DLSS vs FSR vs XeSS: AI Upscaling Explained
A complete comparison of NVIDIA DLSS 4, AMD FSR 4, and Intel XeSS. How they work, which looks best, and which GPU you need for each technology.
What is AI upscaling?
AI upscaling renders a game at a lower internal resolution and uses machine learning to reconstruct a higher-resolution image. Instead of brute-forcing every pixel at 4K, the GPU renders at 1080p or 1440p and an AI model fills in the missing detail. The result: near-native image quality at dramatically higher frame rates.
All three major GPU manufacturers now offer their own version: NVIDIA's DLSS (Deep Learning Super Sampling), AMD's FSR (FidelityFX Super Resolution), and Intel's XeSS (Xe Super Sampling). Each approaches the problem differently, and the quality gap between them has narrowed significantly with their latest versions.
NVIDIA DLSS 4: Multi Frame Generation
DLSS 4, available on RTX 50 series GPUs, is NVIDIA's most ambitious upscaling technology yet. It introduces Multi Frame Generation, which can generate up to three additional frames for every traditionally rendered frame. Combined with DLSS Super Resolution (the upscaler) and DLSS Ray Reconstruction (AI-enhanced ray tracing denoising), it forms a full rendering pipeline.
How it works: DLSS uses dedicated Tensor Cores on NVIDIA GPUs to run a trained neural network. The model is trained on pairs of low-res and high-res images, learning to predict what the high-res version should look like. Because the AI runs on dedicated hardware, it doesn't steal compute resources from the main GPU workload.
Strengths:
- Best image quality of the three, especially in motion
- Multi Frame Generation dramatically boosts FPS (2-4x in supported titles)
- Ray Reconstruction improves RT quality while reducing its performance cost
- Dedicated Tensor Core hardware means minimal overhead
Limitations:
- Requires NVIDIA RTX hardware (no AMD or Intel support)
- Multi Frame Generation adds slight latency (mitigated by Reflex)
- Game developers must integrate it (not a universal driver feature)
AMD FSR 4: Going fully neural
FSR 4 represents AMD's biggest leap yet. Unlike FSR 3 which used a spatial-temporal algorithm, FSR 4 is a fully machine-learning-based upscaler that runs on the dedicated AI accelerators in RDNA 4 GPUs. This brings it architecturally closer to DLSS while maintaining AMD's commitment to broader compatibility.
How it works: FSR 4 uses AMD's AI accelerators (new to RDNA 4) to run a neural network for upscaling. On older AMD GPUs or non-AMD hardware, FSR 3.1 (the algorithmic version) remains available. FSR 4 also includes Frame Generation, though limited to one generated frame per rendered frame (vs DLSS 4's three).
Strengths:
- Massive quality improvement over FSR 3 — competitive with DLSS in many scenarios
- FSR 3.1 fallback works on any GPU (NVIDIA, Intel, even older AMD cards)
- Open source, encouraging broader adoption
- Frame Generation available on RDNA 2 and newer
Limitations:
- Full FSR 4 quality requires RDNA 4 hardware (RX 9070 series and up)
- Frame Generation limited to 1 extra frame (vs DLSS 4's up to 3)
- Slightly behind DLSS 4 in fine detail preservation in side-by-side comparisons
Intel XeSS: The underdog option
Intel's XeSS uses XMX (Xe Matrix eXtensions) AI cores on Arc GPUs for hardware-accelerated upscaling, with a DP4a shader fallback for non-Intel hardware. It's the youngest of the three technologies, but Intel has been iterating rapidly.
Strengths:
- Surprisingly good quality on Arc hardware, especially in XeSS 2
- Shader-based fallback works on any GPU
- Low latency — no frame generation means no extra input lag
Limitations:
- Fewest supported titles of the three
- No frame generation technology (yet)
- Quality on non-Intel hardware is noticeably worse than the hardware-accelerated path
Which should you choose?
Your GPU dictates your best option — you don't get to pick freely.
- NVIDIA RTX 50 series: Use DLSS 4 wherever available. It's the best upscaling technology in 2026. Fall back to FSR 3.1 in games without DLSS support.
- NVIDIA RTX 40 series: Use DLSS 3.5 (Super Resolution + Frame Generation). You won't get Multi Frame Generation but DLSS quality is still excellent.
- AMD RX 9000 series: Use FSR 4 wherever available. In DLSS-only games, you're limited to native resolution, though many developers add FSR alongside DLSS.
- AMD RX 7000 series: FSR 3.1 with Frame Generation. Quality is solid but clearly behind DLSS and FSR 4.
- Intel Arc: XeSS in supported games, FSR 3.1 elsewhere. The Arc B580 is an excellent budget card that benefits greatly from XeSS.
In practice, most major releases in 2026 support both DLSS and FSR, so your choice of GPU brand doesn't lock you out of upscaling. The difference is quality: DLSS 4 > FSR 4 > FSR 3.1 > XeSS, with the gap between DLSS 4 and FSR 4 being smaller than previous generations.
Frequently Asked Questions
Does DLSS work on AMD GPUs?
No. DLSS requires NVIDIA RTX hardware with Tensor Cores. AMD GPUs use FSR instead. However, most games that support DLSS also support FSR, so AMD users have a comparable (if slightly lower quality) alternative.
Does AI upscaling add input lag?
The upscaling itself adds negligible latency (under 1ms). Frame generation does add latency because generated frames are inserted after the fact. NVIDIA mitigates this with Reflex, and AMD uses Anti-Lag+. In practice, the perceived smoothness from higher frame rates offsets the added latency for most players.
Is FSR 4 as good as DLSS 4?
FSR 4 is much closer to DLSS 4 than any previous FSR version. In side-by-side comparisons, DLSS 4 still has an edge in fine detail and temporal stability, but FSR 4 is a massive improvement over FSR 3 and delivers excellent quality on RDNA 4 hardware.
Can I use DLSS at 1080p?
Yes, but the quality trade-off is more noticeable at 1080p because the internal render resolution drops to 720p or lower. AI upscaling delivers its best results when upscaling from 1080p to 4K, where the ratio of rendered to output pixels allows the AI more room to work.
Related Articles
Blackwell vs RDNA 4 vs Battlemage: GPU Architecture Compared
A deep technical comparison of NVIDIA Blackwell, AMD RDNA 4, and Intel Battlemage GPU architectures. What's new, what matters, and which to buy.
CPU Bottleneck Explained: How to Identify and Fix It
Learn what a CPU bottleneck is, how to tell if your processor is holding back your GPU, and the best ways to fix it without overspending.
PCIe 5.0 vs PCIe 4.0: Does It Matter for Gaming?
Does PCIe 5.0 improve gaming performance over PCIe 4.0? We test the real-world difference and explain when PCIe gen actually matters.