Consumer Vs Professional – How To Select The Best Graphics Card

Transcript

Consumer vs Professional – How to Select the Best Graphics Card For Your Workflow Allen Bourgoyne Director, ISV Alliances, AMD Professional Graphics © 2012 Autodesk Learning Objectives At the end of this class, you will be able to:  Understand the modern workstation and how all the pieces contribute to overall application performance  Gain insight into graphics workloads and how they impact the specific features of the graphics card  Understand benchmarks, what they can and can’t measure, how to evaluate results and apply them to your specific workflow  Recognize the difference between consumer and professional level graphics cards and determine which is the right card for your needs © 2012 Autodesk Summary  In this module, we will explore the various components of the modern workstation and understand how they can impact both system and graphics card performance © 2012 Autodesk System Considerations © 2012 Autodesk Factors  When choosing a graphics card, important factors to consider:  What application(s) will I run?  How many simultaneous applications & windows will be running?  How many displays (monitors) will I use?  What is the resolution of the displays?  How large are the models & datasets?  What other components, CPU, memory, OS, etc., are in the system?  Can any of my applications use the graphics card for compute tasks?  Should I choose a consumer or a professional card? © 2012 Autodesk System Considerations  System configurations, other than the graphics card, can affect graphics performance:     CPU Memory PCI bus We will explore in more detail how each of these components can affect graphics performance © 2012 Autodesk System Considerations: CPU  CPUs feed the graphics card    Ideal situation is to feed the graphics card as much as it can handle  Too little, graphics card sits mostly idle  Too much, graphic card is overtaxed Modern CPUs are mostly multi core devices  4 cores per CPU is becoming common  CPUs must handle multiple tasks, multi-cores help with increasing system workloads 2 important CPU performance factors  How many cores  CPU speed (or frequency) © 2012 Autodesk System Considerations: CPU  Multiple cores enables CPUs to run multiple tasks concurrently    The more cores, the more tasks that can run at the same time – course grain parallelism Multiple cores enable single applications to perform multiple tasks concurrently  “Multi-threaded” or “parallelized” applications can realize significant performance gains by taking advantage of multiple CPU cores – fine grain parrallelism CPU speed will determine the rate at which the running task executes  Faster, more efficient CPU will execute the task more quickly © 2012 Autodesk System Considerations: CPU Systems have to run a lot of processes outside of any user programs If your CPU is overtaxed, all system performance will suffer including graphics Multi-core CPUs help with this ever increasing workload Use the Windows® Task Manager to monitor CPU utilization • Watch for peak CPU usage for one or more cores when you experience performance problems © 2012 Autodesk System Considerations: Memory  Model size and complexity affect system memory as well as graphics memory  Windows® provides virtual memory – hard disk space can hold data that exceeds physical memory  32 bit versions of Windows limited to 4GB maximum physical memory, 4GB virtual memory limit (total physical & virtual cannot exceed 4GB, 2**32 = 4GB)  64 bit versions of Windows limited to 128GB maximum physical memory, 16 terabytes virtual memory limit (totally physical and virtual cannot exceed 16 terabytes)  If physical memory is exceeded, OS will “swap” out data to virtual memory on disk to make more room   Best performance is achieved will all data fits into physical memory Virtual memory on disk significantly slower than system RAM © 2012 Autodesk System Considerations: PCI-E® Bus  Peripheral Component Interconnect Express: PCI-E®  Latest standard is PCI-E® Gen 3  PCI-E® Gen 1: ~8 GB/s transfer rate for x16 link  PCI-E® Gen 2: ~16 GB/s transfer rate for x16 link  PIC-E® Gen 3: ~32 GB/s for x16 link  Use PCI-E® Gen 3 systems and graphics cards!!!! © 2012 Autodesk System Considerations  Modern workstations are complex  There are other factors, such as I/O, network, etc., that can affect performance  There is no substitute for testing for performance in your environment, with your applications and your data sets to understand which parts of the system is stressed by your workload © 2012 Autodesk How to Determine Graphics Workload © 2012 Autodesk How to Determine