3D Workstation Performance
No Cheating
Since we’re comparing graphics cards made by more than one manufacturer, their drivers make a significant difference in the results. Thus, we're shaking up the benchmark suite a bit by deliberately not using the SPECviewperf freeware suite. It just doesn’t reflect the current state of professional workstation and CAD applications. Its workloads are just too short and old. Also, some of its benchmarks don’t even start without the presence of professional graphics cards or their professional drivers.
Understandably (but also unfortunately for reviewers), AMD and Nvidia optimize their drivers to target popular benchmarks. This practice diminishes the value of results we generate since performance differences become unrepresentative of other workloads or card behavior in general (we're talking about you, Maya). It all makes for great marketing, though.
Another benchmark that's nearly worthless due to targeted driver optimizations is Cinebench in OpenGL. We tried to run it with current drivers and found that the results (as well as the changes that the drivers have seen over time to get to them) are more than questionable. We’re not going to use tests like this if we’re comparing drivers from different manufacturers.
SolidWorks 2015
What we like to see is a fair comparison using the complete installation of a common software package like SolidWorks. Pitting AMD’s Radeon Vega FE against Nvidia’s Quadro P6000 reveals that the former provides approximately 90 percent of the performance for approximately 30 percent of the price. Nvidia’s workstation card might win in the end, but we don't think it dissuades anyone from considering AMD's card, either.
Ultimately, the Radeon Vega Frontier Edition provides approximately the same performance as Nvidia’s Quadro P5000, which is the workstation version of a GeForce GTX 1080. What's more, the P5000 sells for just under twice as much as AMD's Vega board.
The individual benchmark results show that there aren’t any outliers among the composite sub-scores. These graphs reflect percentage points to make them easier to read, with Nvidia's Quadro P6000 always set as 100 percent.
Creo 3.0
Creo 3.0 runs well on a Radeon Vega Frontier Edition. In fact, it beats the Quadro P6000 in every single sub-score race. This is what good optimization of an application looks like. That is to say these performance results represent real-world performance; they're not just a benchmark sequence that only lasts a few seconds.
Again, converting our results to percentages proves enlightening. An average performance advantage of approximately 20 percent is nothing to sneeze at.
You can bet that Nvidia will optimize its drivers to be more competitive. And that's good news for users, since professional drivers are just as dependent on optimization as gaming drivers when it comes to extracting a given architecture's maximum potential.
3ds Max 2015
The two graphics cards perform almost identically, despite the wide variety of workloads.
Looking at the percentages, AMD's Radeon Vega FE never lags behind by more than 10 percent, and it manages to pull ahead by six percent in an important area. This is what lets the card draw even, since an overall difference of 0.6 percent is almost within the measurement error range.
AutoCAD 2017 - 3D Performance
AutoCAD utilizes DirectX without any fuss: for the most part, it uses a simple wrapper that sits right on top of Microsoft’s libraries.
Our benchmark shows the Radeon Vega FE landing closer to Nvidia's GeForce GTX 1080 than the 1070.
Practically speaking, graphics card drivers can’t be optimized for this benchmark, so the manufacturers aren't able to manipulate its results. If that means this is what we can expect from AMD’s Radeon Vega FE in DirectX-based applications (including games), a lot of folks are going to be disappointed.
MORE: Best Graphics Cards
MORE: Desktop GPU Performance Hierarchy Table
MORE: All Graphics Content