NVIDIA GeForce RTX 5070 Ti vs AMD Radeon RX 9070 XT
Graphics card comparison for gaming — benchmark scores, estimated framerates and which one suits your build.
The short answer
Closely matched. The NVIDIA GeForce RTX 5070 Ti and AMD Radeon RX 9070 XT sit in the same performance class. Our framerate data has both cards limited by the processor rather than themselves at these settings, so we will not put a precise figure on the gap — the benchmark score suggests NVIDIA GeForce RTX 5070 Ti is ahead, but not by a margin we can stand behind for cards this fast.
Separating cards this fast needs 4K framerate data, which our model does not yet cover — it holds 720p, 1080p and 1440p only.
Benchmark scores
| Measure | NVIDIA GeForce RTX 5070 Ti | AMD Radeon RX 9070 XT |
|---|---|---|
| G3D Mark | 32,360 | 26,921 |
| Faster than | 98% of 268 | 93% of 268 |
| 3DMark Time Spy | 27,727 | 29,464 |
| 3DMark Speed Way | — | 6,345 |
| Memory | 16 GB GDDR7 | 16 GB GDDR6 |
| Memory bus | 256-bit | 256-bit |
| Bandwidth | 896 GB/s | 640 GB/s |
| Boost clock | 2452 MHz | 2970 MHz |
| FP32 | 44.0 TFLOPS | 48.7 TFLOPS |
| Board power | 300 W | 304 W |
| Suggested PSU | 750 W | 750 W |
| Power connectors | 1x 16-pin (12V-2x6) | 2x 8-pin |
| Architecture | Blackwell | RDNA 4 |
| Process | 4 nm | 4 nm |
| Launch price | $749 | $599 |
| Released | 2025-02-20 | 2025-03-06 |
Estimated FPS at 1440p
Both measured against the same reference processor — a AMD Ryzen 7 9800X3D — so the graphics card is the variable being tested. Ranges, not exact figures. Measured at 1440p: at 1080p even a fast reference processor caps both cards, and the two columns come back identical.
| Game | NVIDIA GeForce RTX 5070 Ti | AMD Radeon RX 9070 XT |
|---|---|---|
| Fortnite | 272–453 CPU | 272–453 CPU |
| CS:GO | 413–688 CPU | 413–688 CPU |
| GTA V | 145–242 CPU | 145–242 CPU |
| PUBG | 170–284 CPU | 170–284 CPU |
Which suits your build?
These figures hold a reference processor constant. Your own build decides the answer: a faster graphics card buys nothing if the other component is already the limit. Run both against your actual hardware:
Check both in the bottleneck calculator