P500X Versus P900DX: Worth The Money?
And so, we come to the end of our first workstation review using the updated benchmark suite and new baseline test system. We didn’t lose any of you did we?
In comparing these two systems, we had to consider their vital statistics: one is a budget-oriented workstation and the other is a power-user-oriented machine with 16 cores, able to operate on 32 threads concurrently. One is going to serve as the baseline for our future workstation reviews, while the other gets shipped off somewhere else to lend its huge pool of memory and bandwidth to real-world professional problems. Fortunately, fairly similar storage subsystems helped us narrow the number of variables affecting each machine's results. If the P900DX had come with a couple of storage drives in RAID 0, several of the editing tests would have come out vastly different.
In several tests, Intel's Ivy Bridge architecture and higher clock rate allow the P500X to outperform the P900DX, while the higher-end machine leverages its sheer number of x86 cores and extra memory bandwidth allow it to dominate threaded tests. Perhaps the best examples of this are the 3D rendering tests, which split images into tiles (or, in the case of the LightWave 3D renderer, stripes) and let each processor core handle one of them at a time. Tiles that take longer than others merely occupy a single core for longer, while the other cores move on to additional tiles. LightWave 3D’s renderer instead subdivides the stripes that are taking longer, handing pieces of that stripe off to other processor cores, which in most cases keep utilization more balanced.
The P900DX's GPU is marginally faster than the P500X's, giving it a lead in all of the graphics-oriented tests. If the cards were switched, the performance would partially reverse, but not completely. The P900DX's extra memory bandwidth helps when preparing and loading scenes onto the GPU.
Current GPU-based rendering really is outgrowing many of its prior limitations. Early on, iray was limited to photometric lights and couldn't use standard 3D animation lights (point, spot, directional, and area). While this has changed, GPU-based rendering still suffers from other limitations in lighting, shading, and rendering. Another limitation is that the entire scene must fit onto the graphics card's memory. The car scene was originally designed to include two different cars, but that revision made the scene require more GPU memory than the Quadro 2000-equipped P500X could provide.
Another major consideration is that the Windows UI itself becomes virtually unusable when you tax a GPU with rendering duties, since most of its resources are fully utilized by the rendering task. If you are planning on doing GPU-based rendering extensively, we advise you to buy a second GPU, not in SLI, and offload work to the second card for iray or V-RayRT viewport rendering in 3ds Max, for example, and adjust your render interactively. The card doesn’t need to be identical (as it would in SLI configurations), or even similar. Just make sure it's supported by the renderer.
Overall, P500X is significantly slower than the P900DX, just as we'd expect to it to be. Given the differences in system specs, this was a foregone conclusion.
The P500X is still very competitive in single-threaded tasks, which emphasize its Ivy Bridge-based CPU, as well as tasks dependent more on clock rate than core count. The P500X would likely serve very well as an entry-level workstation. Given a more powerful GPU, it could even serve as the primary system for a 3D modeler or someone doing texture painting.
The P900DX on the other hand, is the type of system that would be given to someone doing heavy shading, lighting, or compositing work. With a good RAID array, the P900DX would also make an excellent system for heavy video editing, as its stronger GPU, additional memory bandwidth, and CPU cores make it quite a bit faster for that task.
Is the P900DX worth four times as much as the P500X?
That depends on the work you're doing, and if that work can be distributed across multiple systems. If not, the P900DX's many cores and accommodating memory subsystem is great for applications able to exploit it. It could easily halve the amount of time you spend on a money-making project, allowing you to recoup the additional investment in fairy short order. If your workloads are more easily distributable, it'd be smarter to buy multiple lower-cost systems, and that's where something like the P500X comes into play.