We're all now well accustomed to the fact that a given processor's top stable frequency is mostly determined by the limitations of that particular chip—also known as the silicon lottery—and the CPU cooler one straps on top. This latter point is most certainly true for overclocking Broadwell-E parts, especially the top-end 10-core part, where the cooler will have to dissipate an enormous amount of heat when we shoot for high frequencies. Where the motherboard comes in to play is mostly around how easy it makes the journey from stock frequencies to the blistering heights of overclocking success.
We tested the X99-A II's overclocking prowess using a Core i7-6950X CPU with Cooler Master Nepton's 240M strapped to it. The Nepton has a 240-mm radiator, and before it was superseded by the company's MasterLiquid Pro 240, it had a $110 asking price. This puts it right in the ballpark for coolers one might expect to see in a system built around the X99-A II. As I kicked off the overclocking testing, I wondered how our Nepton would fare in the face of such a beastly processor.
AI Suite got the first crack at boosting our clock speeds. We configured AI Tuner for encoding stability, told it to use AVX instructions during the stress test, and enabled the memory stress test option. After the obligatory warning message came up the system rebooted. We were off to the races.
The software started us out by testing our fastest core, labeled core 9 in AI Suite, at 4.2GHz and all other cores at 3.9GHz. After a single core test and then an all-core test, it proclaimed victory and increased clock speeds by 100MHz. This continued until we were testing core 9 at 4.6GHz and all others at 4.3GHz. Stress testing this configuration ultimately led to a blue screen.
After the reboot auto tuning completed, and we were sitting at a 45x Turbo multiplier for core 9 and 42x for all others, with the base clock left at its default 100MHz. Firing up our Prime95 stress test didn't fare so well, however—we saw thermal throttling rear its ugly head almost instantly. Clearly with all cores humming along at 4.2GHz under a core voltage of 1.369V, our Nepton just couldn't keep up.
Next up to the plate was the firmware's EZ Tuning Wizard. We told it that we primarily used the system for playing games and encoding media, and that we had a liquid cooler. A quick reboot later, and we were sitting at just under 3.9GHz with a 38x Turbo multiplier on all cores (save for core 9, which was set to 42x), for a net result of just under 4.3GHz. The base clock had been bumped to 102MHz, and the core voltage was set to 1.325V. This configuration proved stable in our Prime95 stress test, with no errors nor warnings during the run. CPU temperatures peaked at 91° C, though. Not high enough to induce any thermal throttling, but not exactly comfortable for the processor. A liquid cooler with a 280-mm radiator might be a better choice for this baddest Broadwell-E chip.
With all of the automatic overclocking options out of the way, it was time to see what some manual tweaking in the firmware could do. Using multiplier tweaking alone, with the voltages on their "auto" defaults, we made it all the way to a stable 3.9GHz. At this speed, the firmware supplied 1.235V to the CPU, and temperatures maxed out at 93° C under our Prime95 load.
Moving to 40x all-core multipliers wasn't successful. At this speed, the firmware supplied our processor with 1.27V, and within a few minutes of running Prime95 core temperatures were hitting 100°C and we were seeing thermal throttling. It was time to set voltages manually. Using an offset in the firmware, we were able to set a core voltage of 1.24V. This proved to be perfectly stable during our Prime95 stress test. Temperatures peaked at 96°C, and we saw no signs of thermal throttling.
Having conquered the 4GHz barrier—with the vector units on all 10 cores cranking away on 256-bit fused multiply-add instructions—it was time to get greedy. 4.1GHz was in our sights! A quick trip back into the firmware, and we were looking at 20 threads of Prime95 chugging along at 4.1GHz, all on the same 1.24V core voltage. Temperatures rose a few degrees, to 98° C, but we avoided thermal throttling by the barest of margins.
The limits of our slice of Broadwell-E silicon squashed our dreams when we attempted to push to higher speeds. Prime95 instantly produced errors on a number of threads when we tried for 4.2GHz. And sadly, our Nepton couldn't keep temperatures in check when the core voltage was increased.
Our first Broadwell-E overclocking adventure, therefore, ended at 4.1GHz. To get another data point, I overclocked this same CPU/cooler combo on Gigabyte's X99-Gaming 5P. It also managed 4.1GHz for this same stress test. Two samples doesn't exactly make a pattern, but so far it looks like the X99-A II is not leaving anything on the table for Intel's Core i7-6950X.
After all that excitement, it was time to try our hand at base clock overclocking. Unlike the flexibility afforded by the newer Z170 platform, Intel's X99 chipset supports fixed base clock straps that can be used when overclocking the base clock. These straps act as a reduction gear for the DMI and PCI interfaces to ensure that they remain in spec. This allows overclockers to run the base clock at 125, 166, or 250MHz without messing with the chipset and peripheral links.
Setting the CPU strap to 1.25 in the firmware and leaving all other config options on "auto" gave us a 32x Turbo multiplier for a final CPU frequency of 4GHz.
As expected, given our previous overclocking endeavors, at 4.0GHz the firmware was supplying our CPU with 1.27V. This was too much for the Nepton 240M riding atop our Core i7-6950X. Lowering the core voltage to 1.24V, as we did before, gave us a stable Prime95 stress test run at that 4.0GHz clock speed. Attempting to use the 166MHz or the 250MHz base clock straps left us with a board that failed to boot. Thankfully the firmware caught the error and we weren't forced to clear the CMOS.
It's also worth pointing out that X99 boards supporting Broadwell-E, like the X99-A II, have a firmware option to set a negative offset for the Turbo multiplier that is applied when AVX workloads are run. This gives us some added flexibility when overclocking by letting us run the process twice: once for workloads that execute AVX instructions, and once for those that don't. Once all that's done, you can simply set your desired negative offset to account for the fact that AVX workloads can't hit the same peak clock speeds.
Overclocking our Broadwell-E CPU was a very smooth process on the X99-A II. Manually tweaking clock speeds and voltages was a breeze in Asus' firmware. The board's auto-overclocking functionality gave us mixed results, however. The firmware-based method provided a stable, but relatively conservative, result that could prove to be a good starting point for manual tuning. AI Suite, on the other hand, gave an overly optimistic result that induced thermal throttling.
Now that we've had our fun testing the limits of our Core i7-6950X and Nepton cooler, it's time to turn our attention towards the X99-A II's performance.
|Report: Intel Inside co-marketing program will get a budget cut||3|
|Gingerbread House Day Shortbread||2|
|iMac Pro details and release date come into focus||20|
|Radeon Software Adrenalin Edition: an overview||15|
|Tuesday deals: NVMe storage, a GeForce GTX 1080 Ti, and more||5|
|Intel 15.60 IGP drivers are sitting pretty for Okami HD||6|
|Synaptics Clear ID FS9500 fingerprint sensors slip under phone screens||20|
|TR's 2017 Christmas giveaway: goodies from MSI, Antec, and OCZ||30|
|VESA DisplayHDR attempts to demystify HDR-capable monitors||22|
|Full disclosure: while I work for Intel; the opinions I express here are my own I think I understanding the issue you ran into. For the Braswell platf...||+20|