Quadro M4000—CAD Applications Can't Handle This Much Power
Tom Lansford posted on February 08, 2016 |

When I tested the Quadro M4000 on interactive 3D modeling, the GPU was almost idling at a 10 percent to 20 percent workload, even on large models, while the CPU was ready to blow a gasket. But when it comes to design visualization or immersive design reviews, the Quadro M4000 is a perfect match with the graphics power of 1664 cores, 8 GB of memory, a 3D stereo connector and four 4K DisplayPort outputs. 

The Quadro M4000’s connectors for multiple display synchronization, SLI rendering and 3D stereo all indicate the high-end customer that this GPU targets.
The Quadro M4000’s connectors for multiple display synchronization, SLI rendering and 3D stereo all indicate the high-end customer that this GPU targets.

Day in and day out, CAD modeling on mainstream systems simply can't use this much graphics power. CAD software vendors put their effort into extending the list of useful editing tools, but rarely concern themselves with 3D graphics performance optimization as long as the workstation is “fast enough.” This leads to 3D performance in the design application that is far below optimal performance. Testing shows that the application delivers slow graphics performance, and yet the graphics workload is far below 25 percent of the GPU's capability.


The Quadro M4000 Upgrades the High-End GPU Slot to NVIDIA’s Maxwell GPU Architecture

The NVIDIA Quadro M4000 uses NVIDIA’s Maxwell GPU architecture. The products launched at Siggraph 2015 and showed up in new Intel Skylake workstations this past fall. The Quadro M4000 replaces the Quadro K4200. 

Along with 8 GB of GDDR5, four DisplayPort outputs, 4K monitor resolution support, NVIDIA SLI technology, stereo and sync support, this is a GPU with power and potential that comes in at a price around $1,000.

This table shows the key updated features of the Quadro M4000.

Quadro M4000

Quadro K4200

GPU Architecture

Maxwell (2)

Kepler

NVIDIA CUDA Processing Cores

1664

1344

GPU Memory

8 GB

4 GB

Memory Bandwidth

192 Gbps

173 Gbps

DisplayPort Outputs

4

2

4K Displays Supported (60 Hz)

4

2

Maximum Power Consumption

120 W

105 W

I tested the GPU in a new Skylake workstation with the Intel Xeon E3-1275 v5 clipping along at 3.6 GHz (you’ll find the performance numbers further down). First, let’s start with a look at the outside of this new professional GPU.

The Quadro M4000 doubles the graphics memory to 8 GB. This adds more space for on-board model data and for GPU computing data. The new graphics architecture also enables support for four 4K displays. That is a great feature when you want one graphics card to power a four-surface CAVE installation.

The Quadro M4000 has a total power consumption of 120 W and requires a six-pin connector for the back of the board. Yet, it remains a single-slot design. Three connectors on the top edge of the Quadro M4000 give away the target user. Here, you find connectors for SLI support, 3D stereo and multiple display synchronization. These are high-end professional features needed for CAVEs, virtual reality (VR) and complex display systems.

On the inside, the new Maxwell 2 architecture delivers more performance per watt, which means much higher performance than the Quadro K4200. In fact, the Quadro M4000 approaches the performance levels of the previous generation’s Quadro K5200.

If you are familiar with the SPEC.org benchmark, SPECviewperf 12, then you will know that the results are a weighted frames-per-second (FPS) value for each of the data sets. SPECviewperf is designed to isolate the GPU performance from the rest of the workstations and to load the GPU as much as possible. It is clear from the results below that a Quadro M4000 won't be the choice for simple or even moderately complex 3D work. It is too fast, and as we see in our application testing, the modeling application cannot keep up.

There are applications, however, that mirror SPECViewperf 12 testing. Design visualization and VR applications are designed to keep the GPU running at peak capacity. Applications such as DELTAGEN and Bunkspeed from Dassault Systèmes and VRED and Showcase from Autodesk are examples. This class of visualization applications has extremely efficient 3D engines and less CPU overhead than a full-fledged CAD application. They are able to feed the graphics card with 3D data and render highly realistic images. These applications also support 3D stereo displays, which doubles the graphics rendering requirements.  

