Rendering, Encoding, Compression, Arithmetic
Blender has an efficient rendering module that runs exclusively on the CPU, even though rendering our benchmark file using GPU acceleration is much faster. A tile size of 16 pixels has proven to be the most efficient for CPUs, so we’re using it for our testing. Both Broadwell-based processors fall in line as expected.
This benchmark, which is based on Maxon’s Cinema 4D, provides the interesting option to have the CPU render in single- or multi-threaded mode. The ratio between the two says a lot about efficiency, illustrating the difference between physical cores, modules and simultaneous mult-threading implementations.
That relationship between single- and multi-threaded performance is of particular interest. The former is the same across both new processors, and comparable to Haswell. Multi-threaded performance, however, is better. This leads us to conclude that Broadwell's multi-core efficiency is improved compared to Haswell, as long as the BIOS and microcode play their parts.
Adobe Media Encoder CC
We’re using a UHD video file with an audio track that we’ve recorded ourselves. It’s saved as an H.264 file with 3840x2160 resolution, 25 FPS, progressive VBR, one pass, and a 320 Kb/s and 48kHz AAC stereo soundtrack. We’re using the integrated software renderer, which makes optimal use of all possible threads. The results aren’t surprising.
WinZip 19 Pro – Compression
The trick with this benchmark is to compress different types of content, such as text, pictures, multimedia files, videos and applications, without producing troublesome overhead due to time-sensitive file operations. This is why we copy all 3.02GB worth of data to an ISO file that can be compressed in one go. We’re using the CPU, not the GPU acceleration via OpenCL.
SiSoftware Sandra 2015 – Arithmetic
If the overall results for the single- and multi-threaded benchmark runs are compared, putting heavy emphasis on integer and 32/64-bit floating-point performance, then the Core i7-5775C is found right between Intel's Core i7-4790K and Core i7-4770K, in spite of the new processor’s lower core frequency.
The Core i5-5675C manages to inch out the much higher-clocked Core i5-4690K. This advantage only emerges if an application performs a lot of 64-bit floating-point operations, though. Otherwise, Broadwell isn't quite as fast. It looks like synthetic metrics, at least, can extract more FP64 performance from Broadwell, though integer and FP32 throughput doesn't really improve.
If you're only looking at single-threaded performance, the SMT-equipped processors do significantly worse. This helps explain why Intel’s Core i5-5675C does better in some applications than its bigger brother, the Core i7-5775C.