When Nvidia launched the GTX 1070 and 1080 back in May 2016, it decisively re-established itself as the upper-midrange market, while extending its lead at the top of the graphics stack. Save for the launch of the new 1080 Ti and a modest bump to the GTX 1080’s RAM clocks earlier this year, that’s where the market stayed until August, when AMD’s Vega 56 took a modest lead over the 1070 and Vega 64 landed just below the 1080 (albeit at much higher power). Nvidia obviously wants to close the gap between itself and AMD at the GTX 1070’s price point, and the new GTX 1070 Ti is designed to do just that.
The 1070 Ti is a neat split between the GTX 1080 and the GTX 1070. Fill rates on the 1070 and 1070 Ti can actually exceed the 1080 (depending on your card’s maximum frequency), since some 1070 and 1070 Ti cards actually clock higher than the 1080. All three cards have 64 ROPS, but the 1080 does maintain a noted memory bandwidth advantage. At the same time, the impact of this on games is smaller than you might expect, since Pascal GPUs are historically pretty memory efficient. GDDR5X is only used on the 1080 and 1080 Ti; the 1070 and 1070 Ti still use normal GDDR5. Pricing, again, splits the difference between the 1070 and 1080.
We’ve rounded up reviews from Anandtech, Computer Shopper, PCGamer, and Tech Report. Here’s how they break down: There’s general agreement, on all sides, that this new GPU closes the gap between the GTX 1070 and the Vega 56, though obviously this is still somewhat game-dependent. There are games where the 1070 Ti pulls ahead significantly and games where Vega 56 still leads by a narrow margin.
This brings up a point we want to address in a bit more detail. It is not unusual for GPU reviews to vary between websites depending on a variety of factors, including testbed configurations, the specific games tested, which API the tests are run in, and the detail levels within the games themselves. With that said, the gap between Vega 56 and the 1070 Ti is much wider in some reviews than others. Computer Shopper shows the GTX 1070 Ti and the GTX 1070 beating Vega 56 in almost every test. Anandtech reports that Vega 56 is ~8 percent faster than the GTX 1070, but that AMD’s lower-end Vega is edged out by roughly 5 percent by the 1070 Ti. PCGamer’s results (you can see the aggregate graph below) are between Anandtech and CS, with Vega 56 slightly behind the GTX 1070 and the GTX 1070 Ti well ahead.
Tech Report has published a sneak peak of both its 99th-percentile scatter plot and its average fps/$ . We’ve gone with the scatterplot to offer a different performance perspective (click through to see the average fps results), but TR’s results look a bit more like Anandtech’s in both cases. Again, the Vega 56 is well ahead of the GTX 1070, but the GTX 1070 Ti is ahead of Vega 56.
I don’t bring up these differences to throw shade on any of the sites in question–if I thought they were producing bad data, I wouldn’t include them here–but as a pertinent example of how GPU comparisons can change depending on which tests you use. Testbed configuration is usually a relatively small variable, provided nothing is egregiously misconfigured, but detail settings can shift game results fairly significantly, even without enabling optimizations like GameWorks that are specifically tuned to run well on Nvidia GPUs. AMD maintains an overall compute workload advantage in OpenCL 2.0 (according to Anandtech), but other tests, like Folding, still favor Nvidia GPUs.
The GTX family continues to win on power consumption as well, though how much this matters varies by the gamer and by the base cooler on the GPU. Desktop owners tend to treat power consumption as a proxy for both noise and heat (which is a pretty reasonable way to treat it, in my opinion), but while Vega never hits the 95C temperatures that Hawaii was known for, its coolers still tend to be louder than their Nvidia counterparts (third-party cards based on Vega are few and far between thus far).
The major practical factor is going to come down to price. Cryptocurrency mining has driven the price of GPUs up dramatically, and AMD cards have been hit harder by this than their Nvidia equivalents. Right now, Vega 56 and 64 availability actually looks pretty good, but this can change depending on how the cryptocurrency market shifts from week to week. And of course we aren’t seeing many ODM cards in market yet that aren’t based on AMD’s own reference design, so that could play a factor in selection as well.
The bottom line is this: Vega 56 remains a good counter to the GTX 1070, the GTX 1070 Ti now offers a small advantage for another $ 70 compared with the GTX 1070’s MSRP, and Vega 64 still isn’t a great match-up against the GTX 1080. If you’re already running a Pascal GPU, the 1070 Ti probably isn’t worth the upgrade, but gamers back on the 9xx or earlier cards would notice a dramatic performance improvement.