SPECViewperf 12

Graphics Accelerator

NVIDIA Quadro M4000

Graphics Driver Version

10.18.13.5413

Display Resolution

3840 x 2160 @ 32 bpp

Viewset

Composite

Window

catia-04

85.02

1900 x 1060

creo-01

72.44

1900 x 1060

energy-01

5.74

1900 x 1060

maya-04

69.97

1900 x 1060

medical-01

25.25

1900 x 1060

showcase-01

42.96

1900 x 1060

snx-02

75.50

1900 x 1060

sw-03

111.26

1900 x 1060

NVIDIA is pushing the envelope of GPU possibilities with software tools that help developers build faster applications. NVIDIA helps them leverage the power of the GPU. Along with the launch of the Quadro M4000, NVIDIA announced DesignWorks VR for immersive VR rendering environments. The company also provides Optix software to enable GPU-accelerated ray-traced rendering. Both are available to developers as SDKs and deliver new possibilities in real-time, realistic and immersive rendering.


The Quadro M4000 Has Far Too Much Graphics Power for 3D CAD and Modeling Applications

Unfortunately, they are not built on high-performance 3D engines. 

The Quadro M4000’s 1664 cores, 8 GB of memory and four outputs for 4K displays spell overkill for 3D CAD modeling.
The Quadro M4000’s 1664 cores, 8 GB of memory and four outputs for 4K displays spell overkill for 3D CAD modeling.

CAD applications are designed to create and manage 3D models. Unfortunately, they are not built to optimize performance between the CPU and GPU. Each time you interact with your CAD program, there is the time, X, which the CPU needs in order to prepare the new view, and the time, Y, which the GPU needs to render that information on your display. If Y is essentially zero relative to X, then a faster GPU (pushing Y further towards zero) doesn't deliver any additional performance. X is what needs improvement.

This is exactly the case for the Quadro M4000 when running applications such as Autodesk Inventor and 3ds Max. The result is fine, but it’s not spectacular 3D performance when you interact with large models.

For interactive CAD tests, I used a Ferrari model. I tested with a single car, 30 cars, 50 cars and then 100 cars. For each test, I measured the average FPS, the workload on the CPU and the workload on the GPU.

Test

FPS mean

GPU mean

CPU mean

3ds Max 1 Ferrari

125

25

50

3ds Max 10 Ferrari

12

13

96.5

3ds Max 50 Ferrari

9

20

92.5

3ds Max 100 Ferrari

4

20

92.5

With the simplest model and 125 FPS, the Quadro M4000 is only reaching 25 percent of its maximum capacity. The 30-car model shows how the FPS value drops 10-fold, yet the GPU workload goes down, not up. In all cases the GPU is rendering in zero seconds relative to my interaction with 3ds Max. The same effect is seen in Autodesk Inventor. Unfortunately, Inventor does not display FPS statistics.

Testing with a slower graphics card delivered exactly the same FPS values. This makes two things clear: The workstation is never waiting on the graphics card, and the Quadro M4000 has more power than the 3D modeling applications and the workstation hardware can handle.


When Do You Need the Power of a Quadro M4000?

I like fast graphics, and the Quadro M4000 qualifies. The upgrades from the Quadro K4200 are welcomed. The power of the GPU, the 8 GB of graphics memory, the four DisplayPort outputs and the specialized connectors make the Quadro M4000 the right choice for demanding jobs: interactive visualization, stereo displays, immersive CAVE systems and VR headsets. 3D CAD modeling doesn't need this kind of power without first acquiring much better CPU hardware. This GPU is destined for more demanding applications and hardware configurations.

When you specify the details for your next design review workstation, you can build a real immersive VR and visualization system with the Quadro M4000. And when you define your CAD and modeling workstations, you can save your budget and use a less powerful GPU.

So let's hear from you. Tell me in the comments which GPU you use for CAD and which GPU you use for design visualization. Will I be surprised?

About the Author







Tom Lansford is an international marketing consultant and manages the sites Professional Workstation, CADplace France and CADplace UK. He has been living in Europe since 1992, and previously managed workstation marketing in Europe at NVIDIA. Lansford is a professional videographer and his interests include design visualization, simulation, graphics and GPU computing.

Recommended For You