[songsh0227]

hi i am flow3d modeling two system.

1. intel i7 4790K, ddr3 16G

2. intel xeon e5-2670v2 * 2EA, ddr3 128G

all conditions are same

i think, no.2 system more fast

but in my test, no.1 system more fast

who knows why???

[Sly]

In order to get an answer to your question, you should provide more information on each system.

1. How many memory modules are there?
2. What is the frequency of the memory modules?
3. What is the motherboard model?
4. How many cells does your simulation have?

[flotus1]

And last not least: how many cores were used for the simulation.

[Sly]

You're right. I was assuming he was using all of them.

[songsh0227]

Originally Posted by Sly

In order to get an answer to your question, you should provide more information on each system.

1. How many memory modules are there?
2. What is the frequency of the memory modules?
3. What is the motherboard model?
4. How many cells does your simulation have?

thank you your answer.

my computer detailed specifications

system 1

cpu : Intel(R) Core(TM) i7-4790K @ 4.00GHz

Motherboard : ASRock H97 Performance

Memory : DDR3 8G PC-12800(800 MHz) * 2EA - 2 slot available



system 2

cpu : Intel(R) Xeon(R) CPU E5-2670v2 @ 2.50GHz * 2EA

Motherboard : Asus Z9PE-D8 WS

Memory : DDR3 16G PC-12800(800 MHz) * 8EA - full bank



totel cell about 30 millions

and using all available cpu

[Sly]

From what I can see, both systems could perform better. Here's what I think you should do:

On the i7:
1. Turn off hyperthreading
2. Populate the remaining 2 memory slots with the same kind of memory module.

On the Xeons:
1. Turn off hyperthreading
2. Adjust CPU #2 to run at 2.5GHz. It's at 1.2GHz right now.

[flotus1]

2. Adjust CPU #2 to run at 2.5GHz. It's at 1.2GHz right now.

No need to do that. The screenshot was taken while the CPU was idle or under low load. Its clock speed will increase automatically once it actually has something to do.

Populate the remaining 2 memory slots with the same kind of memory module.

This could be an option, but this CPU only has 2 memory channels. So you just have to make sure that the 2 memory modules are in the right slots to enable Dual-channel mode.
See the memory tab of cpu-z if this is already the case. Otherwise refer to the mainboard manual how to populate the DIMMs to get maximum performance.

Concerning the original question why the simulation runs slower on the second system:
Maybe the software you are using or some of the specific models you used in your simulation just dont scale well on many cores.
Or there is a totally different factor like disc I/O that limits the speed of the simulation.
You would have to run more tests to find out or contact flow3D support.

[songsh0227]

i have one more question.
why hyperthreading turn off??
thank you.

[songsh0227]

i have one more question.
why hyperthreading turn off??
thank you.

[flotus1]

The benefit of hyperthreading in CFD is questionable to say the least.
The performance of of many CFD approaches highly depends on the available memory bandwith which does not increase with HT activated.
On the other hand, the communication overhead does increase when the mesh is split into a higher number of partitions.
That is why the the performance with HT on is usually not much higher and can even be worse than with HT off.
Not to mention the additional cost for the higher number of HPC licenses you need to run the simulation on twice the number of "cores".
I dont know if this is true for flow3d, but you have this issue with software from other companies.

[beer]

flotus is right. I ran basically the same test with the 4790K but with ansys. HT makes the code slower by ~1% (depending on your mesh, memory etc.). Overclocking has no effect either, the speedup was more or less 0. Memory bandwith is (almost) everything for cfd. My 2400Mhz modules are really kicking in.