Switch 2 DLSS | How Tensor Core Upscaling Works on the NVIDIA C2 Chip

1. DLSS on Switch 2 | What the T239 C2 Chip Actually Does

The Switch 2 runs on the NVIDIA T239, a custom Tegra SoC built on the Ampere GPU architecture. The chip ships with 1,536 CUDA cores across 12 Streaming Multiprocessors, each containing second-generation Tensor cores. Those Tensor cores are the reason DLSS Super Resolution is possible.^[1]

DLSS Super Resolution is a two-pass process. The game engine renders a frame at a reduced internal resolution, typically between 540p and 810p depending on the game and DLSS quality preset. The Tensor cores then run a convolutional neural network inference pass, using temporal data from the previous four to eight frames to reconstruct missing pixel detail and output a sharp 1080p image to the dock. The inference pass takes under two milliseconds per frame, leaving the rest of the frame budget for GPU rendering work.^[2]

In handheld mode the math shifts. The target output drops to 720p, so Balanced mode renders internally at roughly 420p. A handful of less demanding titles opt for native 720p handheld without DLSS to minimize latency, but the majority of graphically intensive releases use DLSS across both modes.

2. Frame Generation | Why the T239 Cannot Support It

DLSS Frame Generation, introduced with RTX 40-series in late 2022, requires the fourth-generation Tensor cores and dedicated Optical Flow Accelerators found in the Ada Lovelace architecture and later. The T239's Ampere Tensor cores can execute the matrix math required for Super Resolution but do not include the Optical Flow hardware that makes Frame Generation reliable at low added latency.^[3]

There is also a thermal argument. Frame Generation synthesizes intermediate frames by running an additional GPU pass, which increases overall GPU load by approximately 30 to 40 percent above the base render. On a PC with a 300W total board power budget, that tradeoff is acceptable. On the Switch 2's 10 to 25W docked envelope, the GPU would thermally throttle before the synthesized frames could deliver a net frame rate benefit.

3. Switch 2 vs PS5 Pro | DLSS Super Resolution vs PSSR

The PlayStation 5 Pro uses Sony's PSSR (PlayStation Spectral Super Resolution), a proprietary ML upscaler. Both DLSS SR and PSSR are neural network upscalers, a meaningful distinction from the temporal-only FSR used in the original PS5 and the Steam Deck. The execution environments are radically different.^[4]

PSSR on the PS5 Pro runs on a GPU rated at 33.5 TFLOPS with 576 GB/s of memory bandwidth. DLSS SR on Switch 2 docked runs at approximately 1.5 to 2 TFLOPS with 102 GB/s of bandwidth. The raw gap is roughly 17x compute. Despite that gap, developers working across both platforms consistently report that the visual sharpness delta is smaller than the spec sheet implies, because DLSS Super Resolution has five years of training data over PSSR and produces fewer reconstruction artifacts at equivalent quality presets.

4. For Developers | DLSS SDK Integration on Switch 2

Nintendo distributes the DLSS SDK as part of the Switch 2 licensed developer package. Unreal Engine 5.4 and later includes a DLSS plugin that targets Switch 2 with minimal additional configuration required. Unity 6.1 added first-party Switch 2 DLSS support in its January 2026 update.^[5]

The integration path is straightforward: developers declare a render scale in engine settings and DLSS handles the upscale pass. The primary integration work is ensuring correctly formatted motion vectors and depth buffers are passed to the DLSS plugin, since missing or low-quality motion data produces ghosting on fast-moving objects. Studios porting existing console titles to Switch 2 report the DLSS integration averaging two to four days of engineering work per title.

For Unreal Engine 6 titles, DLSS integration is expected to be tighter still. Epic and NVIDIA co-developed the multi-platform DLSS pipeline as part of UE6's new rendering backend, which shares the same temporal upscaling pathway across PC, console, and handheld targets with platform-specific quality presets set at the engine level.