Choosing a Graphics Card for Premiere Pro CS5.5
Adobe's Mercury Engine, as powered by NVIDIA's CUDA-enabled graphics cards, can be a huge time saver during project preview and rendering. But the performance benefit depends upon the source content and project type. In some instances, a high-end CUDA card delivers the most benefits; in others, investing in a dual-CPU workstation makes more sense.
This detailed analysis looks at various source formats, including DSLR, AVCHD, HDV, AVC-Intra, DVCPROHD and Red, and various project types, from simple, single track one-camera shoots to high-production value multi-layered projects, to identify where and when Adobe's highly-touted Mercury Engine delivers the maximum benefits, and where it doesn't. Learn when investing $1,800 on a high-end card makes sense, where a $400 card will perform just as well, and when a dual-CPU workstation is the best investment.
Suppose you’ve been running Adobe CS5.5 without an NVIDIA graphics card. You’ve heard about the benefits of GPU acceleration with the Mercury Playback Engine and you’re wondering how much time an NVIDIA card could save you. Or you’re buying a new system and you’re wondering whether to buy a dual-CPU system with an inexpensive graphics card or a single CPU system with a high-end card. Or you’ve got an NVIDIA card and you’re wondering whether a higher-end NVIDIA card will deliver substantial time savings. Well, if any of these cases apply to you, you’ve come to the right place.
In this article, I’ll start by explaining the factors that contribute to when and how much NVIDIA-powered GPU acceleration can save in rendering time and preview performance. Then I’ll discuss a series of benchmark tests performed on two HP workstations, a Z400 with a single 2.67 GHz 4-core CPU and a Z600 with two 2.67 GHz 4-core CPUs, both configured with 24 GB of RAM and were running 64-bit Windows 7. The tested graphics cards included the Quadro 2000 ($423 street), Quadro 4000 ($730 street) and Quadro 5000 ($1,700 street) and older-generation Quadro 4800 ($1,154 street).
Let’s define some terms to make sure we’re all on the same page. GPU stands for graphics processing unit, and GPU acceleration refers to a process that the GPU can accelerate, which takes the load off the computer’s CPU. CUDA stands for Compute Unified Device Architecture, and it’s one of the computing engines deployed in NVIDIA GPUs.
Adobe’s Mercury Playback Engine (MPE) describes a bundle of performance enhancements in Premiere Pro that includes 64-bit operation, more efficient multi-threaded performance and the ability to accelerate some functions using some CUDA-enabled graphics cards, which are listed here. Only CUDA-enabled graphics cards can work within MPE to accelerate Premiere Pro. This means that only some NVIDIA cards can accelerate MPE, and (currently) no graphics cards from any other vendor can.
With this behind us, let’s dig in. To start, recognize that the potential benefit of CUDA-enabled GPU acceleration in CS 5.5 depends upon three factors:
- Whether you preview or render with Use Maximum Render Quality enabled. If you do, typically you’ll see significant performance benefit. If you don’t, on some very rare projects, GPU acceleration may actually slow you down.
- The source formats used in the project. Typically, the more compute-intensive the format, the more significant the benefit of GPU acceleration.
- How many layers and effects you’ve applied to your projects. As you would suspect, the more layers and GPU-enabled effects, the better GPU-accelerated encoding performs compared with CPU-only.
I’ll cover the Maximum Render Quality issue first. Then, to explore all three factors in greater detail, I’ll tell the tale of six very disparate projects, describing the source materials, layers, applied effects and the like, as well as detailing the benefits of GPU acceleration for that project. This will give you a feel for how and where GPU acceleration really matters, and which NVIDIA card makes the most sense.
With this as background, let’s jump into the technical discussion.