Modern motherboards typically have less impact on peak CPU frequencies than the characteristics of individual chips and the effectiveness of the associated coolers. That said, mobos still determine how much effort is required to reach top speed—and how much cursing occurs along the way. To get a sense of the overclocking experience, we dropped in a Core i7-5960X processor, strapped on a mammoth liquid cooler, and pushed the CPU as far as we could.
Auto-tuning mechanisms are useful for establishing a starting point for manual overclocking, so we let Asus' AI Suite software have the first crack at the CPU. The automated intelligence ramped the chip up to a peak Turbo speed of 4.5GHz on 1.3V. That speed applied only to single-core loads; with all eight cores occupied, the CPU ran at 4.3GHz.
The auto-tuned config survived our usual torture test, which combines the Unigine Nature benchmark with AIDA64's CPU stress test. CPU temperatures hovered around 56°C, and there was no evidence of throttling. So far, so good.
Feeling cocky, we set the all-core CPU multiplier to 45X and fired up our torture test once more. Everything was OK for a few moments, but a BSOD error soon appeared, forcing a reboot. We had to nudge the CPU voltage up to 1.325V to get the system stable with the CPU pinned at 4.5GHz. At that speed, CPU temperatures crept into the 70-76°C range.
Our test rig booted at 4.6GHz without complaint, but BSOD errors hit as soon as we launched any kind of multi-core load. Increasing the CPU voltage didn't help, and neither did lowering it or adjusting other system voltages. Same deal with the power limits, which were mostly maxed out already. We even kicked the cooler into overdrive by ramping its 140-mm up to full speed, to no avail. 4.5GHz was it... or so we thought.
When we tried to grab some Cinebench scores, the dreaded blue screen returned. The multithreaded rendering benchmark is apparently more strenuous than our usual stress test—or just better at pushing hot-clocked chips over the edge. We had to dial back to 4.4GHz to get Cinebench working, which at least let us drop the voltage to 1.3V.
Asus tells us that 4.4-4.5GHz is an average overclocking result based on the pre-production chips it's tested. Some of the company's samples have gone up to 4.7GHz, while others haven't made it past 4.3GHz. For what it's worth, our Cooler Master Nepton 280L water cooler seemed up to the task of keeping thermals in check. Temperatures spiked up to 84°C at top speed, but there was no evidence of throttling.
Although I will admit that some colorful language escaped my lips after a few overclocking-induced crashes, the overall process was silky smooth with both the auto-tuner and the manual tweaking software. Overclocking with the firmware was pretty pleasant, too. The "auto" voltage and power settings did a good job of keeping up when we increased the CPU multiplier manually.
We intended to compare the Deluxe to another X99 board, but our second subject got stuck in customs and arrived far too late to test for this review. We had to switch to the next best thing, Asus' X79 Deluxe.
The trouble is, we could only pair that motherboard with an older Ivy Bridge-E processor that has fewer cores, higher clocks, and a different architecture than the Haswell-E chip in our X99 rig. The X79 Deluxe uses different memory, too, and it has a different chipset. Those factors make it difficult to isolate the motherboard's effect on system performance.
Scott's Core i7-5960X review is a much better resource for information on how Haswell-E compares to Ivy-E; it's loaded with far more application benchmarks and alternate CPUs than we include in motherboard reviews. We can, however, tell you a few things about how the X99's peripheral performance compares to that of its predecessor.
USB transfer rates are faster, but that's not exactly surprising. The X79 lacks integrated SuperSpeed connectivity, forcing boards to use third-party controllers. Auxiliary peripheral chips have traditionally been slower than Intel's native implementations, and that dynamic has not changed with the X99.
The ASMedia chips on both boards deliver similar throughput—and much lower transfer rates than the X99's integrated controller. We've seen a similar performance disparity when comparing the USB performance of Intel's Z97 chipset to that of third-party controllers.
SATA performance is similar between the two Deluxes, whose chipsets both support 6Gbps speeds. Those results aren't even worth graphing. Neither are the networking scores, which are nearly identical. (The X99 system exhibited slightly lower CPU utilization during GigE transfers, but it also has more CPU cores at its disposal, making the comparison a bit moot.)
There a slight difference in boot times, with the X99 Deluxe reaching the Windows desktop a few seconds after its X79 counterpart:
Both are fairly slow compared to the Z97 boards we've tested, which boot in just 14-16 seconds. The X99 Deluxe and its forebear spend almost that long just winding up before their initial splash screens appear.
We have more X99 motherboard reviews planned, so we'll be able to assess the Deluxe's performance against its true competitors soon. Don't hold your breath for fireworks, though. Motherboards typically have a negligible performance impact when all other system components are the same.
Motherboards can influence total system power consumption, but the numbers we've gathered won't shed light on that fact. They're tainted by differences between the systems' respective Haswell-E and Ivy Bridge-E processors and DDR4 and DDR3 memory. Again, Scott's CPU review is a better source for power consumption data along those lines. Make what you will of the following figures.
The X99 rig consumes less power than the X79 at idle but not under load. The EPU power-saving measures on both boards have little effect on power draw at the wall socket.
We're more interested in how the X99 Deluxe's power draw compares to that of the Haswell-E alternatives. Stay tuned.