Upscaling
Of course, what would a review of a new GPU be without talking about the ever present buzz word, AI, more specifically, AI upscaling, in the form of DLSS and frame generation, or more importantly in the case of the 5060 Ti, multi-frame generation and starting things off with Alan Wake 2, where we can see at 1440p there is a pretty impressive jump over the native performance using DLSS on balanced mode of 88%, which is approaching double the frame rate, but of course this does come with the downside of lower visual fidelity overall. When we enable frame generation, performance predictably increases further, now giving us 171% more performance than we saw at native, with a 17% lower latency to boot. Enabling Multi frame generation and setting it to 3X increases the performance even more, but what is most impressive to me here is at 4X where we can see the latency reaching an identical level as we saw at native, meaning that we’re getting significantly higher frame rates at the same latency so win win, as long as the visuals still look good.
Enabling ray tracing sees pretty much the same kind of thing, with DLSS coming in with over double the performance that we saw at native and each subsequent frame generation setting largely improving the frame rate whilst the latency sees an incremental increase as well, making for a slightly worse experience, but nothing to worry about really considering that the latency is actually much lower than what we saw at native which is something we tend to see with raytracing turned on, that latency does increase ten-fold, though with a game like Alan Wake II, latency isn’t the most important thing, as it’s not a competitive shooter or similar, and instead, visuals while still getting a playable frame rate are arguably more important.
As we move onto Cyberpunk, things are a bit different, as we now have the option for different models to choose from, with CNN being the original model, and transformer now being the newer model, and there’s an argument for using one over the other. The transformer generally will look better than CNN, but CNN will typically give slightly higher performance, which is exactly what we can see across the board with the CNN model, no matter what setting, giving us around another 5 to 6% more performance over the equivalent Transformer version. What we do see from native though, is again, a reduction in latency compared to native, sometimes by around half, when looking at frame generation on x2 for instance on both transformer and CNN models while also increasing the performance by upwards of 125%, though the best increase we saw was using frame generation x4 CNN which came in at 233 frames per second, giving us a 264% increase in performance while reducing latency by 29ms.
When we enable raytracing, latency increases come into play naturally, but we’re still able to see some good reductions, such as DLSS on both models increasing performance by over double, while almost halving the latency, especially in the case of the transformer model which manages to shave off a pretty staggering 49ms of latency. Again, if performance is your bag, then the X4 CNN model is the one to go for, which again, outperforms the transformer model but in this case, does introduce a lot more latency compared to native, and instead, I’d be more inclined to utilise the transformer model which not only increases performance, but reduces latency by around 20ms.
