ASUS Phoenix GTX 1660Ti Review
Published: 22nd February 2019 | Source: ASUS | Price: |
GTX 1660Ti Introduction
With the recent launch of the Turing GPUs Nvidia took the usual improvements in performance that one expects to find on the next generation of technology, and apply all that power towards higher framerates. So far, so typical. However, the RTX range also gained two really useful bits of unique hardware. Those were the RT Cores which were dedicated ray tracing cores that handled light and wowed us with jaw-dropping visuals in everything that took advantage of them, and Tensor Cores which utilised a form of AI to enhance the supersampling and improve antialiasing performance without actually costing you those vital frames per second. In fact, the Tensor Cores - named after the mathematical tensor geometric objects which map vectors to scalars - have so much latent calculative capability that offloading your supersampling to them via Nvidia's own DLSS improves performance when compared to having no supersampling at all. A massively impressive feat.
All this hardware does come at a price though. The current base model RTX 2060 is very affordable starting at around £330 for the most stripped down version, moving up to £430 for the various partner vendors all-singing and dancing models. Still, at this end of the market when budgets are tight, twenty or thirty quid here and there can make all the difference. Enter the GTX 1660Ti. A card which is intended by Nvidia to replace the GTX 1060 by utilising the latest Turing shaders, but without the aid of their Tensor and RT Cores.
The key features of the Turing GPU on the GTX 1660Ti are greatly increased Floating Point Operations Per Second (FLOPS, measured in TeraFLOPS), an equally large buff to the regular Integer Operations Per Second (TOPS, Tera again), and perhaps most importantly the ability to handle TOPS and TFLOPS at the same time. Previously on the Pascal architecture, the GPU could perform one of these tasks at once. For example, something that had 38 Integer calculations and 62 Floating Point calculations per 100 instructions would require a full 100 instructions worth of time to execute. Now, with the concurrent capabilities of the Turing GPU it can handle Floating Point and Integer calculations at the same time, thus taking only 62 instructions worth of time to calculate them all. Still with us? The long and short of that is that it's much quicker than it's Pascal predecessor. Faster calculations equal higher frame rates. 1.5x quicker to be exact than the GTX 1060 to be precise.
It isn't just complicated architecture improvements that the TU116 GPU scores over the GP106. Boost clocks are improved, Memory bandwidth has stepped up from 192 GB/s to 288 GB/s, and a whole host of extra CUDA Cores are available, now 1536 up from the 1280 CUDA cores of the GTX 1060. To fit all that onto a PCB the manufacturing process has reduced from 16nm on the Pascal to 12nm on the Turing, whilst the transistor count has blossomed from 4.4 billion to 6.6 billion.
We have three different models in for review at launch, the two premium models from MSI and ASUS, the Gaming X and Strix, which are very much at the high end of the GTX 1660Ti cost, and a baseline model that shows how affordable the GTX 1660Ti can be if you don't want any bells and whistles.
While the pricing of the Strix and Gaming X start to muddy the waters a little as they are pretty much the same as a baseline RTX 2060, the ASUS Phoenix is significantly more affordable. Does the no-frills approach still bring rich rewards? Let's find out.
The Phoenix is about the smallest amount of card you can buy and still get a GTX 1660Ti. It is a card built for those with extremely tight budgets who are looking to upgrade an ageing card to the latest technology without breaking the bank. Perfect if your funds are limited.