In fact, without DLSS, ray tracing simply wouldn’t be worth it. With DLSS, you’re able to make up much of the performance difference that ray tracing brings to the table, so you can have the awesome image quality benefits of ray tracing, without sacrificing too much in the way of performance.
This helps boost performance by having the SM render a scene in a lower resolution, then having the Tensor Core use data from Nvidia’s Supercomputers to intelligently scale that scene up to full resolution. Through the Tensor Core, developers can integrate DLSS in games, which is essentially an AI-enabled upscaling tech.
Luckily, the other dedicated type of core, the Tensor Core is the other part of the equation that makes it somewhat sensible to enable ray tracing when playing the best PC games. However, without these dedicated RT cores, the performance hit would turn the game into a slideshow.
This doesn’t make ray tracing computationally free, though, as you’ll see from the massive hit in performance that having RT on in Metro Exodus entails. The whole point is to offload the massive computational load that any kind of ray tracing entails, in order to be able to produce playable framerates.īasically, when a light ray is cast in a game engine, the SM will send that information over to the RT core, where it will calculate where that ray bounces, and report that back to the SM so it can render the image. The RT core is what’s responsible for the ray tracing that everyone is talking about. The dedicated among you might have noticed that’s half the amount of Tensor Cores in each SM than Turing had, but there’s a good reason for that – they’re more than twice as fast this time around. Each SM has one RT core and four Tensor Cores. Outside of CUDA cores, each SM also has Tensor Cores for AI workloads like Deep Learning Super Sampling (DLSS), and dedicated ray tracing cores. With Ampere, and thus the RTX 3060 Ti, Nvidia was able to engineer the SM to have one of the datapaths essentially able to pull double duty, which is why the amount of CUDA cores has effectively doubled per SM with Ampere. It’s more, but it shouldn’t be so much more that you’ll need to worry about upgrading your power supply to accommodate – assuming you have a competent one in the first place, that is.īack in Turing, each SM had two datapaths – one dedicated to Floating Point (FP32) workloads, and the other dedicated to Integer. The total graphics power (TGP) is 200W for the RTX 3060 Ti, up from the 175W of the RTX 2060 Super. Power consumption sees an increase over the RTX 2060 Super, but it’s not as pronounced as other cards in the Nvidia Ampere lineup.
That means there are now 4,864 CUDA cores in the RTX 3060 Ti, up from the 2,176 found in the RTX 2060 Super. However, because of changes Nvidia has made to its SM since Turing, each of these now has 128 CUDA cores, double that of the 64 in each Turing SM. With this graphics card, you’re getting 8GB of the same GDDR6 memory found in the RTX 3070, paired with 38 Ampere Streaming Multiprocessors (SM). This isn’t exactly going to be the graphics card that’ll push Watch Dogs Legion with ray tracing at 4K, but it is definitely up to doing it at 1080p. The Nvidia GeForce RTX 3060 Ti is based on the same Ampere architecture as the RTX 3080, and gives us a look at what the Ampere lineup will look like in the mid-range and budget segment of the GPU market. (Image credit: Nvidia) Features and chipset