Results: Synthetics
We begin our benchmark analysis with a handful of synthetic tests designed to emphasize differences between the various subsystems that changed between last quarter's configuration and this PC. Notably, the Core i7-4770K should perform better in well-threaded metrics.
It remains to be seen how the GeForce GTX 780 Ti and Radeon R9 290 fare against each other, though I'd expect the GK110-based board to be faster in most measurements.
When we break down the test's individual components, we can see the Physics suite, in blue, heavily favors the Core i7, while the Graphics component, in black, puts the GeForce in first place. Those two scores play into the overall 3DMark result, in red, which naturally reflects an advantage for last quarter's PC.
My previous machine also enjoys a clear advantage in PCMark 8's Home and Creative benchmark tests, while the Work suite appears to be limited by some other variable.
This quarter's build suffers greatly for its mechanical disk drive. You're not going to see storage play a huge role in the rest of our benchmark results. However, the above chart illustrates that you're going to wait a lot longer for Windows to boot and your favorite applications to load.
Intel's Core i5-4670K can't keep up with the Core i7-4770K in Sandra's Arithmetic module, which obviously takes advantage of as many logical processors as possible. Both Haswell-based CPUs support similar ISAs.
When it comes to the Cryptography test, however, other factors come into play. The Encoding/Decoding benchmark is accelerated by AES-NI, so performance is dictated by the rate at which system memory can feed data into the CPU. Sandra's Hashing test depends more on ISA support and clock rate.
And here are the numbers largely responsible for that Cryptography score. My newer build's memory defaults to 1333 MT/s, while the XMP profile increases it to 1600 MT/s. Last quarter, I also used a 1333 MT/s default, while XMP helped me overclock to 1866 MT/s. Haswell isn't a particularly bandwidth-starved architecture though, so we're not expecting these exaggerated results to manifest as pointedly in the real-world tests.
