Board & Power Supply
In our opinion, the most interesting part of EVGA's board is the GPU's power supply and its five doubled phases.
But before we go into more detail on that, check out the PCA's sparsely populated right side. Aside from six half-height polymer solid capacitors, two chokes for smoothing currents on the input side, and a handful of connectors, there aren't any other large components over there. That might explain cut-outs in the cooling plate between the heat sink and board.
Indeed, overlaying the heat pipe and copper sink over our bare board shot shows how these pieces fit together. Clearly, EVGA's thermal solution helps cool the chokes as well, just as we proposed during our review of the company's GeForce GTX 1080 FTW2.
The surface area available for cooling is much greater now. That's appreciated, since the coils do dissipate lots of waste heat.
GPU Power Supply
A quick glance at the components EVGA picked for its five (doubled) phases might suggest a power supply that's wildly overkill. But this isn't a configuration built for extreme overclocking; still, there's real math behind EVGA's design.
The ON Semiconductor NCP81274 that EVGA uses is a multi-phase synchronous buck converter able to drive up to eight phases. Thanks to a power-saving interface, it can operate in one of three modes: all phases on, dynamic phase shedding under average loads, or a lower (fixed) phase count for situations when the entire power supply isn't needed. This is important for distributing loads and hot-spots intelligently across the card.
Since EVGA deemed eight real phases insufficient, the company struck a compromise with five phases plus doubling, yielding 10 control circuits. This is achieved using ON Semiconductor NCP81162 current-balancing phase doublers, which monitor two phases and determine which one should receive the next PWM pulse output sent by the controller. So, when a phase receives a 40A load, the doubler splits it into two 20A loads.
One ON Semiconductor NCP81158D dual MOSFET gate driver per phase drives a pair of Alpha & Omega Semiconductor AOE6930 dual N-channel MOSFETs, which combine the high- and low-side FETs in one convenient package. The decision to employ two of those AlphaMOS chips per circuit (totaling 20 of them) isn't necessarily related to EVGA's pursuit of overclocking; there are other technical reasons to go this route...
Let's get back to our 40A load example, split into two 20A loads via phase doubling. Through the use of two dual MOSFETs in parallel, this number then drops to just 10A per package, facilitating a much more even distribution of hot-spots across the board, reducing the parallel circuits' internal resistance, and cutting down on power loss converted to waste heat. The high number of control circuits also cuts the switching frequency, further alleviating thermal load.
The AOE6930s operate efficiently at temperatures through 75 to 80°C, and up to about 20A. Beyond that, you'd be looking at worrying thermal loads. But even in the NCP81274's all-on mode, this provides up to 400A of current, which should be sufficient for any imaginable operating condition. EVGA's real motivation here was clearly to make those control circuits work as effectively as possible. Without spoiling our environmental measurements, the concept turns out well.
Thermal Sensors
EVGA uses an 8-bit flash-type controller made by Sonix that does a good job of capturing data almost in real-time. A total of nine small thermal sensors are positioned above and below possible hot-spots on the board. These feed information to the Sonix chip, which controls the iCX cooler's three fans in an asynchronous manner.
Due to the way those AOE6930 MOSFETs change the control circuits, however, EVGA should have moved the VRM's thermal sensors accordingly. Because the company didn't, you end up with temperature readings that sometimes don't match the measurements taken directly below the components. Still, the differences we saw are quite acceptable. The sensor readings do approximately resemble the true values.
EVGA's Precision XOC software tool displays the output of all nine sensors, going so far as to enable logging. The software also lets you fine-tune the fan controls to help reduce specific hot-spots.
If you'd like to know more about iCX, check out Testing EVGA's GeForce GTX 1080 FTW2 With New iCX Cooler.
Memory Power Supply
The GeForce GTX 1080 Ti FTW3 Gaming employs half-height ferrite core chokes that may be of average quality, but are sufficient for this application. Nvidia even uses them. You'll find these coils in the GPU and memory power supplies.
A total of 11 Micron MT58K256M321JA-110 GDDR5X ICs are organized around the GP102 processor. They operate at 11 Gb/s data rates, which helps compensate for the missing 32-bit memory controller compared to Titan Xp. We asked Micron to speculate why Nvidia didn't use the 12 Gb/s MT58K256M321JA-120 modules advertised in its datasheet, and the company mentioned they aren't widely available yet, despite appearing in its catalog. Because Nvidia sells its GPU and the memory in a bundle, EVGA has very little room to innovate in this regard.
The memory's power supply is located to the left of the GPU's voltage regulation circuitry. It consists of an ON Semiconductor NCP81278 two-phase synchronous buck converter with integrated gate drivers and the same AOE6930 dual N-channel MOSFETs.
Miscellaneous Components
Current monitoring is handled by a triple-channel Texas Instruments INA3221. Those two shunts in the input area are taking the current flow to be monitored.
And with two coils behind the eight-pin power supply connectors, EVGA even adds some kind of filtering against spikes.
The card has a dual BIOS that comes with a small sliding selector switch (master/slave). We advise against force-flashing the master BIOS to a different version, though.
MORE: Best Graphics Cards
MORE: Desktop GPU Performance Hierarchy Table
MORE: All Graphics Content