Graphics Workload  The size and complexity of the graphics workload will determine overall graphics performance  Graphics workload includes both memory requirements and GPU work  Can’t focus on just one aspect! © 2012 Autodesk Understanding Graphics Memory Usage  Each application window requires graphics resources   Consume memory for geometry, graphics elements, shader programs, pixels, etc. Require graphics processing power to create & render graphics, regular refresh of the window  The more windows on the display, the more graphics resources and processing power required  Multiple displays will require more graphics resources and processing power © 2012 Autodesk Graphics Memory Usage: Multiple Displays  Graphics displays require memory to hold the graphics images that need to be displayed (frame buffer).  Formula: (display_width & display_height * bits_per_pixel) / 8 = total bytes  19” display: (1024 * 1280 * 32) / 8 = 5,242,880  22” display: (1600 * 1200 * 32) / 8 = 7,680,000  24” display: (1900 * 1200 * 32) / 8 = 9,120,000  30” display: (2560 * 1600 * 32) / 8 = 16,384,00 (This is the memory needed just to hold the pixel data itself, more is needed manage the actual display)  30” display requires > 3x the amount of frame buffer memory of 19”!  Dual monitor configurations require twice the frame buffer memory  32MB for dual 30” displays! © 2012 Autodesk Measuring Graphics Memory Usage  How can I tell how much graphics memory I am using?  AMD provides a tool that can help determine graphics memory usage   AMD FirePro Memory Viewer shows memory usage in real time Available for download at: http://www.amd.com/us/products/workstation/graphics/tools/Pages/tools.aspx GPU Tool shows graphics & system memory usage in real time CPU © 2012 Autodesk Understanding GPU Workload  The GPU workload is the amount of processing work the GPU needs to do to render your model  Includes items such as:  Number of triangles or polygons that need to be drawn  Applying colors, textures, and materials to surfaces  Rendering reflections, shadows, lighting effects  The GPU needs to be able to complete the workload fast enough to provide interactive updates  Somewhere between 30 to 60 frames per second will provide for smooth, interactive update rates that simulate real-time motion to the eye © 2012 Autodesk Understanding GPU Workload  Many applications will tell you how many triangles (polygons) for your model.   For example, Maya choose “Display” → “Heads Up Display” → “Poly Count”  Note: disk size of your data file not indicative of graphics memory usage Once you know the triangle size, you can determine how many times per second your graphics card can render this model:  graphics card max tri/sec / (number of triangles in model * desired frames per sec) = frames rendered per second  Example:  500 M tri/sec / (1M triangles * 30 frames per sec) = 16 frames/sec can be rendered by the graphics card © 2012 Autodesk Understanding GPU Workload  Understanding the graphics demands of your workload is key to selecting the right card    Too little memory, your models won’t fit! Too little GPU and your performance will be poor! Only you can determine the demands of your workflow! © 2012 Autodesk Graphics Benchmarks © 2012 Autodesk Benchmarks  Benchmarks only measure what they are designed to measure    Important to know the original design goals  What is being measured?  How is it being measured? They may have limited lifespan: age over time  New technologies may alter benchmark usefulness  When was the benchmark created?  What version of software was used to create the benchmark? Benchmarks designed for a particular version of an application may not be relevant for future versions  Later versions of software may contain different features & workflows  Does the benchmark test features that I use? © 2012 Autodesk Benchmarks  After the benchmark is complete, you need to analyze the results  Mistakes here can result in incorrect conclusions  Let’s take a look at some sample benchmarks to see how lack of careful analysis can result in incorrect conclusions © 2012 Autodesk Benchmark Example #1 Test A Test B 350 800 300 700 600 250 500 200 400 150 300 100 200 50 100 0 GFX Card 1 GFX Card 2 0 GFX Card 1 GFX Card 2 Test A is CPU limited, not a good test for graphics cards • Test B is a better indicator of graphics performance – if you relied on Test A to make a buying decision, you could make a mistake! © 2012 Autodesk Benchmark Example #2 GFX Composite 3 GFX subtest 6 2.75 5 2.5 4 2.25 2 3 1.75 2 1.5 1 1.25 1 0 GFX Card 1 GFX Card 2 GFX Card 1 GFX Card 2 Overall graphics composite shows about a 10% difference