DLSS, FSR, and XeSS are upscaling technologies that render a game at a lower internal resolution and then reconstruct it to a higher output resolution, raising frame rate while approximating the image quality of native rendering. Nvidia DLSS uses the Tensor cores on RTX graphics cards, AMD FSR is an open technology that runs on graphics cards from any vendor, and Intel XeSS is optimized for Intel Arc cards while running more broadly on other hardware. This article defines upscaling, explains how each technology works, compares image quality and performance, lists hardware support, and covers the frame generation variants each one offers.
A comparison table sets the three side by side across method, hardware, and image quality. Upscaling lets a graphics card produce a high-resolution image from less rendering work, which raises frame rate at a given output resolution. Each section answers one question about the three technologies, building a complete comparison of how DLSS, FSR, and XeSS differ in approach, hardware requirements, and results.
What Are DLSS, FSR, and XeSS?
DLSS, FSR, and XeSS are upscaling technologies that render a game internally at a lower resolution and reconstruct it to a higher output resolution, increasing frame rate while approximating native image quality. Each technology takes a lower-resolution frame and fills in the missing detail to match a target output resolution, so the graphics card renders fewer pixels than the final image contains. The three technologies share one core function:
- The lower internal render draws each frame at a reduced resolution, cutting the pixel count the graphics card processes.
- The reconstruction rebuilds the frame to the target output resolution using motion vectors, prior frames, or trained models.
- The frame-rate gain follows from the reduced render work, since fewer rendered pixels free the graphics card to produce more frames.
Upscaling raises frame rate without lowering the output resolution, an alternative to reducing settings covered in the explanation of what causes FPS drops. The video memory each technology uses for its buffers relates to the capacity described in the explanation of VRAM.
How Does Nvidia DLSS Work?
Nvidia DLSS, Deep Learning Super Sampling, reconstructs a higher-resolution frame from a lower-resolution render using a trained neural network that runs on the Tensor cores of RTX graphics cards. According to Nvidia’s DLSS documentation, the technology uses motion vectors and prior frames with an AI model to produce an output frame that approximates native resolution. DLSS has three defining traits:
- The Tensor core requirement restricts DLSS to Nvidia RTX graphics cards, which contain the dedicated hardware the AI model runs on.
- The neural reconstruction uses a trained model and motion data to rebuild detail, according to Nvidia’s documentation.
- The quality modes trade internal resolution against frame rate, with Quality rendering closer to native and Performance rendering lower for more frames.
DLSS runs only on Nvidia RTX cards because it depends on Tensor cores, the hardware basis described in the explanation of how GPUs work. DLSS is frequently paired with ray tracing to recover the frame rate that ray tracing consumes, a technique covered in the explanation of ray tracing.
How Does AMD FSR Work?
AMD FSR, FidelityFX Super Resolution, is an open upscaling technology that reconstructs a higher-resolution frame from a lower-resolution render and runs on graphics cards from any vendor, since it does not require dedicated AI hardware. According to AMD’s FSR documentation, early FSR versions used spatial and temporal upscaling, while later versions added a machine learning model. FSR has three defining traits:
- The open compatibility lets FSR run on AMD, Nvidia, and Intel graphics cards, since it does not depend on vendor-specific hardware.
- The temporal reconstruction in current versions combines motion vectors and prior frames to rebuild detail, according to AMD’s documentation.
- The quality modes set the internal render resolution, trading image detail against frame-rate gain in the same manner as competing technologies.
FSR runs broadly because it does not require dedicated AI cores, which makes it available on hardware that cannot run DLSS. The open approach contrasts with the Tensor-core requirement of DLSS, a difference in hardware support compared in the table below and tied to the GPU architectures in the explanation of how GPUs work.
How Does Intel XeSS Work?
Intel XeSS, Xe Super Sampling, is an AI-based upscaling technology that reconstructs a higher-resolution frame from a lower-resolution render, running at full capability on Intel Arc graphics cards while supporting a broader path on other vendors’ hardware. According to Intel’s XeSS documentation, the technology uses a trained model that runs on Arc XMX cores at full quality and on a DP4a code path on other compatible cards. XeSS has three defining traits:
- The Arc optimization runs XeSS on the XMX matrix engines of Intel Arc cards for full performance, according to Intel’s documentation.
- The broad fallback uses a DP4a code path so XeSS also runs on many non-Arc graphics cards, with reduced acceleration.
- The AI reconstruction uses a trained model with motion vectors to rebuild detail toward the target output resolution.
XeSS reaches its highest quality on Intel Arc hardware while remaining usable on other cards through its fallback path, placing it between the RTX-only DLSS and the fully open FSR in hardware support. The matrix engines that accelerate XeSS on Arc relate to the GPU compute units in the explanation of how GPUs work.
What Are the Image Quality and Performance Differences?
The three technologies differ in image quality and performance because their reconstruction methods differ, with AI-based methods on dedicated hardware generally resolving fine detail and motion more cleanly than methods without dedicated acceleration. All three raise frame rate by rendering fewer pixels, but they vary in how cleanly the reconstructed frame matches native rendering. The differences appear in three areas:
- The detail reconstruction varies because AI models running on dedicated hardware can resolve fine edges and textures that simpler methods soften.
