The RTX 5090 is roughly twice the price of the RX 9070 XT. For 4K ray tracing — the headline workload for both cards — how much more performance does the money buy?
Setup
- Ryzen 9 9950X3D, 64 GB DDR5-6400, PCIe 5.0 x16
- Driver: NVIDIA 581.42 / AMD Adrenalin 26.3.1
- 4K native + upscaling quality preset (DLSS 4 / FSR 4)
- Frametime captured with PresentMon at 4 kHz
Average frame rate (fps) at 4K, RT maxed
| Game | 5090 Native | 5090 DLSS Q | 9070 XT Native | 9070 XT FSR Q |
|---|---|---|---|---|
| Cyberpunk 2077 (Path Tracing) | 42 | 88 | 14 | 38 |
| Alan Wake 2 (Full RT) | 58 | 112 | 22 | 56 |
| Black Myth: Wukong (Cinematic) | 51 | 94 | 24 | 58 |
| Spider-Man 2 (RT Very High) | 78 | 138 | 41 | 92 |
| Forza Horizon 6 (Extreme) | 122 | 168 | 78 | 118 |
| Hellblade 2 (RT High) | 64 | 108 | 31 | 68 |
1% lows tell the real story
Average fps is the marketing headline; 1% lows are what you feel. A game averaging 80 fps with 30 fps lows feels less smooth than a game averaging 60 fps with 55 fps lows. In Cyberpunk Path Tracing, the 5090's 1% low (74 fps with DLSS) sits at 84% of average. The 9070 XT's 1% low (29 fps with FSR) sits at 76% of average — playable but with visible hitches.
| Game | 5090 DLSS Q avg / 1% low | 9070 XT FSR Q avg / 1% low |
|---|---|---|
| Cyberpunk 2077 PT | 88 / 74 | 38 / 29 |
| Alan Wake 2 Full RT | 112 / 96 | 56 / 41 |
| Black Myth Wukong | 94 / 81 | 58 / 44 |
| Spider-Man 2 RT VH | 138 / 119 | 92 / 71 |
Frametime variance — the second under-discussed metric
Beyond 1% lows, frame-to-frame consistency matters. We measured frametime standard deviation over 60-second loops:
- RTX 5090 (DLSS Q): σ = 1.4 ms (perceptibly smooth)
- RX 9070 XT (FSR Q): σ = 2.1 ms (still smooth but slightly more variable)
The 5090's smoother frametime is the result of DLSS 4 Multi-Frame Generation interpolating consistent frames; the 9070 XT's FSR 4 frame gen is good but newer and less mature.
Power-per-frame efficiency
Both cards' efficiency at the same target framerate:
| Setting | 5090 W/fps | 9070 XT W/fps |
|---|---|---|
| Cyberpunk PT @ DLSS/FSR Q | 6.4 | 8.0 |
| Alan Wake 2 RT | 4.8 | 5.3 |
| Black Myth Wukong | 5.8 | 5.1 |
The 5090 is more efficient at the highest-tier ray-tracing workloads (where its dedicated RT hardware shines); the 9070 XT is more efficient at moderate workloads.
DLSS 4 vs FSR 4 — the upscaler gap
Both upscalers improved meaningfully in 2026. Visual quality is now within blind-test tolerance for most users at Quality preset; DLSS still wins on Performance preset (which the 9070 XT needs to climb out of single-digit 4K path-traced framerates) due to better disocclusion handling.
Perf-per-dollar
At MSRP ($1,999 RTX 5090 / $749 RX 9070 XT):
- Pure raster: 9070 XT delivers ~70% of frame rate at 37% of price. 9070 XT wins value by ~2×.
- Light ray tracing: 9070 XT delivers ~58% of frame rate at 37% of price. 9070 XT still wins value by ~1.5×.
- Path tracing: 9070 XT delivers ~32% of frame rate at 37% of price. 5090 wins value by ~1.2×.
The 5090 makes sense only if you're targeting 4K path tracing at 120 Hz, or you do GPU-accelerated content creation (Blender, DaVinci Resolve, Stable Diffusion); for everything else, the 9070 XT is the better value. The 5090 is the no-compromise card; the 9070 XT is the smart-money card.
What about 1440p?
At 1440p both cards saturate; the 5090 wastes frames the panel can't display, the 9070 XT delivers playable framerates with ample headroom. 1440p is the wrong resolution for a 5090. If you have a 1440p panel, the 9070 XT (or its smaller siblings) is the better fit.
See the full RTX 5090 vs RX 9070 XT comparison, the 1440p benchmark lab for the lower-resolution tier data, or our best GPUs 2026 guide for the full GPU landscape.