12GB · AI Score 3.1/100 · first-party measured on 12 AI workloads
3.1 AI Score ✓ Measured
Every number on this page is first-party: NVIDIA GeForce RTX 4070 was run on our pinned 12-workload AI suite on 2026-07-11, with under 0.5% run-to-run variance. On Llama 3.1 8B (Q4_K_M) NVIDIA GeForce RTX 4070 delivers about 93.16 tokens/sec. Llama 3.3 70B does not fit. It needs roughly 42GB and this card has 12GB. 8 of the 12 workloads won't fit on 12GB at the tested precision, Qwen3 32B, Llama 3.3 70B, Z-Image Turbo, FLUX.1-dev and others. We publish those as hard gates rather than quietly dropping to a smaller quant.
Most efficient card in its class that I measured, 1.36 tokens/watt on Qwen3 4B, and it never broke 57°C on anything. But the 12GB ceiling is real: 8 of my 12 workloads simply don't fit. Great small-LLM card, wrong card for image/video generation. Quick note on the setup: all my AI benchmarking was done on rented cloud GPUs, I used all three of Vast.ai, RunPod and Modal depending on which had the card, and they all have their pros and cons. Same pinned harness on every run, and everything here got double-checked before it went up.
| Workload | Result | Telemetry | Data |
|---|---|---|---|
| Qwen3 4B | 149.68 tok/s | 2.6 GB peak110 W53°C1.36 tok/WQ4_K_M | ✓ Measured |
| Llama 3.1 8B | 93.16 tok/s | 4.8 GB peak125 W54°C0.75 tok/WQ4_K_M | ✓ Measured |
| Qwen2.5-Coder 14B | 50.74 tok/s | 8.6 GB peak131 W57°C0.39 tok/WQ4_K_M | ✓ Measured |
| Qwen3 32B | ✕ Won't fit needs ~23 GB | VRAM-gated at this precision | ✓ Measured |
| Llama 3.3 70B | ✕ Won't fit needs ~46 GB | VRAM-gated at this precision | ✓ Measured |
| Workload | Result | Telemetry | Data |
|---|---|---|---|
| Z-Image Turbo | ✕ Won't fit needs ~13 GB | VRAM-gated at this precision | ✓ Measured |
| FLUX.1 dev | ✕ Won't fit needs ~26 GB | VRAM-gated at this precision | ✓ Measured |
| Workload | Result | Telemetry | Data |
|---|---|---|---|
| FLUX.1 Kontext dev | ✕ Won't fit needs ~26 GB | VRAM-gated at this precision | ✓ Measured |
| Qwen-Image-Edit | ✕ Won't fit needs ~42 GB | VRAM-gated at this precision | ✓ Measured |
| Workload | Result | Telemetry | Data |
|---|---|---|---|
| LTX-Video (distilled) | ✕ Won't fit needs ~14 GB | VRAM-gated at this precision | ✓ Measured |
| Wan 2.2 5B (720p) | ✕ Won't fit needs ~18 GB | VRAM-gated at this precision | ✓ Measured |
| Architecture | Ada Lovelace |
| CUDA cores | 5,888 |
| VRAM | 12GB GDDR6X |
| Memory bus | 192-bit |
| Memory bandwidth | 504 GB/s |
| Boost clock | 2,475 MHz |
| TDP | 200 W |
| Process | 4nm (TSMC 4N) |
| Interface | PCIe 4.0 x16 |
| Release date | 2023-04-13 |
| Launch MSRP | $599 |
NVIDIA GeForce RTX 4070 scores 3.1/100, #66 of 102. It ran 3 of 12; 8 exceeded its 12GB. Every figure here is our own measurement.
100% = this card, AI & Machine Learning headline metric (AI Score). #29 of 61 desktop cards in this vertical.
| GPU | Relative | % | AI Score |
|---|---|---|---|
| AMD Radeon RX 7800 XT | 106% | 3.3 | |
| NVIDIA GeForce RTX 4060 Ti 16GB | 103% | 3.2 | |
| GeForce RTX 5070 | 103% | 3.2 | |
| NVIDIA GeForce RTX 4070 Super | 100% | 3.1 | |
| NVIDIA GeForce RTX 4070 | 100% | 3.1 | |
| NVIDIA GeForce RTX 4070 Ti | 97% | 3 | |
| Intel Arc A770 Limited Edition | 94% | 2.9 | |
| NVIDIA TITAN V | 94% | 2.9 | |
| GeForce GTX 1080 Ti | 84% | 2.6 |
Same card, other workloads: NVIDIA GeForce RTX 4070 Gaming benchmarks
← All AI & Machine Learning GPU rankings
| Transistors | 35,800 million |
| Die size | 294.5 mm² |
| Process node | 4 nm |
| Fabricated by | TSMC |
| Transistor density | 121.6 million per mm² |
Denser than 91% of the 76 cards we have silicon data for. Density is the clearest measure of what a process node bought: a card that gained it without growing the die got its speed from the fab rather than the architecture.
Silicon figures from Wikipedia (CC BY-SA 4.0). Benchmarks on this page are our own. Compare every chip.
Whole-job timings, composed from our measured per-model results on this card.
| Workflow | Time | Energy | Basis |
|---|---|---|---|
| Full codebase review | 19.7 min | 43.13 Wh | measured |
Can't run: 60-second AI short film (needs Qwen3 32B), 60-second AI short film, narrated (needs Qwen3 32B), 10 short social clips (needs Qwen3 32B), 6-panel comic page (needs Qwen3 32B), 20 long-form articles (needs Llama 3.3 70B), Character sheet, 12 poses (needs FLUX.1 dev), 100-photo restoration batch (needs FLUX.1 Kontext dev), 24-frame storyboard (needs Qwen3 32B), 100-photo restore and enlarge (needs FLUX.1 Kontext dev).
This card is $599 to buy. The cheapest listed rate on Vast.ai is $0.076/hour, but that is the floor: we budget $0.091/hour, a 20% premium, because idle time, storage and unavailable cheap instances all land on the same bill. At that rate buying wins after 6,568 GPU-hours. Below it you are paying for idle silicon.
| How you would use it | GPU-hours a year | Rental cost a year | Time to break even |
|---|---|---|---|
| 2 hours a day, hobby | 730 | $67 | 9.0 years |
| 8 hours a day, working on it | 2,920 | $266 | 2.2 years |
| 24/7, always-on agent | 8,760 | $799 | 9.0 months |
At hobby usage this card is very unlikely to pay for itself before it is superseded. Rent it. Rental figures include a 20% premium over the cheapest listed rate. Ignores electricity, resale and the fact that a rented card can be a newer one tomorrow.
Cheapest of the RunPod and Vast on-demand rates we see, sampled daily. Spot and interruptible pricing runs lower.