Results: Workstation Applications
OpenGL: Cinebench R15
We decided to run one more small synthetic benchmark before getting to the heavy hitters: Cinebench R15’s OpenGL benchmark. In this test, clock rate rules and more than four cores aren’t necessary. The results aren’t exactly newsworthy, but they do make for a nice transition to the larger applications.
The Creo test’s biggest advantage (and disadvantage, to be honest) is its long run time. The longer a benchmark runs, the more reliable its results tend to be. If the results of different CPUs are still the same after an hour, then there shouldn't be any noticeable differences between them during daily use.
|Workflows||CPU Composite (1)I/O Composite (1)Graphic Composite (6)|
|Content||- Open, Generate, Close and Delete Workflows- GPU- and CPU-based Output Computations (Tessellation, Mass Properties Analysis)- All Possible Types of Shading Functions and Movements (Rotate, Pan, Zoom)- Usage of Screen Space Ambient Occlusion (SSAO)- Bump Mapping and Image Background Support- Snapshots|
Let’s take a look at the three composite results of the individual benchmarks, starting with CPU performance. Once again, the eight simulated threads are more of a hindrance than an asset. True quad-core CPUs fare significantly better. However, this is only the case for pure CPU testing, which doesn’t provide us with the whole picture.
If we’re looking at the visible results instead, which is limited to the graphical output, then SMT doesn’t matter that much and higher clock rates are all you want. This means that the equivalent Kaby Lake and Skylake models with the same number of physical cores generate the same performance.
The I/O composite numbers round out the results and paint a similar picture. Kaby Lake’s only advantage over its predecessor comes from higher frequencies. Anyone considering an upgrade should spend some time thinking about the value story here. Otherwise, you'll end up spending a lot of money going sideways instead of stepping up.
Solidworks by Dassault Systems is also very CPU-dependent. Additionally, it’s a benchmark that runs for long periods of time and doubles as a stress test. Overclocked CPUs are actually more likely to show errors during this benchmark than during the more common Prime95 or OCCT runs.
|Workflows||CPU Composite (2 Tests)Graphic Composite (9 Tests)|
|Content||CPU- Tessellation- PhotoView360 RenderingGPU- RealView- Ambient Occlusion On- Shadows- Up to 4.75 Million Triangles- Order Independent Transparency (OIT)|
The CPU composite results show very clearly that processors able to work on lots of threads at high clock rates have the advantage. SMT is a big help as well. In this case, the benchmark tasks consisted of CPU-based tessellation and rendering.
Conversely, graphics output performance depends primarily on a CPU’s frequency and IPC throughput. This makes the comparison between Kaby Lake and Skylake at the same clock rates a worthwhile endeavor.
This software suite is a classic that runs well on consumer graphics cards since DirectX is used for 3D rendering instead of OpenGL.
Ever since the introduction of the unified shader model, graphics cards haven’t included specialized 2D hardware. Also, hardware-based 2D commands are all but extinct due to the driver model of all Windows versions since Vista. This means that many steps need to be computed by the CPU before a direct output can be executed. So, the 2D test exposes even the smallest weakness in IPC performance.
There’s a small difference between Kaby Lake and Skylake CPUs in 2D, but it vanishes completely in 3D. Once again, multi-threading doesn’t really provide an advantage. The clock rate's really all that matters.