GeForce RTX 5090 vs Radeon RX 9070 XT
GeForce RTX 5090
Winner98#1 of 3 desktop GPUs · top 34%
Fastest consumer GPU by a wide margin; 575 W and $1,999 say it all.
vs
GeForce RTX 5090 wins by 33 points
Radeon RX 9070 XT
65#2 of 3 desktop GPUs · bottom 67%
Beats the RTX 5070 in raster with 16 GB; ray tracing finally competitive.
Overview
Key Differences
GeForce RTX 5090
- ≈ estimated268% higher Blender 4.x samples per minute: 12,500 vs 3,400
- ≈ estimated127% higher 3DMark Speed Way (ray tracing): 14,500 vs 6,400
- ≈ estimated126% higher 3DMark Port Royal (ray tracing): 35,000 vs 15,500
- ≈ estimated59% higher 3DMark Time Spy graphics: 47,000 vs 29,500
- ≈ estimated56 points higher Ray tracing score: 100 vs 44
Radeon RX 9070 XT
- $1,400 cheaper at launch: $599 vs $1,999
- 47% lower rated power: 304 W vs 575 W
- ≈ estimated8 points higher Efficiency score: 100 vs 92
Review
GeForce RTX 5090 scores 98 overall to Radeon RX 9070 XT's 65: #1 against #2 of 3 desktop GPUs.
GeForce RTX 5090 leads on Raster by 59% (100 (estimated) vs 63 (estimated)) and Ray tracing by 127% (100 (estimated) vs 44 (estimated)).
Radeon RX 9070 XT leads on Efficiency by 9% (100 (estimated) vs 92 (estimated)).
GeForce RTX 5090 is ahead in 4 of 4 benchmarks, the widest gap in Blender 4.x samples per minute by 268% (12,500 (estimated) vs 3,400 (estimated)).
GeForce RTX 5090 is rated at 575 W against 304 W; launch prices $1,999 and $599.
- Raster≈ estimated100 +59%≈ estimated63Traditional rendering: most games at high frame rates.
- Ray tracing≈ estimated100 +127%≈ estimated44Ray-traced lighting and reflections.
- Efficiency≈ estimated92≈ estimated100 +9%Raster performance per watt of board power.
- Overall9865
Benchmarks
- 3DMark Time Spy graphics≈ estimated47,000 +59%≈ estimated29,500
- 3DMark Speed Way (ray tracing)≈ estimated14,500 +127%≈ estimated6,400
- 3DMark Port Royal (ray tracing)≈ estimated35,000 +126%≈ estimated15,500
- Blender 4.x samples per minute≈ estimated12,500 +268%≈ estimated3,400
Specs
| Spec | GeForce RTX 5090 | Radeon RX 9070 XT |
|---|---|---|
| CPU | ||
| Architecture | Blackwell (GB202) | RDNA 4 (Navi 48) |
| Shaders | 21,760 | — |
| RT cores | 170 (4th gen) | — |
| Tensor cores | 680 (5th gen) | — |
| Base / boost | 2.01 / 2.41 GHz | — |
| Stream processors | — | 4,096 |
| Ray accelerators | — | 64 (3rd gen) |
| AI accelerators | — | 128 |
| Game / boost | — | 2.40 / 2.97 GHz |
| GPU | ||
| L2 cache | 96 MB | — |
| ROPs | 176 | 128 |
| TMUs | 680 | 256 |
| Infinity Cache | — | 64 MB |
| Memory | ||
| Type | GDDR7 | GDDR6 |
| Capacity | 32 GB | 16 GB |
| Bus width | 512-bit | 256-bit |
| Bandwidth | 1,792 GB/s | 640 GB/s |
| Connectivity | ||
| Interface | PCIe 5.0 x16 | PCIe 5.0 x16 |
| Outputs | 3× DP 2.1b, 1× HDMI 2.1b | 3× DP 2.1a, 1× HDMI 2.1b |
| Power connector | 1× 16-pin | 2× 8-pin |
| Power | ||
| Total board power | 575 W | 304 W |
| Recommended PSU | 1000 W | 750 W |
| Misc | ||
| Node | TSMC 4N | TSMC N4P |
| Launched | January 2025 | March 2025 |
| Launch price | $1,999 | $599 |