System Builder Marathon Q4 2014: $1600 Performance PC

Benchmark Results

The new build’s stock settings look a little slow in 3DMark’s CPU-based Physics score, which is odd since its motherboard defaults to 100.5 MHz (a 0.5% “stock” O/C). That loss turns into a gain when overclocked, even though both processors were set to the same multiplier when overclocked. Perhaps a variance of 100 3DMarks is normal?

Or maybe I’m just burying the lede. Enormous gains in GPU performance led to large victories for the new system in 3DMark’s overall score.

Similar inconsistencies in PCMark’s Home and Creative scores aren’t a big deal, since I really don’t use those scores. Based on the transfer patterns of real-world programs, storage is the only PCMark score that carries any weight in our overall evaluation. The new system loses that one by 1%.

The new build also trails the Q3 machine in Sandra Arithmetic. This is starting to sting a little, but I still hold hopes that it won’t suffer any losses in our real-world applications.

Sandra Cryptography shows mixed benefits for the Q3 machine’s higher baseline DRAM data rate.

The higher baseline memory bandwidth is far more apparent in Sandra’s memory bandwidth test. Both sets of RAM reach the same overclock, so both get the same average overclocked bandwidth.

Battlefield 4 shows surprisingly little differentiation between the Q3 and Q4 systems at medium details and high resolution. Lower resolutions reach the game’s 200 FPS cap.

The GeForce GTX 980 shines at Battlefield 4 Ultra settings, though its biggest gains are found at medium resolution. Perhaps it could have taken a bigger lead at medium details too, if not for the game’s 200 FPS cap.

Well known for its CPU and DRAM bottlenecks at medium resolution and “High” details, Grid 2 still shows some type of benefit from the Q4 build’s GeForce GTX 980. The most likely reason for these results is that AMD graphics have more CPU overhead, and this is something I’ve proven in previous-generation tests.

Relatively large gains for the Q4 machine’s GTX 980 diminish at higher Grid 2 resolutions, where the performance benefit would have been most-useful. Aggressive overclocking helps though.

Arma 3 puts the squeeze on the new system at standard settings, where it appears to scale in proportion to CPU and DRAM clocks. Slightly larger gains at lower resolutions may point to the Q4 card’s lower CPU overhead.

Aside from the weird results at 1600x900, Arma 3 numbers finally break out in the Q4 machine’s favor at the game’s Ultra quality preset.

Far Cry 3 appears to have a 125 FPS cap, but I bet The Donald (Woligroski, SBM builder) has a workaround. GPU overclocking is the new machine’s claim to glory at the game’s “High Quality” settings.

The Q4 machine’s lead expands at Far Cry 3 Ultra Quality, but not as much as we’d like at 5760x1080. With a 23 FPS reported minimum at stock settings and a few long frame times scattered throughout, overclocking is needed to eliminate choppiness.

The Q4 machine’s stock settings appear slightly slower than Q3’s in multi-threaded loads, so I checked to see why. It turns out that while the previous system supported “enhanced” turbo modes by default (4.40 GHz with all cores loaded), the new machine follows Intel’s official (non-enhanced) Turbo specifications. Both machines use the same multiplier when overclocked, allowing the new system’s faster RAM and slight BCLK O/C (0.5% by default) to push a slight lead.

OpenCL shines on the Q4 PC’s GTX 980.

The benefits of “enhanced” turbo again appear for the Q3 default configuration throughout our productivity suite. Fortunately, the new build gets a couple miniscule wins when comparing both overclocks.

The Q4 system falls behind the Q3 build in three out of four file compression tests, and the differences are tougher to define. I’m fortunate that those losses are small.