A little while back I introduced the NVIDIA 5000M graphics card to you (housed in the Hewlett Packard EliteBook 8740w mobile workstation), and gave some background for the hardware and software that makes that awesome little package go. My question was, “Can I break the little bugger?” Well, I’m a bit sad to say, no.
Why sad? I can usually bog a graphics card’s output with large selections, simulation, and rendering. I’d love nothing more than to say I broke it, and exactly where the breaking point was. Unfortunately, as of this time, I haven’t reached the Quardo 5000M’s limits. (I’m still hopeful though).
Comparisons and Datasets
In order to explain how well the Quadro performed, I’ll later refer to the SPECViewperf benchmark. Unfortunately, that score does little to convey a real feeling of what can be expected while using the graphics card. Therefore, I will occasionally reference the NVIDIA Quadro FX 2700M as a unit of comparison. In this way, more of you can get a tangible feel for what kind of improvements we are talking about. The Quadro FX 2700M is hosted by a Dell Precision M6400.
Throughout the testing process, I modeled various components to evaluate the stability and general feeling of how the Quadro 5000M would behave during normal everyday activities. I was also furnished with some more complex datasets, such as the multi-thousand component Industrial Machinery Mechatronics model, and the RC car model, which contained just shy of 300 components. These were the test subjects of numerous Stress Analysis (FEA), Simulation, and Rendering workflows.
First let me say that the overall feel of the graphics was smooth.
- Anti-Aliasing was sharp
- Renders were quite beautiful, and colors were rich and deep
- Translucent objects were displayed well
- Selection Previews were snappy, with sharp edges and vivid colors
- Meshes were sharply defined
- General component movement, manipulation, and zoom functionality was smooth
In terms of modeling performance, I couldn’t identify an area where the Quadro 5000M was lacking when compared with any other graphics card I have experienced. While a graphics card isn’t going to fix that slow CPU or limited RAM, it is interesting to note that the comparison Dell Mobile Workstation was equipped with a faster CPU than that of our subject HP 8740w. The NVIDIA Quadro 5000M still outperformed the comparison machine.
In short, there was nothing in the performance of the Quadro 5000M that would be considered a let down, and it was definitely a pleasure to use.
Autodesk Inventor 2011
One thing that has always plagued graphics cards has been selection preview updates. The waiting is hell. The more components piled into an assembly, the slower the visual cues become. The Quadro 5000M’s reaction time was fast. Selecting the entire 300 part RC car assembly resulted in a slight pause that was less than a second. The multi-thousand part factory assembly took a bit more time, but was still very fast. When compared to the FX 2700M, the selection highlights were a bit crisper, and somewhat faster, especially with the larger assemblies.
Renders went quite well. The image to the right contains the factory sample dataset, with over a thousand parts. It was rendered in Inventor Studio at 1024×768, with good anti-aliasing. It took 2 minutes, 30 seconds to complete.
Inventor Simulation and Analysis
The NVIDIA Fermi technology pumps some of the Finite Entity Analysis (FEA) calculations to the GPU, subsequently the CPU is freed up do more of the general instructions. While I have no clear method to measure these values, I ran the same simulations on both machines and came up with very nice results.
The Stress Analysis results shown to the right are the displacement of the Industrial Machinery base after placing 2000 pounds on it. The result times were quite satisfying. The colors were vibrant and edges were well defined. Navigating from one result page to the next was quite fast.
The Zoom Test
Another set of tests that generally gets the GPU to bog down begin by adjusting the view angle such that multiple translucent components are between the viewer and a very tightly meshed FEA subject. Add slice graphics to the view and use the Steering Wheel zoom function. Then while keeping both mesh and translucent edges overlapped in the view, zoom in at a consistent speed.
I am looking for two factors: How hesitant the zoom is, and how far I can zoom.
The Comparison Quadro FX 2700M was a bit edgy with only Inventor running (and this document in the background). Nothing bad by any means, just occasional, barely noticeable pauses. The zoom distance went to the ‘Steering Wheel Pivot Point’ but no farther. I tried numerous methods to get past this, but I could not.