in the cards • If you examine the detailed benchmark results, you will see that most of the scores are similar, but one of the GFX tests shows a 40% difference! • Detailed examination of test results show most tests are CPU limited • You need to run test multiple times, varying CPU and GPUs to understand results © 2012 Autodesk Graphics Benchmarks  From the previous examples, you can see that there are more to benchmarks than just running a test and comparing a number  It’s easy to misinterpret results  Benchmarking takes a lot of work  In order to understand the relevance of the results, you have to run many tests, varying difference pieces of hardware in a systematic manner  Benchmarks can provide useful data – you need to analyze it carefully  There’s no substitute for testing on your workflow with your data! © 2012 Autodesk Consumer vs Professional Graphics Cards © 2012 Autodesk Professional Graphics Cards Professional graphics cards are designed for professional users and optimized for professional applications  Professional graphics cards are developed, tested, and tuned against professional applications  Professional software developers, like Autodesk, develop on, test with, and support professional graphics cards © 2012 Autodesk Professional vs Consumer Graphics Cards  A number of people choose consumer graphics cards over professional cards when using workstation class applications  Several factors contribute to this decision:  Consumer cards have a lower price  Perception is that there is no difference between consumer and professional cards with respect to performance and features  A consumer card came with my system  That’s what I was told to buy  I got a free one! © 2012 Autodesk Professional vs Consumer Graphics Cards  The reality is that professional cards are the correct choice for professional users and for professional applications      Consumer cards have a lower price  There are professional cards to fit every budget! Perception is that there is no difference between consumer and professional cards  Professional cards, drivers, and applications are high tuned to deliver performance, reliability, and stability beyond consumer cards A consumer card came with my system  Fortunately, that can be remedied! That’s what I was told to buy  Check your sources! I got a free one!  Congratulations, great card to use at home! © 2012 Autodesk What do I get for the extra $$$’s? Benefit Professional Graphics Consumer Graphics Drivers tuned, tested, & optimized for professional applications by AMD Professional applications certified and tested by Autodesk Direct engineering efforts between Autodesk and AMD Supported by dedicated trained support team Average 2-3 year availability 3 year warranty © 2012 Autodesk What about performance? Let’s compare performance of a $369* consumer card vs $120* professional card: Cadalyst 2011 3D Graphics 1400 Viewperf 11 composite 25 1200 20 1000 15 800 600 10 400 5 200 0 0 FirePro V3900 Radeon HD 6970 FirePro V3900 Radeon HD 6970 Professional card offers almost 2x the performance at 1/3rd the cost! *Original MSRP for FirePro™ V3900 and Radeon™ HD 6970 Cadalyst 2011 benchmark, SPEC Viewperf 11 benchmark. System config: Intel Xeon 6-core 1xW3680 @ 3.33 GHz, 16GB RAM, Windows 7 64-Bit SP1; FirePro 8.88 driver, Radeon Catalyst 11.8 © 2012 Autodesk Conclusion  Modern workstations are complex   Before you can analyze graphics performance, you need to understand the demands your specific workload puts on the workstation   Need to understand what parts of the workload stress graphics Benchmarks can be useful tools to help determine which card to buy   Need to understand all of the pieces in order to understand graphics performance Need to understand the benchmark and analyze results carefully Use professional graphics cards for professional applications  Professional graphics cards have the performance, stability, and reliability professionals demand – and priced to fit every budget © 2012 Autodesk Autodesk, AutoCAD* [*if/when mentioned in the pertinent material, followed by an alphabetical list of all other trademarks mentioned in the material] are registered trademarks or trademarks of Autodesk, Inc., and/or its subsidiaries and/or affiliates in the USA and/or other countries. All other brand names, product names, or trademarks belong to their respective holders. Autodesk reserves the right to alter product and services offerings, and specifications and pricing at any time without notice, and is not responsible for typographical or graphical errors that may appear in this document. © 2012 Autodesk, Inc. All rights reserved. © 2012 Autodesk