on 3 cores = 35.6 sec/iter (2nd order disc.)

Second batch of test for the i7-2600k:

1. HyperThreading = on (real + virtual cores used) @ overclocked 4.5 GHz

on 8 cores = 28.0 sec/iter (1st order disc.)
on 8 cores = 29.5 sec/iter (2nd order disc.)
on 7 cores = 28.7 sec/iter (1st order disc.)
on 7 cores = 30.4 sec/iter (2nd order disc.)

Tests on the FX 8150:

on 8 cores = 25.2 sec/iter (1st order disc.)
on 8 cores = 26.0 sec/iter (2nd order disc.)
on 7 cores = 25.8 sec/iter (1st order disc.)
on 7 cores = 27.1 sec/iter (2nd order disc.)

Conclusions

Even though ANSYS says HyperThreading should be turned off since Fluent only uses physical CPU cores, there is a performance increase of 12.8 % when using HT, however, the main reason to leave HT on is if you want to do something on your pc during the simulation. If HT is off you are going to run the simulation on 3 cores, but if HT is on you can run it on 7 (since Windows will take advantage of the HT and you will have the same performance as with 1 real core) with increase of 22.6 % in simulation speed.
The overclocking aspect doesn't seem to have much effect on the Fluent speed, the increase in CPU clock speed of ~33 % gained about 3.5 % increase in simulation speed. The increase in electric power usage of 12 % (cpu only) hardly makes this worth the effort, especially when you factor in that at 4.5 GHz the cooling system was running at maximum settings (3/3) and at stock speed 1/3 setting was more than enough to keep the temperature levels same.
The FX8150 is slower per physical core, but since you do have 8 of them it is around 11-15 % faster than the i7-2600k. This all holds if you're not paying the licenses for Fluent yourself, if you are than the price of license cannot justify the slower per core performance of the AMD and you'd be better off getting two i5-2500k's since they'll scale linearly up to 4 nodes even on regular GigE ethernet.

This corrects my previous statements from the other thread (I've had apps using the CPU power from the i7 during tests which slowed it down quite a bit), the FX8150 doesn't really have a 25 % advantage over the i7. It's more along the lines of 12-13 % in realistic scenarios.


All times are GMT -4. The time now is 01:20.