Nvidia GTX 1660 Ti: Finally, a Turing Worth Paying for
Five months ago, Nvidia launched its new RTX Turing family, with new ray tracing and anti-aliasing capabilities and no improvement whatsoever in terms of performance per dollar. Nvidia may have felt it had little choice — the company appears to have systemically underestimated its exposure to the cryptocurrency market, which led to an enormous glut of cheap Pascal cards throughout the end of 2018 — but enthusiasts largely chose not to take the bait. The company has been forced to acknowledge that sales of its RTX 2080 and 2070 series failed to meet expectations.
The GTX 1660 Ti is the GPU Nvidia hopes will change things. Like its RTX cousins, the GTX 1660 Ti includes the architectural enhancements and improvements that were baked into Turing. Unlike them, it ditches RTX and DLSS compatibility. The net result of this is a smaller GPU, a higher clock speed, better power consumption, and the first significant performance-per-dollar improvement Nvidia has brought to market this entire generation. This problem of flat performance-per-dollar isn’t unique to Nvidia, a fact we addressed in a recent article.
As rumor suggested, the GTX 1660 Ti is a midrange step between the RTX 2060 and the GTX 1060 from last generation. Its launch price of $ 279 is still an increase over the $ 249 that the 1060 debuted at, but not as large of a jump as the other Turing cards have carried.
The GPUs 6GB of RAM is a bit concerning given that competing cards from AMD in this price range now carry 8GB, but the bump to GDDR6 at least significantly improves overall memory bandwidth. The TU116 GPU is also more power-efficient than the GTX 1060 it replaces.
Anandtech’s review of the card can be read here, but it’s not difficult to summarize. This is, by far, the best Turing GPU that Nvidia has launched. More specifically, it’s the only card to offer an actual performance per dollar improvement.
Anandtech’s table sums things up nicely. The problem with Turing, up until now, is that Nvidia raised prices on every Turing card relative to its Pascal counterpart at the same position in the product stack. This forced gamers into a binary choice: You can buy an RTX card for roughly what you paid for your previous GTX card and enjoy no extra performance or you can spend more money for a performance jump.
The 1660 Ti changes this calculus for the first time this generation, and it’s a change I’m genuinely glad to see. As for whether the removal of RTX features matters, I don’t know that it does. The bottom line, for now, is that RTX is a capability that will only be used in a relative handful of games in the next year or two. You can decide, based on the list of supported titles, whether you want to invest in the capability to play specific games or not, but the gap in implementation between Battlefield V and Metro Exodus illustrates how risky this can be. Metro’s implementation of RTX is considered to be better, overall, than BFV’s. In other words, simply knowing that a game will support a feature doesn’t tell you anything about how well the game will support the feature.
As far as how effectively AMD competes, the answer is, “It’s complicated.” The Vega 56 is generally ahead of the 1660 Ti by a small margin, but AMD told us yesterday that the $ 280 price point it was advertising on a Vega 56 was not a permanent price cut. If the GPU were to hold this price, it would actually offer reasonable competition for the 1660 Ti. If it jumps to over $ 300 again, it’s not going to be the case.
The RX 590, meanwhile, is generally outclassed. Again, this isn’t necessarily surprising — the RX 590 has been overpriced, relative to the performance gain over the RX 580, ever since it launched. If AMD trims the price, the RX 590 could continue to offer a relatively strong price/performance point on its own basis, probably in the $ 220 – $ 240 price bracket.
Regardless of what AMD does, however, we’ve finally gotten a Turing GPU that offers a genuine improvement to both performance and performance-per-dollar. Apparently, the only thing that required was dumping the features that were supposed to make Turing shine in the first place. Depending on how you feel about those features, this may not even count as a negative.