Best PC for 4NEC2 Modeling Load? (see NEC2MP post #39)

[nikiml]

Is there anything special that I need to do to get the optimizer to use multiple cores?

I am trying multi-variable frequency swept optimization and it only uses 1 of 4 cores.

OS WinXP 64
4nec2/X 5.7.6

[holl_ands]

See post #21: I did NOT find any improvement in 4nec2 using multiple cores.

[300ohm]

From post #8

Well, I just upgraded my motherboard to one with more overclockability (I havent done a clean install yet, waiting for my 640 GB drive to arrive) and at 3.1 ghz, the time is down to 283.609 seconds. More tweaking to do, heh.

I created a 2GB RamDisk from DataRam (free for up to 4 GB)
http://memory.dataram.com/products-a...wnload-Ramdisk

and installed 4nec2 and the benchmark nec file to it.

The result : 22.885 seconds !!!!!!

I also put my temporary IE10 files on the ramdisk for a bit faster browsing.

The small downside to using ramdisk is that on shutdown, the files and folders on the drive are saved. Then on booting, the files and folders are recopied to the ramdisk, which results in bit longer shutdown and bootup times, but not by very long, maybe 15 to 20 seconds more.

[nikiml]

I have also ran the nec2 engine with its input and output on a ram disk, but never measured any significant speedup. Are you really talking about a 10 fold speed increase?

[300ohm]

Are you really talking about a 10 fold speed increase?

Yes. !!!

Just to be sure, I ran another large nec file, the KISS 6 X 6 mesh reflector with 1 MHz step sweep from 470 to 698 MHz, which now takes about 35 minutes. IIRC a 6 MHz step sweep used to take me over an hour.

If you didn't see a speed increase, are you sure all the program files were on the ramdisk ? It only takes one file out of place to gum up the works.
Oh, and make sure you uninstall the 4nec2 programs and files that were on the hard drive, so you have a single install to the ramdisk.

I used the newest 4nec2 version, 5.5.8

My CM4251 coarse model here:
http://www.wuala.com/300ohm/Document...oarseModel.nec
now only takes 161 seconds to do an AGT test. Before, it took close to a half hour.
Interestingly, Task Manager doesn't show any reads or writes to the ramdisk as the program generates the nec output. Only at the end do I see a little blip.

[300ohm]

Stampeder:

Does anyone know if there is a lot of disk thrashing going on while the processing is happening, or does it happen mostly in ram?
Can someone who has run 4NEC2 on different Windows OSes give a brief summary of performance and UI issues or differences they saw?
Cheers!

300ohm :

Its mostly in ram. And ram usage is generally below 512 mb.

I kind of take back my answer on that one, and I think you were on the right track Stampeder. Task Manager doesn't measure a whole lot of disk activity in either case, but from my experiments it seems like a lot of small ones are going on all the time in 4nec2, and that's probably why RAMDisk helps.

[stampeder]

Hmmm... the cost of RAM is not much these days, so your method is a great time saver! Another option is that with the cost of SSD (Solid State Disk) dropping it might be worthwhile for devout 4NEC2 users to run everything on one of those.

[300ohm]

Yep, I recently got 8 GB of PNY PC16000 DDR3 for $39 and since I noticed win8 preview really didnt use much more than 2 Gb at any one time, that's when I decided to play around with RamDisk. (I recall paying about $400 for 16K of ram for my work Apple 2, serial #0000052, back in 1979, heh.)

Overall, Ive been pleased with win8 preview, once I found workarounds for the stupid Metro interface. Its been super stable, runs all my WinXP programs, and has a good updated task manager and updated Windows Media Center for my Aver TV card. The Aver Media Center would occasionally crash my system, whereas Win 8 Media Center has not yet. Rumor has it that the cheaper Win8 versions wont include Media Center, so that may be a deal killer for me.

[300ohm]

http://users.otenet.gr/~jmsp/

Looks good, might have to give this a try.

