ProsIncredible performance, eclipsing other recent GeForce offerings by a significant margin; quiet and effective cooler; relatively compact at 10.6" long
ConsHigh power use; significant heat output
We always think a little bit of history is important to appreciate the full story behind every video card product on the market today. Back in late 2013, AMD's Radeon R9 290 and 290X shook things up quite a bit, overtaking Nvidia's Titan and GTX 780 for the overall performance lead. Nvidia countered quickly, however, and its ultra-high-end GTX 780 Ti, released a month after the Radeon duo, salvaged a small win for Nvidia going into 2014. The standings remained much the same until from this very tenous postion of power Nvidia launched its high-end Maxwell 2.0 design in September 2014. Coming in the form of the GTX 970 and GTX 980, Nvidia had a commanding position in the video card market for the first time in years. The GTX 970 more or less equalled the R9 290X and nearly reached the level of the 780 Ti, while the GTX 980 shot well above it. AMD had no response at that time, so the GTX 980 reigned supreme until Nvidia launched its fast but very expensive Titan X in March 2015. That card wasn't meant for the masses (and was nearly impossible to actually find in stock for several months), but it showed that Nvidia was intent on keeping its dominant position.
Entering June 2015, all eyes were on AMD, as rumors had been flying regarding its next big video card launch. At first it was believed this product would be called the R9 390X, but that card in fact ended up being a simple refresh of the 290X. The true next-gen card would be called the Fury X, and in using a catchy premium naming scheme, it borrowed a page out of Nvidia's marketing handbook. It was scheduled to land near the end of June, but of course, Nvidia wasn't going to let AMD have all the fun. In early June, it preemptively launched its Fury fighter, the comparatively maintream-sounding GTX 980 Ti. First available only in a reference design with a blower-style cooler, the 980 Ti was nonetheless impressive, and indeed put a damper on the subsequent Fury launch. Unlike with the Titan X, Nvidia decided to allow board partners to install aftermarket coolers on the 980 Ti, which meant more performance could be extracted from the powerful GPU. In today's review, we'll be looking at the ACX-based model from EVGA, one of the first aftermarket models out of the gate. At left, you can see our sample sitting on top of its product box.
We'll be looking at a number of different cards in our benchmarks, as follows:
- Sapphire Radeon R9 290 Tri-X 4GB at reference clocks of 947MHz/5000MHz (our Tri-X sample runs cooler than the reference version and thus can better maintain this core clock)
- Sapphire Radeon R9 290 Tri-X 4GB emulating a Radeon R9 390 8GB at clocks of 1000MHz/6000MHz. In our benchmarks, this card is labeled as the "R9 390*" (note that 4GB VRAM deficit versus the real R9 390 doesn't come into play at the 1080p resolution we used).
- EVGA GeForce GTX 780 Ti 3GB, now discontinued, with reference in-game clocks of 1006MHz/7000MHz
- EVGA GeForce GTX 980 FTW 4GB, at reference in-game clocks of 1240/7000MHz
- EVGA GeForce GTX 980 FTW 4GB, overclocked to 1497MHz/7500MHz
- EVGA GeForce GTX 980 Ti ACX 2.0+ 6GB, at reference in-game clocks of 1202MHz/7000MHz
- EVGA GeForce GTX 980 Ti ACX 2.0+ 6GB, overclocked to in-game clocks of 1427MHz/7600MHz
For a few of our benchmarks, we also included data for several older cards that we had on file:
- Sapphire Radeon HD 7870 2GB, now discontinued, with clocks of 1000MHZ/4800MHz
- Sapphire Radeon R9 270X 2GB, a mainstream gaming card, with clocks of 1050MHz/5600MHz
- EVGA GeForce GTX 670 FTW 2GB, now discontinued, with reference in-game clocks of 1058MHz/6000MHz
- EVGA GeForce GTX 780 ACX 3GB, now discontinued, at reference in-game boost clocks of 1006MHz/6000MHz
In all of our tests, we provide performance at the popular 1080p resolution. Where possible, we show not just average framerates but also minimum framerates, which are often a better indicator than averages of the perceived smoothness of gaming performance. While we typically also test at 1440p, for this article we chose to bench with a greater number of games instead. The test bench used consisted of the following:
- CPU: Intel Core i7-4770K overclocked to 4.4GHz
- Motherboard: ASRock Z97 Extreme4 (thank you to ASRock and Newegg for providing this review sample)
- SSD: Samsung 850 Evo 500GB
- RAM: 2x G.Skill TridentX 2x4GB DDR3-2400
- Case: NZXT S340
- CPU Cooler: Noctua NH-U12S (thank you to Noctua for providing this review sample)
- Power Supply: EVGA Supernova 1000 PS
- OS: Windows 8.1
- Drivers: Nvidia GeForce 353.30; AMD Catalyst 15.6 Beta (older drivers were utilized for the discontinued midrange cards)
Overclocking, Power Use, and Temperature
Our sample card ran at reference clocks, which are officially rated at 1000MHz core, 7000MHz memory. The card in fact runs at 1202MHz core in-game, as a result of Nvidia's Boost 2.0, assuming power and temperature are within pre-set limits. We found that the card would step down 13MHz when temperatures hit 70C, or when the power limiter was exceeded (it defaults to 100%, but can be adjusted up to 110%). And while the ACX cooler is far more effective at managing heat than the reference blower-style cooler, when installed in a closed mid-sized case, the card was unable to stay under 70C for an extended period of testing. This was particularly noticeable during overclocking. As a result, our 1427MHz overclock, achieved with a +0.4V to voltage, actually settled down to 1414MHz after a few minutes. Note that 1380MHz was achieved without the voltage bump.
Overall, power use is high, but impressively-low if taken within the context of the performance, which you'll see shortly. At 359W, this card draws nearly as much power as the GTX 780 Ti, and far more than the GTX 980 or even the power-hungry Radeon R9 290. Jumping ahead a bit, the 980 Ti performed 24% faster overall than the GTX 980 in our full benchmark suite, and in that light its 27%-higher power draw actually is quite reasonable. Still, this isn't the ideal card for lower-end systems sporting power supplies under 500W or minimal ventilation.
In addition to using 3DMark for power testing, we use it as a baseline as well as a stability test for our overclocking. It's not a perfect comparison between video card generations or manufacturers, so don't use it to judge overall performance - it's better as a utility to indicate potential scaling of a single card's overclock. We use the Graphics Score only here, rather than the overall score, which is heavily-influenced by CPU performance.
Note that were are reporting the "Graphics Score" here, which is what's relevant when comparing video cards. While the 3DMark score is often what's reported, it takes into account the CPU and memory subsystems, and we want to take that out of the equation here to focus on the GPUs. We see that the 980 Ti takes a commanding lead on all the other cards tested, including an impressive 54% lead over the 780 Ti which it ostensibly replaces. It's also 30% faster than the 980, which is impressive given that they use the same Maxwell architecture (the 980 Ti is essentially a scaled-up version of the 980). The delta is big enough that no amount of overclocking on the 980 is going to allow it to catch up to the 980 Ti, an unusual situation when it comes to cards one step apart in the product stack. The GTX 780, for instance, could easily catch the 780 Ti once overclocked, although we don't show that here. Note that this benchmark, along with Tomb Raider on the next page, include several older cards for reference: the HD 7870, the R9 270X, the GTX 670, and the GTX 780. That may be of interest to those upgrading from older or more mainstream cards.