In the early days of PC-based CAD, we knew that high-performance graphics cards and large-screen monitors were required to run CAD software effectively. Desktop computers of the day came standard with low-resolution (320x200 or 640x350), few-color (4 or 16) displays. But with 20-inch tube monitors (running 800x600- or 1024x768-resolution 256-color graphics) weighing in at 75 pounds and costing $4000 or more, the solution wasn’t cheap -- or fun to lift. We nevertheless took the plunge to attain the productivity we needed.
Fast forward to today, and 21-inch flat screen monitors are a throwaway commodity. But what about the graphics cards: are high performance ones still necessary?
On my job as a CAD manager, I see a variety of inexpensive gaming cards used in computers used for CAD work, and sometimes even no special graphics card at all (through the use of embedded graphics that are bundled with the CPU). There is little thought put towards a key component of the CAD workstation. I’ve observed firsthand problems related to inadequate graphics cards, including slow performance, driver upkeep issues, and lack of support for key 3D display-based commands in CAD software.
It seems logical to me that specifying the proper professional-level graphics hardware for CAD workstations would rid us of these problems and so make the lives of CAD users more productive and pleasant. The question then becomes: How do I educate myself on how to select the right graphics hardware, and convince clients that it is worth their time and money to do so?
To answer these questions, I interviewed Adam Scraba, NVIDIA Senior Product Manager, and Sean Kilbride, NVIDIA Technical Marketing Manager. I wanted to get their take on how professional graphics (specifically their Quadro product line; see Figure 1) might help CAD and IT managers get the most from the CAD software they support.
Q: Let’s start at the very beginning by asking what is the difference between a professional graphics processor card, and the typical consumer/gaming card we see in desktop PCs today?
A: The key differences in professional graphics (as compared to gaming graphics) are the hardware itself, and the presence of features that support deployment into professional environments.
Q: OK, then let’s first dig into the hardware. Your Quadro boards are the professional series, and the GeForce boards the consumer ones. What’s the difference between them?
A: Quadro and GeForce, at the most basic level, share a common set of graphics processing units (GPUs) and rough board design; however, there are significant differences. The Quadro line is a carefully curated collection of hardware that is tuned and supported for deployment into an office environment, and is manufactured by NVIDIA.
In contrast, GeForce is our chip technology that is licensed by other board manufacturers.
Q: So Quadro could be described as more robust, or a solution customized for professional production environments?
Q: From a hardware point of view, what do Quadro boards provide that consumer-level products don’t?
A: First, you can qualify a Quadro card, and then scale up your workstations over multiple years without worry. A given Quadro card is the same card regardless of where or how you purchase it, because it is made by NVIDIA to a single specification.
Contrast this with gaming cards, each of which is offered by multiple hardware vendors with differing specifications and configurations. Each gaming card often has well over ten versions of it on the market, making it difficult to know which version to qualify.
Second, Quadro offers extended availability and support, and a warrantee that lasts three years. Gaming card lifecycle, support, and warrantees are much shorter and vary by the board manufacturer.
[Ron Green comments: From my point of view as a CAD manager, I find it much easier to have all CAD workstations configured the same, to reduce support burdens. The fact that I can purchase a single graphics card that runs consistent drivers saves me time – especially as software is updated annually.]
Q: You mentioned that there are features that help CAD and IT managers deploy Quadro cards. Can you elaborate on that?
A: This is primarily through driver management.
Drivers are how applications harness the acceleration power of the GPU. Quadro drivers and hardware are certified & tested to be compatible with and flawlessly accelerate over 100 of the most-used professional applications. For example, when Dassault Systemes and NVIDIA say there is a certified Quadro board and driver for SolidWorks 2014, we mean that the board and driver combination has been tested to work flawlessly on 100% of SolidWorks’ features and views. It also means that the board-driver combo is supported by both NVIDIA and SolidWorks, and so any issues reported by users will be fixed with a high priority by both parties.