Features

As you would expect, performance is the main feature of NEC/MP. Take a look to the following performance tables to see why.
•Performance gain up to 1000% (10x) or even more
(Typical boost is 500-800% on a dual/four core processor)
•True multicore - multithreaded SMP support
•Vectorized loops across the whole program
•Built against the highly accurate Intel Math Kernel (MKL)
and BLAS/LAPACK matrix function library
•Dual precision FP supporting up to 11.000 segments
(even more in the upcoming commercial x64 editions)
•SIMD hardware support (SSE2, SSE3, SSSE3, SSE4.1, SSE4.2 and AVX)
•Asynchronous buffered I/O disk writes
•Command line interface for easy integration with third party tools and end user workflow
•Self-contained Windows executables (XP-Windows 8)
•Fully compatible with 4NEC2 input & output file format
(executables can replace the original NEC2DXSxxx files)
•Fully compatible with standard NEC data files
•FREE as free beer (NEC2/MP x86 edition)

•Affordable NEC-2 native 64-bit edition (Due 3rd quarter 2012)
•Commercial NEC-4 (4.1 or 4.2) versions available for purchase upon presenting proof of valid NEC-4 license ownership
•Custom made builds upon request

Gave it a try.



Yup, I got about about a 500% increase in speed in the benchmark.
http://www.digitalhome.ca/forum/show...t=96747&page=3

Whoa, I just ran a frequency sweep of my CM4251 coarse model from 470 to 806 MHz step 6 in 763 seconds. IIRC, It used to take me more than 24 hours to run it !!! (Im also using RamDisk now)
The new NEC engine is certainly making more use of the processor(s), as its hitting 92% CPU utilization. IIRC before, it didn't go over 50%.

Looks like we're cooking with real gas now, heh.

[300ohm]

The result : 22.885 seconds !!!!!!

OK, using Athan Papadimitriou's faster NEC2 engine http://users.otenet.gr/~jmsp/

, along with my RamDisk, my time to do the benchmark is now 4.617 seconds !!!!!!!!!!!! About a 500% increase over the stock 4NEC2 engines.

That's an awfully long ways from my original time of 283.609 seconds to do the benchmark, heh.

[stampeder]

Awesome multi-threaded goodness, 300ohm! Clearly anyone using 4NEC2 should adopt NEC2MP to get their work done.

http://users.otenet.gr/~jmsp/#dnld

[300ohm]

Another option is that with the cost of SSD (Solid State Disk) dropping it might be worthwhile for devout 4NEC2 users to run everything on one of those.

OK, last month I bought an Intel 330 120GB SSD drive, to use as my system drive. The Intel SSDs seem to be the most reliable, which is more important to me than getting the highest benchmark speed. (The Samsung 830 series is a close second in reliability.) Toms Hardware took apart the Intel 330 and the more expensive Intel 520 and found that they both used the exact same PCB and the exact same synchronous MLC NAND, and Sandforce controller. The only difference being a sticker on the PCB that said bin2, heh.
http://www.tomshardware.com/reviews/...rk,3190-2.html

So far, its been working perfectly. My boot times, counting from when the PC beeps to the desktop have decreased from about 45 seconds to about 8 seconds. Shutdown times have decreased from about 20 seconds to about 5 seconds. Open Office loads in about 3 seconds instead of about 20.

The Intel 330 120GB SSD drive specs on SATA3 are 500MB/s read and 450MB/s write.
My motherboard is only SATA2, so I only get 280MB/s read and 260MB/s write, which is still faster than 2 WD Raptors in raid 0.

My 4nec2 benchmark on it, using the NEC2/MP engines, is 49.328 seconds.

[300ohm]

I have also ran the nec2 engine with its input and output on a ram disk, but never measured any significant speedup.

Nikiml, I think I know why you didn't measure any significant speedup. By default, Windows allows contents on a drive to be indexed. By unchecking that option (no indexing), speed is at least 10 times faster.

I found that out after installing Windows and running the 4nec2 benchmark. It was as slow as a hard drive. Unchecking the indexing box got the benchmark back into fast speeds.

[nikiml]

good catch - i'll try it

[300ohm]

Yeah, for all the years I was using WinXP, I turned the file indexing on the drives off. Windows uses file indexing on the drives to make future searches a little bit faster. But there is an overhead cost for doing so. Since I generally know where I put stuff, I would only do a search once in a blue moon, heh.

Most SSD optimization guides will also say to turn off drive indexing on SSD drives. For one, its not needed. For another reason, it makes unnecessary writes, technically making the drive longevity shorter (but not by a significant amount).

My 4nec2 benchmark on it, using the NEC2/MP engines, is 49.328 seconds.

I think I may have left file indexing on the drive when doing that benchmark, so Ill have to re-do.