Is NVIDIA’s Latest Graphics Board Too Good for You?

Quadro P6000 great for rendering, VR but may be overkill for ordinary 3D modeling.

The new Quadro P6000 is the fastest NVIDIA graphics board ever. (Image courtesy of NVIDIA.)

The new Quadro P6000 is the fastest NVIDIA graphics board ever. (Image courtesy of NVIDIA.)

NVIDIA is moving fast. The new Pascal architecture has just begun shipping; last week, the new Titan X was launched based on the Pascal architecture; this week, the top-of-the-line Quadro boards were launched.

This is a faster introduction across product lines than NVIDIA did for the last two major GPU releases: the Maxwell and Kepler architectures.

Alongside the Quadro P6000, NVIDIA announced the Quadro P5000. Here are the highlights:

  • The Quadro P6000 and P5000 are based on NVIDIA’s GP102 graphics processor
  • The Quadro P6000 has 24 GB of memory and 3840 compute unified device architecture (CUDA) cores—nearly 300 more cores than the Titan X
  • For virtual reality (VR) and 3D stereo application, simultaneous multi-projection allows for left and right eye projections to be created in a single geometry pass
  • Loaded with GDDR5X memory, which delivers twice the bandwidth of the GDDR5 found on the previous generation board (the Quadro M6000) and is critical for GPU computing
  • Unified virtual memory, available on Linux, will accelerate GPU-computing problems with very large data sets
  • Dynamic load balancing of graphics and computing applications delivers better GPU-computing and graphics mixed-mode operations
  • Designed to accelerate GPU-based ray tracing, video rendering and high-end color grading
  • Boasts 8K display resolutions with support for DisplayPort 1.4

 The basic characteristics of the two new Quadros are laid out below.

Feature

Quadro P5000

Quadro P6000

GPU

Pascal, GP102

Pascal, GP102

CUDA Cores

2560

3840

Memory

16 GB GDDR5X

24 GB GDDR5X

Display Outputs

4x DP 1.4 & 1x DVI

4x DP 1.4 & 1x DVI

Display Support

4 x 4K resolution at 120 Hz

4 x 5K resolution at 60 Hz

4 x 4K resolution at 120 Hz

4 x 5K resolution at 60 Hz

Available

October 2016

October 2016

Pricing

Not available

Not available

It is worth pointing out that the Quadro P6000 is absolutely the fastest, most powerful graphics board in the NVIDIA family. Not only does it have 24 GB of GDDR5X memory, twice the memory of the Titan X, it also has 3840 CUDA cores compared to the Titan X with 3584 CUDA cores. There is no faster GPU in the NVIDIA line up.

Another key point for professionals is that the Quadro P6000 and P5000 unified memory architecture is perfect for compute applications with large data sets running on Linux. The unified memory architecture allows tasks of unlimited size to be calculated and rendered. The unified memory architecture is only possible on Linux. It does not exist under Windows.

Virtual reality is all the rage in the consumer space. Professionals, on the other hand, have been struggling with VR, stereoscopic 3D displays and the relevant technical problems for over two decades. The new Quadro products use a technique called single-pass multi-projection. It allows the Quadro P5000 and P6000 to process the 3D scene one time and generate two perspective views: one for the right eye and one for the left eye. This doubles the performance for stereo projections which, in turn, doubles your budget for complexity and fidelity in the VR, 3D stereo image.

The Pascal architecture can dynamically balance graphics and computing work on the GPU. This enhances the Quadro’s ability to work interactively with realistic, GPU-computed ray tracing in the application viewport. Imagine Iray’s realistic rendering in a Maya viewport for a faster lighting and illumination workflow in visual effects (VFX) scenes.

The professional world is moving beyond 4K. The new Quadro GPUs support DisplayPort 1.4 and 8K resolutions. And they support four simultaneous 5K resolution displays. That is a boon to professionals in video and special effects today, and it will be for engineers, too. It is not hard to imagine 5K displays, currently running USD$1,600, replacing 4K displays at the sub-USD$800 price point in the future.

NVIDIA recommendations for Quadro models gives software applications quite a bit of headroom. (Image courtesy of NVIDIA.)

NVIDIA recommendations for Quadro models gives software applications quite a bit of headroom. (Image courtesy of NVIDIA.)

Users, especially those with budgets to consider, may elect to look at graphics hardware a level below NVIDIA recommendations. NVIDIA seems to have taken care to spec a card for the power user for each software application, not wanting such a user to be hardware constrained. But let’s consider a typical user, say for 3ds Max, who does 3D modeling but rarely, if ever, uses the Iray plugin for GPU ray tracing and rendering. For that user, Quadro K1200 may perform well and be preferred over the recommended Quadro M4000.

Users should also check to see if their particular application can function with or take advantage of the NVIDIA hardware being recommended. For example, ANSYS Fluent’s results are more accurate using double-precision floating point operations, but the GP102 GPUs in the Quadro P6000 recommended above are optimized for single-precision floating point operations. Heavy use of simulation solvers should be steered toward the new Tesla P100 GPUs, which are optimized for double-precision floating point computing and meant for graphics cards in HPC nodes in data centers rather than in desktop workstations.

A Final Word

NVIDIA is upgrading the Quadro family with the Pascal architecture faster than any architecture change that I can remember.

The Pascal architecture is faster than any previous architecture.

The Quadro P6000 surpasses the raw performance of its consumer/gamer counterpart, the USD$1,200 Titan X, adding nearly 300 additional CUDA cores and doubling the graphics memory. In addition, the new architecture, combined with NVIDIA’s excellent GPU-computing support for ray-tracing as well as power to accelerate VR, make the Quadro P6000 and the P5000 worth considering as professional workstation graphics for those doing rendering or creating/viewing VR content.

NVIDIA does not expect to ship the Quadro P6000 and the P5000 until October and has not released any pricing. Keep in mind the NVIDIA Quadro M6000, which the P6000 presumably replaces, was selling for a low cost of USD$4,000.


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.