Secondly, when companies deploy and qualify graphics into their IT infrastructure, they want to qualify the hardware-driver combination. As well, they want the ability to update drivers with incremental improvements -- while minimizing the risk of new driver additions that ‘break’ some functionality. This requires the maintenance of “stable” drivers that incorporate small incremental bug fixes and feature/performance optimizations through the idea of small “dot releases.” NVIDIA offers these under the name of Optimized Drivers for the Enterprise (ODE) at www.nvidia.com/drivers.
Q: With respect to software certification, is this included in the warrantee?
A: While certifications and warranties are ways in which NVIDIA stands behind its professional solutions, one is not connected to the other.
Certifications of Quadro products for professional software are a result of NVIDIA’s close work with software ISVs (independent software vendors) to guarantee that our boards -- in combination with a certified Quadro driver -- pass quality assurance tests of both NVIDIA and the ISV. Certification provides users peace of mind in knowing that they can focus on getting work done, rather than solving hardware and software compatibility problems.
I would point out that while Quadro cards have a three-year warranty, in most cases older graphics cards continue to be certified for many years beyond this time. In fact, many Quadro cards which are two to three generations old continue to be certified with professional applications.
Q: So rather than me fighting bugs or finding the right drivers for a consumer graphics board driver, I let you and the software companies worry about it, right?
Q: How long does it take for certified drivers to be available for new software releases?
A: NVIDIA works with closely with software partners to assure that certified drivers are available as close to the release of new software as possible. In many cases, certified drivers will be available at launch or immediately afterwards. In some cases the availability of certified drivers are dependent on the ISV’s certification schedule.
Q: I’ve found that simply not having to mess with a graphics board or experiment with drivers saves me a lot of time over the life of the workstation. Do you have any ROI [return on investment] metrics or tools that help users calculate the potential savings?
A: IT managers and CAD engineers often discount or ignore completely how much their time is worth when considering the value of hardware uptime. Let’s say you are a CAD engineer making $60,000 per year working on CAD modeling for half of your eight-hour day. A meager 10% time savings – from either the 3D CAD workflow being faster, or a reduction in system downtime due to certified drivers, or a combination of both – is worth $3,000 per year.
Q: I’d like to point out the savings in reduced CAD manager and IT staff time due to lack of trouble shooting. This works in much the same way, as our labor rates are similar to that of engineering time.
A: Good point.
Q: I find that having exactly the same hardware on the users’ desks keeps maintenance costs down. So, my question is: If over the course of several months I buy several of a certain model of NVIDIA card, can I be assured all are, in fact, the same?
A: The short answer is yes. Let’s take the Quadro K5000 as the example. Quadro offers an availability and stability advantage over gaming cards. The same Quadro board you purchased two years ago will still be available today. Since workstation hardware is designed to be long-life products, Quadro maintains extended availability. Standard availability is three years or greater. Additionally since we maintain the same specifications, a Quadro K5000 purchased one year ago is identical to the one you purchase today, regardless of geographic location. In contrast, gaming boards tend to exist in the market with many variations of components, cooling solutions, size, and so on.
Q: From a specification point of view, I find the perception is that more megabytes of RAM on the graphics board equates to a better board. Is the reality that the RAM simply supports the GPU’s bandwidth?
A: Not necessarily. While both bandwidth and amount of RAM are important aspects of a graphics card, the amount of RAM is the most important factor in determining how large a CAD model you’ll be able to fit into memory on the graphics card. Loading the entire model into graphics memory helps maximize performance by eliminating the need to swap data between system memory and graphics memory.
Quadro cards are designed with memory configurations ranging from 1GB all the way up to 12GB on the high-end Quadro K6000. The memory for each type of Quadro card is optimally configured based on the types of workflows commonly used by the majority of professional users.
Q: So bigger models tend to point towards graphics boards with more RAM then?
Q: Could you briefly explain the benchmarks you use to ascribe graphics performance? I know I don’t always know how to interpret the benchmark data I see.
