EVGA is well known for making very good GPUs. Their ACX coolers, now in the third generation, are both attractive and functional. They don’t attempt to look like some alien creature or something that it’s not, but simply a a giant heatsink, two redesigned fans and a bevy of LEDs to spruce things up a bit. Sometimes LEDs get a bit ostentatious, becoming a large selling point and even driving up the cost without actually providing any real-benefits. But is there a balance between adding flashy pieces and making a quality GPU?

It’s unfortunate that the early samples of EVGA Pascal GPUs came sans VRM cooling pads, something that should have been planned for so as to forestall any unfortunate accidents from those wishing to push the boundaries of their GPUs well beyond the specifications of the actual hardware. The fan does actively cool the VRM components, though this seems to not be enough causing outrage among some who simply are not satisfied with the design. Despite that shortcoming, evident on our sample here, we’ll explore the many facets of what makes a good GPU. The GTX 1070 certainly makes a good value proposition, but does EVGA have a good argument for their interpretation of NVIDIA’s GTX 1070?


GTX 1070 FTW is bright and fast

And hot, though that’s an issue for a bit later. On top we see an identification plate that accouterments that’s quite bright with built-in LEDs. It’s not unattractive in the least, and certainly not too not terribly flashy. There’s also six quadrants filled with RGB LEDs that can be customized at your leisure. Instead, they opted for a massive cooler, a great overclock and two 8-pin and power connectors. Those connectors alone speak to the potential lying in wait. This is clearly built to be overclocked, and they urge you to. There has been some mention of VRM’s that may get a bit too hot, though the heatsink and fans do provide coverage over them, they aren’t touching and initially had no means of easily transferring heat away. This is of the older design and thus we’ll see if there truly are any issues present.

Looking at the board itself it’s fairly well built with a mostly understated motif. The LEDs are noticeable, though they don’t detract from the actual usefulness of the heatsink itself. It’s built atop a 10-phase design and a massive heatsink on top to help dissipate the heat.


The fans are of their third generation ACX and are more than sufficient for the job for which they’re installed for. As they’ve done with all their coolers already for some time, the fans are completely off when idle. These fans are much improved over their predecessor, with newer ball bearings and a slightly tweaked design. The heatsink, too, is larger and much better designed than before. The heat pipes provide more direct contact than the previous design as well as having more of them to boot. This will hopefully allow more heat to dissipate away.


The two eight pin power connectors combine for a theoretical maximum of 300W that can be drawn by the card. That’s far and above the TDP of 180W. This was designed, at least as it appears from an outside observer, to be an overclocking focused card, and one that would likely benefit from being underwater if you wished. The MSRP of $679 is a bit higher than other competing variants that seem to offer more. But in reality, such as with cars it seems, LEDs do nothing to increase the speed of the card. I want something that’s built to go fast, not that looks like it goes fast. This may fit that bill.



The design is understated in it’s beauty. And it looks great in nearly any situation. The problem of LEDs becomes less so when you look at how robust all the components are. This seems like an overclockers dream card, at first glance. The components that EVGA has used are all very high quality, meaning that even without increased cooling on the VRM’s themselves, they can withstand quite a bit of heat and have larger tolerances. The potential for high overclocks is definitely there, and ASIC quality may be the only limiting factor, that and cooling. Regardless, this looks good and though the price is a bit higher than a snazzier and more lighted version, but it look good.


Pascal itself has a number of massive steps forward in the interest of compute. As NVIDIA has said in the announcement of Pascal during GDC, it’s one of the most advanced architecture around. Though it seems they’ve seen fit to optimize this for the use in AI and deep learning applications. The large Pascal die has a massive amount of transistors, 15.3 billion, and dedicated 16FP hardware. That left a lot of questions in our minds, such as how that was to translate to the consumer versions made to display pretty graphics.

Taking the FP16 hardware out, a few SP’s and adding in the necessary ROPs and TMMs gave us a sophisticated GPU that promises performance that’s faster than even the mighty Maxwell-based Titan X. All while drawing less power and potentially being much cooler. NVIDIA has also seen fit to properly enabled Asyncrhonous compute, though as we’ve seen in our recent DOOM Vulkan benchmarks, it doesn’t seem to share quite the advantage that AMD does.

For years NVIDIA has pushed their own version of Async compute that revolves around their implementation of proprietary APIs through GameWorks and the like. They could leverage CUDA to perform compute and rendering tasks asynchronously in DX11 and below. DirectX 12 is far more advanced, and enables such on the lowest possible levels, something that isn’t quite in line with NVIDIAs architecture at the moment. But that doesn’t mean it performs poorly, just that the increase is slightly less than competitors.


But enough of that, let’s get on with how this performs. It’s one tough board, supposedly, so let’s put it through its paces. We’ll use the same default-clocked system as always and put it through a good compilation of games to show you how well this card is compared to the competition. Let’s test, see if those VRM’s are really an issue