微波EDA网,见证研发工程师的成长!
首页 > 研发问答 > 微波和射频技术 > 电磁仿真讨论 > EM simulation on X5482 or X5472 Xeon processors

EM simulation on X5482 or X5472 Xeon processors

时间:03-31 整理:3721RD 点击:
Dear all,

I was wondering if anyone has done any EM simulation, particularly CST MWS and HFSS, on systems with high-end Intel Xeon processors, i.e. X5482 (3.2 GHz) or X5472 (3.16 GHz). They?re both of quad-core type with 1600 FSB. I would like to know if the extra simulation speed they provide is worth their price. Related data would be appreciated.

Thanks in advance,

wave-maniac

Hi wave-m

Now that intel come with the new LGA775 CPU I think that better to buy one of them
less price better performance. using chipset X38 and with over clocking to cpu and ddr 1333 you can get 1600MHz and more and for the Xeon you are limited to use
up to ECC DDR 800MHz an expensive memory.
the benefit of Xeon that you can get 2(workstation) or 4(server) CPU's on single machine but now that you have Quad in LGA775 you can get almost the same performance in less price.

I attached here the chipset of the two to show you the difference. and spec. of CPU's

pl

Dear plasma,

Thanks for your post. As far as I have read in various computer technology websites, despite their very similar structure, "Intel 775pin" processors are tuned for gaming and normal computer applications whereas Intel Xeon series 5400 have been tweaked to provide highest massive number crunching throughput.

Regards

From my experience some games use massive computation more than EM
Simulation. That's the reason why a Game companies use the graphic card for computing. And now you can see that's EM companies is following them.
I attached here some part from acceleware,com
Companies typically purchase large amounts of hardware in an attempt to reduce runtimes of computing tasks; multiple servers that occupy a lot of space and generate additional costs associated with housing, operating and maintaining this equipment.
In a case where the user's software can run on only one computer at a time, companies are then constrained by the available resources in that particular machine.
The use of GPUs to address general-purpose computing tasks is an emerging competitive technology in fields where users engage in intensive computer modeling, such as seismic data processing, biomedical imaging and electromagnetic simulation. Instead of a computer's CPU handling the data processing required to complete the task, the GPU is used. Graphics cards possess much greater computational parallelism than single or multi-core CPU computing platforms. For example, while today's Intel or AMD CPUs can perform four or eight tasks in parallel, today's GPUs can perform 128 or more tasks in parallel.
Now Intel to Prevent emigration towards graphic cards put more effort to 775CPU and if you compare it to Xeon it will have better performance than Xeon.
The benefit of Xeon that you can use more than one CPU on same machine. But now
That you have 775CPU 64bit quad you can get almost the same performance in a less price.
What we have to do now is to ask members in forum to run some benchmark to
New CPUs, Xeon and non Xeon and to see the results.

pl

I read somewhere that the GPU in a high-end GeForce 8800 graphics card can do a couple hundred gigaflops/sec. That's enormously faster than any motherboard. However, the graphics card's limited data I/O bandwidth is a bottleneck for many number crunching applications -- the GPU idles most of the time, waiting for data transfer. I don't know the details, though. nVidia provides a MATLAB library so you can do some processing tasks easily in the GPU.

That GPU chip consumes around 180 watts. Ouch!

Dear plasma,

Thanks for the info. As was mentioned in my first question, I was looking for a comparison between the latest Xeon processors and their predecessors. My comment about those non-Xeon Intel processors being tweaked for gaming does not have anything to do with application of GPUs for simulations. As to this last topic, I would like to point out:

Of the three main EM simulation methods, i.e. FDTD, MOM and finite element, only FDTD, or better say time domain techniques have benefited from GPU or GPGPU. For moderate to large boundary value problems the other two need much more memory than GPUs can have integrated on their boards (I normally use at least 10 GB and up to 32 GB of memory for finite element). Using the memory from outside graphic card substantially compromises their speed. On the other hand, as of now there is no efficient implementation of finite element or moment method for GPUs or GPGPUs. By efficient, I mean a speed increase of at least a few times and also one which does not fully involve the CPU, and is suitable for a wide range of cases. If you know of any, please let me know.

Anyway, FDTD does not cover all EM simulations. For example, for high-Q structures, particularly where the simulated device does not have a good impedance matching (such as when an antenna is designed for a goal other than low VSWR and is matched later), ?full convergence? does not mean accurate results. Prediction using signal processing techniques cannot resolve the issue either. For such cases the best alternative is either FE or MOM, hence no related use for GPUs.

I think it's a matter of time that EDA companies using FE or MOM will use this graphic card the price of DDR dropping and you can find graphic cards with memory of 3Gb and 2GPU. the problem is only to integrate between the software and the GPU. I don't
think this GPU parallel computing limited to FDTD.
look at
http://www.nvidia.com/object/tesla_deskside.html
http://www.nvidia.com/object/tesla_features.html

pl

Until the time those EDA companies come up with an efficient solution for accelerating MOM and FE using graphic cards, I have to stick to existing fastest CPUs. Anyway, I've already paid for a dual Xeon X5482 system. I hope it's not going to let me down.

Hey Wave Maniac, how much RAM is going to sweat with the 8 cores? And is it again DELL mashine? I think you're not going to build-assemble it by yourself, are you?

I must say, DELL 690 is stable as a rock so far on the WinXP 64-bit OS.

Dear rfmw,

For the start, I've chosen 32GB but it's expandable to 64 GB if necessary. It is a T7400 workstation.

Copyright © 2017-2020 微波EDA网 版权所有

网站地图

Top