why the simulation speed doesn't increase with hfss 10

[dingjingfeng]

vista 64 hard disk controller speed slow

now I use lunix server with 2GHz 64bit processor and >10GB memory.
I notice when the used memory is up to 2GB, the simulation speed is very slowly.
the Version of HFSS is 10.1.1

what's the problem behind???

 

[Azulykit]

why so much disk when ram available in hfss

Check with Ansoft tech support. Maybe you need to adjust an environmental variable setting to tell your computer that it has a big brain.

It sounds like you have a nice machine.

 

[dozy_walia]

hfss starts not swapping to disk

wts RAM?

 

[dingjingfeng]

hfss requirement

>10GB memory.
this is a lunix server

 

[maxwellian]

access more than 2 gb on a 32 bit system hfss

Are you using a 64-bit version of the Linux OS, or a 32-bit version?

--Max

 

[dingjingfeng]

hfss does not allocate virtual memory

Sorry my server is unix
and the software envierment is KDE
I think this is a 32bit system
because this version is 2002.
is it better to update the OS, to improve the speed of HFSS?

 

[maxwellian]

hfss compatibility results 64 32 bit

Upgrading to the 64-bit OS might help. I am not sure about HFSS, but other memory-intensive programs are sometimes limited by 32-bit OS in how much contiguous memory can be allocated at one time. In Windows, you start to see problems when you try to allocate around 1.8 GB, unless you use the /3GB bootswitch.

I am not sure, but I think that most 32-bit Linux platforms would at least let you allocate 3 GB for a single process. However, I am not sure what other issues may be going on because of the 32-bit limitataion on memory allocation for the OS. For Linux, I would have expected that you would hit the limitation at a larger problem size.

I am not an expert with KDE (or most any other type of Linux) kernel configuration. Maybe you can check for kernel settings that you could change to make it handle larger process sizes more efficiently? Pagefile size, or something else?

Good luck,

--Max

 

[Azulykit]

To take full advantage of the RAM you have available you need to be running a 64 bit OS. In my case I am using the 64 bit version of XP Pro.

The Ansoft folks told me that they were working on a Linux version but I don't know if it is available.

I would suggest discussing your hardware and OS options with them to get you the best performance.

 

[rfmw]

@Azulykit

how much RAM do you have in your 64-bit WinXP maschine?

If more than 4GB, when does the HFSS start to swap to local disk?


For example, I have 32-bit WinXP with 2GB of RAM. Large disk swaping begins at 1.6GB of RAM usage...

 

[Azulykit]

You are on target and see the problem with a 32 bit OSas it is limited to 2 GB RAM. I believe that the 64 bit XP Pro OS will address up to 16 GB RAM. The box you mentioned is just a little bigger than mine ( 8 GB RAM) and should be just fine for all but the largest (big and high frequencies).

I usually just give up if I see swapping going on. The solution times are just too long. When that happens I try to reduce the magnitude of the problem with symmetry planes, less detail, etc.

There is no such thing as a computer that is too big or too fast. Unfortunately, there are plenty that are too expensive.

 

[HOHOH]

I have a experiance in XP(32bit) and Vista(64bit).

CPU : AMD Athlon 64 X2 4400
RAM : 4 GB

I applied to simulate waveguide to coaxial adapter by hfss 10.0.

The result of hfss in XP(32bit) is 2 times fast more than vista(64bit) condition.

It's a simple comparison of result and I don't know accuracy method for comparison.

Please recommend method. I can re-try evaluation.

 

[Azulykit]

I wanted to focus my comments a bit. For large problems a considerable amount of memory is needed. What HFSS does when the requirement for RAM exceeds that available is use the hard drive for additional space. This is what is referred to as swapping and it really slows the analysis. My experience is that once this starts happening one may as well give up finding a solution. The only choice its to rework the model.

More RAM in my mind is valuable in that it facilitates the solution of larger problems while avoiding swapping. A 64 bit OS can address significantly more RAM than a 32 bit OS. I believe that the number is 2 GB vs 32 GB. I am sure someone will correct this comparison if I missed.

