| Author | Message | ||
     Rainer Maassen (rmaassen) New member Username: rmaassen Post Number: 2 Registered: 07-2005 |
We are an R&D group developing sensors for conductivity measurement in aquatic solutions and are modelling electrodynamics, heat transfer and fluid dynamics systems. Our goal is prototyping by simulations before we go to the hardware iteration steps. Up to now FlexPDE supports our needs ideally because we are free to define the structure of our models without limitations to a number of preconfigured "physics". Computing time begins to play an important role of our developing process now. We use quite conventional low-cost PC machines under Windows 2000, Windows XP and 32-bit Linux. Now we plan to invest in hardware to speed up our design cycle. What are the key components for configuring a high-speed machine for the use of FlexPDE? Is there an advantage of 64-bit Linux over 32-bit Windows? Will a 64-bit version of FlexPDE be available in near future and would it be fundamentally faster? Any experience or recommendations would be much appreciated. Regards, Rainer P.S.: Sorry for this post first placed in the wrong forum. | ||
| Robert G. Nelson (rgnelson) Moderator Username: rgnelson Post Number: 401 Registered: 06-2003 |
We have not yet installed a 64-bit machine in house, so I can only relate what I have read. 1) The 32/64 issue relates to the amount of addressable memory that the operating system can manage and that an application can use. Under Windows XP/32 for example, the limit is 3 GB for any single task, and presumably 4GB total. If your problems are using over 1GB of memory (as reported on the Status window or the .log file), then 64 bits may be important to you. 2) There is no significant advantage to be expected for Linux over Windows or 64-bits over 32-bits as far as execution speed. The speed is controlled predominantly by the processor clock rate. Memory transfer rate will also be important. Depending on the problem, a larger memory cache may be a significant factor. To a smaller extent, speed will be affected by the quality of code compiled by the C++ compiler. The Windows compiler is slightly better than the GCC compiler we use on Linux. 3) FlexPDE is currently a 32-bit application, so a 64-bit machine will not increase the 3GB maximum memory addressable by FlexPDE, but may be much superior in the handling of overall memory use by all tasks in the system if you install over 4GB total memory. In theory, the 64-bit operating systems can run 32-bit applications seamlessly. 4) We plan to generate a 64-bit version of FlexPDE this summer. We do not expect performance gains in this effort, but only an addressable memory increase. There are reasons to expect slightly slower performance due to larger instruction size, but we will see. 5) Dual processors can be used by FlexPDE to a limited extent, with performance improvements of 30 to 50 percent. Dual processors are much better for interactivity of other tasks running simultaneously with FlexPDE. We have ordered a dual-core Athlon 64 for our development work, and will be investigating the dual-core machines for performance improvement. |