State of CAD and Engineering Workstation Technologies
Abbreviations
CAD is Computer Aided Design
CAE is Computer Aided Engineering
CEW is Computer aided design and Engineering Workstation
CPU is Central Processing Unit
GPU is Graphics Processing Unit
Hardware for CPU-Intensive Applications
Computer hardware was designed to support programs and it is a typical but simplistic view that higher spec hardware will enable all computer programs to do better. Up to recently, the CPU was indeed really the only device for computation of computer programs. Other processors embedded in a PC or workstation were dedicated to their parent devices for instance a graphics adapter card for display, a TCP-offloading card for network interfacing, along with a RAID algorithm chip for hard disk redundancy or capacity extension. However, the CPU is not the only processor for software computation. We are going to explain this in the next section.
Legacy programs still rely on the CPU to do computation. That is certainly, the regular view is valid for applications that have not taken advantage of other kinds of processors for computation. We’ve got done some benchmarking and feel that applications like Maya 03 are CPU intensive.
For CPU-intensive applications to execute faster, the rule is to have the highest CPU frequency, more CPU cores, more main memory, and possibly ECC memory (see below).
Legacy software had not been meant to be parallel processed. Therefore we shall check carefully with the software vendor about this issue before expecting multiple-core CPUs to create higher performance. Irrespectively, we’ll achieve a higher output from executing multiple incidences of the application but this is not just like multi-threading of a single application.
ECC is Error Code Detection and Correction. A memory module transmits in words of 64 bits. ECC memory modules have incorporated electronic circuits to detect just one bit error and correct it, but aren’t capable to rectify two items of error happening in the same word. Non-ECC memory modules do not check whatsoever – it is constantly work unless a lttle bit error violates pre-defined rules for processing. How often do single bit errors occur nowadays? How damaging would an individual bit error be? Let’s see this quotation from Wikipedia in May 2011, “Recent tests give widely varying error rates with 7 orders of magnitude difference, ranging from 10-10-10-17 errors/bit-hour, roughly one bit error hourly per gigabyte of memory to 1 bit error per century per gigabyte of memory.”
Hardware for GPU-Intensive Applications
The GPU has now been developed to realize the prefix of GP for General Purpose. To become exact, GPGPU represents General Purpose computation on Graphics Processing Units. A GPU has many cores you can use to accelerate an array of applications. As outlined by GPGPU.org, the industry central resource of GPGPU information and facts, developers who port their applications to GPU often achieve speedups of orders of magnitude in comparison to optimized CPU implementations.
Many software applications happen to be updated to exploit the newfound potentials of GPU. CATIA 03, Ensight 04 and Solidworks 02 are samples of such applications. Because of this, these applications are a lot more understanding of GPU resources than CPU. That is certainly, to own such applications optimally, we ought to spend money on GPU in lieu of CPU for the CEW. Based on a unique website, the newest Abaqus product suite from SIMULIA – a Dassault Systemes brand – leverages GPU to perform CAE simulations two times as fast as traditional CPU.
Nvidia has released 6 member cards in the new Quadro Fermi family by April 2011, in ascending sequence of power and cost: 400, 600, 2000, 4000, 5000 and 6000. In accordance with Nvidia, Fermi delivers up to 6 times the performance in tessellation of the previous family called Quadro FX. We shall equip our CEW with Fermi to accomplish optimum price/performance combinations.
The potential contribution from the GPU to performance depends on another issue: CUDA compliance.
State of CUDA Developments
As outlined by Wikipedia, CUDA (Compute Unified Device Architecture) is really a parallel computing architecture produced by Nvidia. CUDA may be the computing engine in Nvidia GPU offered to software developers through variants of industry-standard programming languages. For instance, programmers use C for CUDA (C with Nvidia extensions and certain restrictions) compiled by way of a PathScale Open64 C compiler to code algorithms for execution for the GPU. (The most recent stable version is 3.2 released in September 2010 to software developers.)
The GPGPU website has a preview of your interview with John Humphrey of EM Photonics, a pioneer in GPU computing and developer of the CUDA-accelerated linear algebra library. The following is an extract from the preview: “CUDA provides for very direct expression of exactly how you want the GPU to do a given unit of labor. Ten years ago I was doing FPGA work, the place that the great promise was the automated conversion of advanced languages to hardware logic. Needless to say, the large abstraction meant the actual result wasn’t good.”
Quadro Fermi family has implemented CUDA 2.1 whereas Quadro FX implemented CUDA 1.3. The newer version has provided features that are significantly richer. By way of example, Quadro FX did not support “floating point atomic additions on 32-bit words in shared memory” whereas Fermi does. Other notable improvements are:
Up to 512 CUDA cores and 3.0 billion transistors
Nvidia Parallel DataCache technology
Nvidia GigaThread engine
ECC memory support
Native support for Visual Studio
State of Computer Hardware Developments
Abbreviations
HDD is tough Disk Drive
SATA is Serial AT Attachment
SAS is Serial Attached SCSI
SSD is Solid State Disk
RAID is Redundant Array of Inexpensive Disks
NAND is memory according to “Not AND” gate algorithm
Bulk storage is an essential a part of a CEW for processing instantly and archiving for later retrieval. Pushes with SATA interface increasingly becoming bigger kept in storage size and cheaper in hardware cost with time, although not getting faster in performance or smaller in physical size. To obtain faster and smaller, we need to select hard disks with SAS interfaces, using a major compromise on storage size and hardware price.
RAID has existed for years for providing redundancy, expanding how big is volume to well beyond the confines of one physical harddrive, and expediting the velocity of sequential reading and writing, in particular random writing. You can deploy SAS RAID to cope with the massive storage size issue nevertheless the hardware price will go up further.
SSD has resulted in recently being a bright star beingshown to people there. It has not replaced HDD due to the high price, limitations of NAND memory for longevity, and immaturity of controller technology. However, they have found a place recently as being a RAID Cache for two main important benefits not achievable along with other means. The very first is a greater speed of random read. The second is economical point when employed in addition to SATA HDD.
Intel has released Sandy Bridge CPU and chipsets which can be stable and bug free since March 2011. System computation performance is over 20% more than the last generation called Westmere. The top CPU model has 4 editions that are officially capable of over-clocking to around 4GHz as long as the CPU power consumption is within the designed limit for thermal consideration, called TDP (Thermal Design Power). The 6-core edition with official over-clocking should come out in June 2011 timeframe.
CurrentState & Long run
Semiconductor manufacturing technology has improved to 22 x 10-9 metres this year 2011and is heading towards 18 nanometres in 2012. Smaller means more: we are going to read more cores plus much more power from your new CPU or GPU made on advancing nanotechnology. The current laboratory probe limit is 10-18and this sets the headroom for semiconductor technologists.
While GPU and CUDA are receiving big impacts on performance computing, the dominant CPU manufacturers aren’t resting on their laurels. They’ve got started to integrate their particular GPU in to the CPU. However, the amount of integration is really a long way away from the CUDA world and integrated GPU will not likely displace CUDA for design and engineering computing sometime soon. What this means is our current practice as described above will stay the prevailing format for accelerating CAD, CAE and CEW.
END
Related posts:
