

NVIDIA GeForce RTX 5090 wins 7 of 9 benchmarks, averaging 255.4% faster.
Both cards were measured first-party on our bench, same suite, same test rig.
The gap is widest in Stable Diffusion XL, where NVIDIA GeForce RTX 5090 leads by 461% (3.76 vs 21.08 images/min); the closest fight is Qwen3 4B (143% apart); VRAM decides part of this one: NVIDIA GeForce RTX 5090 runs 7 of our 12 AI workloads while the other card runs 3, models that don't fit score zero.
| Benchmark | NVIDIA GeForce RTX 3070 Ti | NVIDIA GeForce RTX 5090 | Difference |
|---|---|---|---|
| Qwen3 4B tok/s | 154.65 | 375.6 | -59% |
| Llama 3.1 8B tok/s | 102.13 | 268.14 | -62% |
| Qwen2.5-Coder 14B tok/s | 0 | 149.75 | n/a |
| Qwen3 32B tok/s | 0 | 71.15 | n/a |
| Llama 3.3 70B tok/s | 0 | 0 | n/a |
| Stable Diffusion XL images/min | 3.76 | 21.08 | -82% |
| Z-Image Turbo images/min | 0 | 10.725 | n/a |
| FLUX.1 dev images/min | 0 | 2.036 | n/a |
| Qwen-Image-Edit images/min | 0 | 0 | n/a |
| NVIDIA GeForce RTX 3070 Ti | NVIDIA GeForce RTX 5090 | |
|---|---|---|
| VRAM | 8GB | 32GB |
| Transistors | 17,400M | 92,200M |
| Die size | 392.5 mm² | 750 mm² |
| Transistor density | 44.3 M/mm² | 122.9 M/mm² |
| Architecture | Ampere (GA104) | Blackwell (GB202) |
| Memory bandwidth | 608 GB/s | 1792 GB/s |
| Boost clock | 1,770 MHz | 2,407 MHz |
| TDP | 290 W | 575 W |
| Launch MSRP | $599 | $1,999 |
| Release | 2021-06-10 | 2025-01-30 |
NVIDIA GeForce RTX 3070 Ti full review · NVIDIA GeForce RTX 5090 full review · All AI & Machine Learning rankings