The Quadro 5000M however was a slightly different story. I had Inventor running with Fusion (and a download) in the background. The zoom smoothness was about 100% better than the comparison Quadro. Here’s the kicker: the ‘Steering Wheel’ zoom went past the Pivot Point, through the wall of the part, up to and beyond the next wall of the part, and kept going….and going.
I repeated the process, but stopped the zoom just short of the second wall, and orbited the view to look back where I came. Smooth as greased owl poop. That was amazing.
Autodesk Inventor Publisher 2011
Those that use Inventor Publisher know that it is tremendously dependent on RAM and processing power. People are still finding it hard to nail down just how much hardware is required to make Publisher run smoothly.
The Publisher workspace responded fairly well. Considering how much of a resource hog that the application is, I’d judge the Quadro 5000M and Publisher as a great match. The selection previews were good, but not really any crisper or faster than the comparison FX 2700M. The colors however were warmer and slightly deeper in comparison, when the Autodesk Materials Library was referenced. The remainder of the component manipulation was reasonably smooth.
The real improvement was in the Storyboard playback. That was slick; So much so that I had to run it again and again just to be sure it wasn’t a glitch. If you have ever used this feature, you know how choppy it can become. I tested the RC car with Auto-Explode feature on both one-level, and all-levels. Using the auto-assign snapshots method, I created a fast storyboard of every component move in the model. I played it backwards and forwards, at slow speeds and fast. This never became choppy, and was quite impressive.
Autodesk Fusion and Algor Simulation
Fusion behaved wonderfully. Occasionally, Fusion has behaved a bit edgy with various hardware combinations I have experienced, like it needed a bit more video ‘oomph’. I found none of this as I added features to this pinion gear set. Everything from modeling to visual styles went very well.
Algor performed well, with no video related issues. The results speed was fair, considering the slow CPU speed. Analysis display output was very responsive as I altered result display options from one to another. No delays and everything was sharp.
Multiple applications went without a hitch. There were no noticeable artifacts or conflicts when moving from one application to another. I ran Inventor, Publisher, and Alias simultaneously with no issues whatsoever. Additionally, Inventor, Fusion, and Algor were tested together. I routinely left a model in simulation, and just shut the lid on the laptop, something that I’d never do on other workstations. I returned to it later in the day, and after the system woke up, the simulation display updated between result pages as if it had never gone into standby. This procedure is not exactly a definitive test, but does give a feel for how well the Quadro 5000M encapsulates the video instructions.
Soon I will include more information about how NVIDIA handles the separate video output of each application and discuss the NVIDIA nView Desktop Display Manager, which I have come to find out is a pretty cool tool.
SPECviewperf 11 is the newest release of the well known benchmark tool that measures performance of the graphics cards in numerous areas, and then delivers a composite score. The results of the test ran on the HP 8745W containing the NVIDIA Quadro 5000M are as follows:
Catia-03: 32.83 Ensight-04: 30.11 Lightwave-01: 46.89 Maya-03: 69.27 ProE-05: 8.19
Solidworks-02: 44.03 Siemens Teamcenter Vis-02: 31.19 Siemens NX-01: 30.94
The NVIDIA Quadro 5000M is a mobile workstation dream. It performed beautifully and I could not find fault with its performance in any of the areas that I investigated. The largest improvements I can put my finger on were simulation and component movement. In these areas, the Quadro 5000M seemed to hold the greatest edge over the comparison graphics card. Additionally, there were no instances of driver conflict or problems with the Windows 7 operating system.
The largest hurdle anyone will face is the cost of this Ultra-High end graphics card, and higher end mobile workstations that it is packaged in. However, when considering the time lost waiting, and often recovering from crashes resulting from CPU overloads while simulating large numbers of components simultaneously, the cost begins to look like a positive investment.
From a purely performance standpoint, I don’t think you will be finding another mobile graphics solution to make a similar comparison with.
Check back for the performance review of the Quadro 5000M while using AutoCAD Civil 3D 2011, and various model forms including Corridors, Surfaces, Revit Buildings and Pont Cloud datasets. I have a few tricks up my sleeve, and the next round should give the mesh heavy AEC crowd something useful of compare.