Processor speed is an issue too but I am more interested in being able to analyze large problems than have the absolutely highest calculation rate possible. I wonder if the waveguide test case cited was processor dependent rather than available RAM limited.

My suspicion is that it is more useful to have more RAM than multiple processors and higher speeds. This is particularly the case as both the hardware and the license cost is higher for multi-processor computers.

 

[dingjingfeng]

In my opnion, With 64bit cpu, calculation speed should be much higher than 32bit!

because a word of 64bit can experss much high accrate number than 32bit word

correct??


for a same frequency cpu, the calculation speed of 64bit cpu should much faster than that of 32bit cpu
???

I have a experiance in XP(32bit) and Vista(64bit).

CPU : AMD Athlon 64 X2 4400
RAM : 4 GB

I applied to simulate waveguide to coaxial adapter by hfss 10.0.

The result of hfss in XP(32bit) is 2 times fast more than vista(64bit) condition.

It's a simple comparison of result and I don't know accuracy method for comparison.

Please recommend method. I can re-try evaluation.

 

[HOHOH]

I tested same pc.

In my case, 32bit is faster than 64 bit.

I don't know reason.

 

[maxwellian]

I have a experiance in XP(32bit) and Vista(64bit).

CPU : AMD Athlon 64 X2 4400
RAM : 4 GB

I applied to simulate waveguide to coaxial adapter by hfss 10.0.

The result of hfss in XP(32bit) is 2 times fast more than vista(64bit) condition.

It's a simple comparison of result and I don't know accuracy method for comparison.

Please recommend method. I can re-try evaluation.

I think there is a popular misunderstanding that 64-bit processors are faster for a lot of reasons. Some people think that they perform more multiplications per cycle because they handle a larger accumulator size. Other people think that there are special features in the way the ALU works.

In the latest class of 64-bit processors, there are a lot of things that work to provide faster analysis:
- Bigger L2 Cache size. This makes a very big difference in that you keep more data closer to the CPU where intense calculations are taking place. Less fetches have to be made to the memory to move data close to the CPU for processing.
- Better Front-Side Bus (FSB) - New FSBs are faster and more efficient. For EM simulation, faster FSB has a big effect on overall processing speed.
- Some EM simulators are starting to use more advanced features in the new 64-bit processors that were not available in older processors.
- Faster clock speeds. Processing time in EM simulators scale more or less with the clock speed.

Changing to a 64-bit OS shouldn't change any of these factors.

I really agree with Azulykit that if you have 2 GB of RAM on your computer, and you have solver process that requires around 1.5 GB of RAM (or maybe a little larger) you eventually get to the point where the OS cannot hold everything in RAM; it has to "borrow" virtual RAM space from the hard disk. This is what we sometimes call "Swapping to disk." And when it starts to happen, you might as well kill the job and rework the model, as has been stated already. Swapping to disk can slow the job 100x or more because instead of accessing RAM, your computer is now trying to access some of that memory through your very slow hard disk controller (at least it's very slow compared to RAM access).

Slower simulation times on Vista can be from several reasons:
1. Vista, as an OS, requires a *lot* more basic RAM just for the OS than did Windows XP. So the largest effective job you can run is probably smaller on your PC under Vista than it was under XP because you simply have less RAM space left to use. Vista is very wasteful of RAM!
2. Vista also has a few more things going on in the OS than the older OSs.
3. HFSS might not be optimal for Vista. A lot of software developers are finding that Vista is forcing them to change a lot of stuff to be properly compatible. (Does Ansoft actually say that they support 64-bit Vista in this release? I don't think so, so this might be part of the problem. While the software might run in Vista/64, it isn't likely to have been optimized for it.)

You might want to try benchmarking other applications besides just HFSS. This might tell you if the problem is HFSS, or if you are finding that your computer simply runs slower under Vista/64.