POPPUR爱换

标题: matlab Release 2010b 发布,正式集成 CUDA 加速 [打印本页]

作者: Edison    时间: 2010-10-13 18:54
标题: matlab Release 2010b 发布,正式集成 CUDA 加速
http://www.mathworks.com/product ... _9b_homepage_268513

What's New in Release 2010b

Release 2010b includes new features in MATLAB and Simulink, one new product, and updates and bug fixes to 85 other products. Subscribers to MathWorks Software Maintenance Service can download product updates. Visit the License Center to download products, activate software, and manage your license and user information.
New capabilities for the MATLAB product family include:

    * Custom enumerated data types, 64-bit integer arithmetic, and desktop enhancements in MATLAB
    * GPU computing with CUDA-enabled NVIDIA devices in Parallel Computing Toolbox
    * Support for the GigE Vision hardware standard in Image Acquisition Toolbox
    * Automated PID tuning in Control System Toolbox
    * New System objects for communications design in MATLAB, supporting 95 algorithms in Communications Blockset
    * Spline Toolbox capabilities merged into Curve Fitting Toolbox
    * OAS and CDS calculations in Fixed Income Toolbox, Reuters Contribute functionality in Datafeed Toolbox, and credit risk enhancements in Financial Toolbox
    * Graphical tool for fitting dynamic networks to time-series data in Neural Network Toolbox
    * Next-Generation Sequencing Browser in Bioinformatics Toolbox and time lag, error models, and covariate analysis in SimBiology

New capabilities for the Simulink product family include:

    * Arrays of buses with For Each subsystems for reducing Simulink model size, simulation time, and data copies for large-scale models
    * Subsystem variants in Simulink and atomic subcharts in Stateflow to create reusable model components
    * Test harness generation, data logging, and scripted test execution in Simulink Verification and Validation
    * Parallel builds of model reference hierarchies with Real-Time Workshop and MATLAB Distributed Computing Server
    * SimRF, based on Simscape, including circuit envelope simulation and RF Blockset capabilities
    * FPGA Workflow Advisor, critical path highlighting, and DO-254 support in Simulink HDL Coder
    * Multicore support for multirate models in xPC Target and xPC Target Turnkey systems

New capabilities for Polyspace code verification products include:

    * Web interface for viewing project metrics in Polyspace Server for C/C++ and Polyspace Server for Ada
    * Support for all statically enforceable MISRA-C++ coding standards in Polyspace Client for C/C++
作者: Edison    时间: 2010-10-13 19:06
Limitations and Restrictions

The following limitations apply to the code within the function that arrayfun is evaluating on a GPU.

    *

      The arrayfun function argument must be a handle to a MATLAB function, whose function file (not a script) defines a single function.
    *

      The code can call only those supported functions listed above, and cannot call scripts. Overloading the supported functions is not allowed.
    *

      Indexing (subsasgn, subsref) is not supported.
    *

      The following language features are not supported: persistent or global variables; if, for, while, parfor, spmd, switch, try/catch, and return.
    *

      All double calculations are IEEE-compliant, but because of hardware limitations, single calculations are not.
    *

      The only supported data type conversions are single, double, int32, uint32, and logical.
    *

      eps is supported without input or with numeric input, but not with a string input. For example, you can call eps and eps(x), where x evaluates to a number; but you cannot call eps('single') or eps('double').
    *

      Functional forms of arithmetic operators are not supported, but symbol operators are. For example, the function cannot contain a call to plus, but it can use the + operator.
    *

      Like arrayfun in MATLAB, matrix exponential power, multiplication, and division (^, *, /, ) perform element-wise calculations only.
    *

      There is no ans variable to hold unassigned computation results. Make sure to explicitly assign to variables the results of all calculations that you are interested in.
作者: 凹特    时间: 2010-10-13 19:07
据说2010b精简了不少工具箱,我还在下载,改天换上260+试试效果
作者: sonicgoboy    时间: 2010-10-13 19:10
终于等到了,等了很久啊
作者: gzpony    时间: 2010-10-13 19:50
好消息。终于官方的来了。

正在用2010a
作者: pikaqiuuuu    时间: 2010-10-13 20:21
提示: 作者被禁止或删除 内容自动屏蔽
作者: 94117239    时间: 2010-10-13 20:27
我毕业论文就是用matlab做的
作者: zaknafein    时间: 2010-10-13 20:45
GPU computing with CUDA-enabled NVIDIA devices in Parallel Computing Toolbox
一千刀又没有了...
作者: dalao123    时间: 2010-10-13 20:52
这个不错啊,支持
作者: gzpony    时间: 2010-10-13 21:13
http://www.mathworks.com/help/toolbox/distcomp/bsic3by.html

GPU Computing
作者: ocuo1985    时间: 2010-10-13 22:42
同LS,毕业论文是用MATLAB,可惜到现在已经什么都不会了
作者: asdfjkl    时间: 2010-10-14 19:25
这东西不错,cuda加速极为实用。
作者: metapost    时间: 2010-10-14 19:56
正在下载,看着文档感觉用起来很方便的,不知性能如何
作者: crazystone    时间: 2010-10-14 20:05
很久没有用过matlab了
作者: 越光宝盒    时间: 2010-10-14 20:49
性能有几倍的提升啊?
作者: potomac    时间: 2010-10-14 21:54
提示: 作者被禁止或删除 内容自动屏蔽




欢迎光临 POPPUR爱换 (https://we.poppur.com/) Powered by Discuz! X3.4