[hitzhwan]

Does the calculation speed of Ansys in the Linux system is higher than that in the window system?

[aerosayan]

Scientific code depends on ram and CPU being free.

on Linux the code runs natively.

I don't know what's "wl". I think you meant "wsl" i.e windows subsystem for Linux. There, both the windows is and the Linux is has to run together.

We don't even know how much virtual emulation takes place. So, slowdowns are expected.

[LuckyTran]

For all practical purposes, the speed is the same. You have to get very creative for the speeds to become significantly different.

[hitzhwan]

Originally Posted by aerosayan

Scientific code depends on ram and CPU being free.

on Linux the code runs natively.

I don't know what's "wl". I think you meant "wsl" i.e windows subsystem for Linux. There, both the windows is and the Linux is has to run together.

We don't even know how much virtual emulation takes place. So, slowdowns are expected.

Do you mean the calculation speed only depends on the cpu and ram, it does not depends on the system?

[hitzhwan]

Originally Posted by LuckyTran

For all practical purposes, the speed is the same. You have to get very creative for the speeds to become significantly different.

1.Are you sure they are the same? Have you compared in the Ansys calculation?

2. What is the meaning of "creative"?

[LuckyTran]

I have done the test many many years ago. I built two almost nearly identical Nehalem system. One was a 9x0 and the other was a 9x0+10 build. Both ran triple channel DDR3 RAM back when this was state of the art. Both were ram bandwidth limited. One ran windows and the other ran Ubuntu. Today, you are still RAM bandwidth limited unless you have poured a ton of $ into your setup and are using hex/oct-channel setups. If you're not bandwidth limited, then your problem is simply too small to need any RAM and speed was never an issue. I left LES cases running for months and both had identical run times, iteration for iteration, over 3+ months.

I also switched to lower cas latency RAM at one point from latency of 9 to 8 and immediately got a 20% faster calculation time. The operating system is insignificant in a production environment. Millions of other factors are much more meaningful, like what hardware you are actually using.

The difference between a linux environment and windows environment is which compiled variant you are using. Unless you work for Ansys, you are going to be using a pre-compiled distribution where they have already compiled the code for you (because they're not gonna hand over the source code) and to get the compiled version to work, you just need to make sure you have the correct libraries. For windows, this mean you have a specific version of visual c++ installed. And for linux, this means you have the equivalent libraries. The thing is, the visual c++ libraries are derived from the same unix ones. And Ansys has compiled their code with this in mind already. All this is a longwinded way of saying you need to install the prereqs.

What I mean by creative is... you need to compile it from sourcecode with optimizations for your setup. Unless you actually have the source code, you'll never be able to do this. And even if you did, it's not trivial what compilation changes even need to be made to achieve a difference in performance.

So yes, I am sure they are the same. I have done the test. And yes, I wasted several months of cpu time to test something so trivial, I am absolutely ashamed of it.

[hitzhwan]

Originally Posted by LuckyTran

I have done the test many many years ago. I built two almost nearly identical Nehalem system. One was a 9x0 and the other was a 9x0+10 build. Both ran triple channel DDR3 RAM back when this was state of the art. Both were ram bandwidth limited. One ran windows and the other ran Ubuntu. Today, you are still RAM bandwidth limited unless you have poured a ton of $ into your setup and are using hex/oct-channel setups. If you're not bandwidth limited, then your problem is simply too small to need any RAM and speed was never an issue. I left LES cases running for months and both had identical run times, iteration for iteration, over 3+ months.

I also switched to lower cas latency RAM at one point from latency of 9 to 8 and immediately got a 20% faster calculation time. The operating system is insignificant in a production environment. Millions of other factors are much more meaningful, like what hardware you are actually using.

The difference between a linux environment and windows environment is which compiled variant you are using. Unless you work for Ansys, you are going to be using a pre-compiled distribution where they have already compiled the code for you (because they're not gonna hand over the source code) and to get the compiled version to work, you just need to make sure you have the correct libraries. For windows, this mean you have a specific version of visual c++ installed. And for linux, this means you have the equivalent libraries. The thing is, the visual c++ libraries are derived from the same unix ones. And Ansys has compiled their code with this in mind already. All this is a longwinded way of saying you need to install the prereqs.

What I mean by creative is... you need to compile it from sourcecode with optimizations for your setup. Unless you actually have the source code, you'll never be able to do this. And even if you did, it's not trivial what compilation changes even need to be made to achieve a difference in performance.

So yes, I am sure they are the same. I have done the test. And yes, I wasted several months of cpu time to test something so trivial, I am absolutely ashamed of it.

Hi，thank you so much for your detail information, so in your opinion and experience, there is no difference for the calculation speed of Ansys between the Linux system and Windows system? It has no meaning to compare with them?