A: Ideally, the best benchmark is how a board performs for your particular workload. However, users have particular needs in terms of the types and sizes of models they work with, model complexity, and required features -- so no single benchmark is going to truly be representative of any particular user’s workflow.
That said, benchmarks provide the ability to make some general comparisons using some relatively standard usage models. Benchmarks should use the actual software and real-world models, like the SPECapc benchmarks developed by the SPEC group. Other “synthetic” benchmarks, such as SPEC Viewperf 12, use application traces to simulate user workflow. It can be used as a general benchmark for comparing boards, but may not represent the real software or models that represent your specific workflow.
[Robert Green comments: CAD managers may find the SPEC page an interesting read, if only to get an idea of how to construct in house benchmark tests for other software programs. See http://www.spec.org/benchmarks.html#gwpg.]
Q: Let's get beyond the buzzwords to determine the level of graphics most appropriate for the following types of users:
- AutoCAD user primarily working in 2D
- Regular Solidworks user working with 3D parts and assemblies
- Advanced modeling user doing renderings
Assume an i7 or Xeon quad core processor, with Windows 7 64-bit operating system as a base platform.
A: This question is asked of us frequently. While actual workflow and usage of everyone is quite different, we can make some general recommendations as starting points:
- For CAD users working with AutoCAD, we recommend either a Quadro K600 for basic CAD work or a K2000 for a more advanced use
- For Solidworks users, the recommendation is a K2000 for small- to medium-size parts and assemblies, and a K4000 for users working with large assemblies.
- Once users begin working with GPU-accelerated rendering applications, they are going to want a more powerful high-end GPU, like the K5000.
Q: Is the processor and memory selection of the workstation a significant consideration for a given graphics card?
A: While there isn’t a specific relationship between a graphics card and a workstation’s CPU or memory, it’s best to think of a workstation as a complete solution. As such, the operating effectiveness of a workstation is only be as strong as its weakest component. It makes little sense to put a high-end CPU into a system and then bottleneck its performance with too small a GPU.
Similarly, if the graphics system is constantly waiting on a weak host CPU to handle application processing tasks, it won’t be able to deliver its full performance.
[Robert Green comments: In addition to RAM, be sure to consider the types of monitors (and how many of them) you need to connect to the graphics card, the board’s power consumption, and the slots available in your workstations. Some boards need extra power or two slots. The table in Figure 2 summarizes these requirements for the Quadro line of graphics cards.]
Q: We all know that software changes. How does this impact my graphics card selection? Is there a formula, such as “buy one higher model than you need today” that staves off obsolescence?
A: There is, unfortunately, no simple formula, because everyone’s needs and workflows are different. This makes specific statements difficult, but we can make some general recommendations.
One common trend is that workloads tend to get more complex over time. Either you’re working with larger models than you previously did, or you’re being asked to submit more design iterations on a project. Both mean that being able to increase your productivity is vitally important.
I recommend that CAD managers take a look at how workloads have increased in complexity over the one to two years, and then use this as a guideline as to the type of workload increases they can expect to see in the near future.
Q: So the main thing that CAD or IT managers can do is to really know their software well, to talk to management teams, and explain that new workstations should be specified for projected future software usage -- as opposed to current needs? I hope your answer is yes, because I’ve been preaching this for years.
A: Yes. With the longer timeframe in mind, they can decide which professional graphics solution best fits not just their current workloads, but also their future workloads as well.
One additional thing to keep in mind is that newer versions of professional applications by and large are taking greater advantage of GPUs than before. In the past, a GPU may have been used only for viewport 3D displays. Applications that use a GPU for rendering and/or simulation may require more GPU horsepower than previous versions of the same application, even though the workflow may seem the same.
If you’re a CAD or IT manager responsible for specifying workstations and their components, I hope this discussion has opened your eyes to some of the reasons why a bargain graphics card isn’t always a bargain.
When you take all the variables into account over the three-year (or maybe more) lifespan of your CAD workstations, it becomes apparent that a robust and well supported professional graphics card can be seen as an investment in uptime and productivity.