- The motion handling varies because temporal methods differ in how they treat fast motion, fine detail, and transparent surfaces.
- The performance gain is similar at matching quality modes, since each renders at a comparable internal resolution before reconstruction.
Image quality depends on the game’s implementation and version as much as the underlying technology, since each receives ongoing updates. The frame-rate gain from any of the three is one path to a target frame rate defined in the explanation of FPS in gaming, distinct from the frame generation that adds whole frames, covered in the explanation of frame generation.
Which Hardware and Games Support Each Technology?
DLSS requires an Nvidia RTX graphics card, FSR runs on graphics cards from any vendor, and XeSS runs at full quality on Intel Arc cards and through a fallback path on many others, with game support depending on developer integration. Hardware support is the clearest dividing line between the three, since DLSS is the most restricted and FSR the most open. The support differences are set out below:
| Technology | Vendor | Method | Hardware Support | Frame Gen Variant |
|---|---|---|---|---|
| DLSS | Nvidia | AI on Tensor cores | Nvidia RTX cards only | DLSS Frame Generation |
| FSR | AMD | Open temporal, later AI | Any vendor’s GPU | FSR 3 Frame Generation |
| XeSS | Intel | AI on XMX, DP4a fallback | Full on Arc, broad fallback | XeSS Frame Generation |
Game support is added by developers, so a given title may include one, two, or all three technologies depending on integration. The frame generation variants listed in the table raise the displayed frame rate by a different mechanism than upscaling, explained in the explanation of frame generation, and the upscaled frame rate pairs with the targets in the guide to a good FPS for gaming.
What Are the Frame Generation Variants?
Each upscaling technology has a frame generation variant that inserts additional generated frames between rendered ones to raise the displayed frame rate, distinct from the upscaling that rebuilds each rendered frame at a higher resolution. Frame generation and upscaling are separate techniques that combine in current versions, with upscaling lowering render cost and frame generation adding displayed frames. The variants work as follows:
- DLSS Frame Generation from Nvidia inserts generated frames using optical flow and motion data on supported RTX cards, according to Nvidia’s documentation.
- FSR 3 Frame Generation from AMD inserts generated frames and runs on a broad range of hardware, according to AMD’s documentation.
- XeSS Frame Generation from Intel inserts generated frames on supported hardware, complementing the XeSS upscaling path.
Frame generation raises the displayed frame rate but does not reduce input lag, a trade-off detailed in the explanation of frame generation. The added latency relates to the delay between input and display covered in the explanation of input lag in gaming.
Key Takeaways
- All three upscale by rendering at a lower internal resolution and reconstructing to a higher output, raising frame rate.
- DLSS uses Tensor cores on Nvidia RTX cards, restricting it to that hardware.
- FSR is open and runs on graphics cards from any vendor without dedicated AI hardware.
- XeSS is optimized for Intel Arc and runs more broadly through a DP4a fallback path on other cards.
- Image quality depends on method and implementation, with AI methods on dedicated hardware resolving detail more cleanly.
- Each has a frame generation variant that inserts generated frames, distinct from upscaling and without reducing input lag.
What is the difference between DLSS, FSR, and XeSS?
All three upscale from a lower internal resolution. DLSS runs on Nvidia RTX Tensor cores, FSR is open and runs on any GPU, and XeSS runs at full quality on Intel Arc with a fallback path elsewhere.
Which is better, DLSS or FSR?
DLSS often resolves fine detail more cleanly because it runs an AI model on dedicated Tensor cores, but it requires an Nvidia RTX card. FSR is open and runs on any GPU, making it more widely available.
Does FSR work on Nvidia cards?
Yes. AMD FSR is an open technology that runs on graphics cards from any vendor, including Nvidia and Intel, since it does not require vendor-specific AI hardware to reconstruct frames.
Does XeSS only work on Intel GPUs?
No. XeSS runs at full quality on Intel Arc cards using XMX engines, but a DP4a code path lets it run on many non-Arc graphics cards from other vendors with reduced acceleration.
Does upscaling reduce image quality?
Upscaling renders at a lower internal resolution, so it can soften fine detail compared with native rendering. Quality modes render closer to native, while performance modes favor frame rate over detail.
Is upscaling the same as frame generation?
No. Upscaling reconstructs each rendered frame at a higher resolution to raise frame rate, while frame generation inserts new generated frames between rendered ones. The two techniques combine in current versions.
Last Thoughts on DLSS vs FSR vs XeSS
DLSS, FSR, and XeSS all upscale from a lower internal resolution to raise frame rate, differing mainly in hardware support and reconstruction method: DLSS runs on Nvidia RTX Tensor cores, FSR is open across any vendor, and XeSS targets Intel Arc with a broad fallback. Image quality depends on the method and the game’s implementation, and each technology adds a frame generation variant that inserts displayed frames. Readers can continue with the explanation of frame generation, the explanation of VRAM, the explanation of ray tracing, or the PC gaming guide hub for related concepts.
