The GTX 1660 Ti Arrives To Replace NVIDIA’s GTX 1070, But Not With Ray-Tracing

The largest failing of NVIDIA’s Turing-based product stack was the lack of an affordable option. Late last week, the company moved to partially solve that problem, announcing the GTX 1660 Ti.

You’ll notice that it’s an 1000-series card, and that’s not by mistake. The 1660 Ti doesn’t have any on-board Tensor Cores, meaning it lacks the hardware required for real-time ray tracing. It also isn’t capable of powering the other neural network techniques NVIDIA has been touting with their RTX cards, like the deep learning supersampling that’s appeared in Final Fantasy XV and Metro Exodus.

[referenced url=”https://www.kotaku.com.au/2018/11/nvidia-rtx-2080-ti-rtx-2080-gtx-1080-ti-benchmarks/” thumb=”https://www.kotaku.com.au/wp-content/uploads/sites/3/2018/11/IMG_20181017_143933-410×231.jpg” title=”Nvidia’s RTX 2080 Ti, RTX 2080 And GTX 1080 Ti Compared” excerpt=”Ray tracing is all well and good, but how many frames for your buck do you actually get from Nvidia’s new RTX 2080 Ti and RTX 2080 cards?”]

But the 1660 Ti isn’t meant to knock those games out of the water, anyway. NVIDIA’s pitch was that it would be at least twice, if not three times better, as powerful in some games for those rocking a GTX 960.

The pitch for those who are comparing between discounted GTX 1060 cards and a GTX 1660 Ti is a little technical. Take a game like Shadow of the Tomb Raider. Over the course of 100 instructions that the game might send to the graphics card, some of those might be floating point instructions and others might be integer point instructions.

On older cards with older architecture, the GPU would have to alternate between which instruction could be carried out. Turing-based cards can run these concurrently, and while the RTX GPUs do this using the Tensor Cores, NVIDIA introduced some specialist cores for floating point instructions on the GTX 1660 Ti. The supposed benefit is improved frame rates, at least in games that utilise vastly more complex shaders (which modern games are tending to do).

How that works out in practice depends on the game in question. For the most part, third-party reviewers have found it pretty much goes toe to toe with the GTX 1070 and is a better value buy than the RX 590.

Locally, the GTX 1660 Ti is available from $469 which is about $30 more than what it’ll cost you to get an RX 590 in Australia. Given that an RTX 2060 will set you back approximately $579, and some retailers are still selling old GTX 1070 cards for more than that. And given that the Vega 56 will set you back at least $600, the GTX 1660 Ti is sitting fairly pretty price-wise when it comes to the Australian market.

Here’s the full specs:

NVIDIA stressed that the card won’t support SLI – a feature exclusively restricted to the RTX 2080, RTX 2080 Ti and RTX Titan cards – but it would get the same NVENC benefits as the rest of the RTX lineup. The argument there is that you could use a single GTX 1660 Ti-powered PC to stream, getting better output and frame rates than what you would have otherwise gotten with the CPU-based x264 encoding method.

Part of this is also dependent on a NVIDIA-optimised beta build of Open Broadcaster Software, which was released earlier this month. It’s worth noting that the new NVENC implementation isn’t supported by Windows 7, according to the OBS patch notes. But the NVENC improvements will also benefit all GeForce GPUs from the 700 series and beyond.

More details can be found on the official website.