16GB · AI Score 4.7/100 · first-party measured on 12 AI workloads
4.8 AI Score Includes estimates
Every number on this page is first-party: NVIDIA GeForce RTX 4080 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 4080 delivers about 127.09 tokens/sec. Llama 3.3 70B does not fit. It needs roughly 42GB and this card has 16GB. For image generation, SDXL runs at 5.08 it/s, while FLUX.1-dev won't fit at BF16 (needs ~26GB). 6 of the 12 workloads won't fit on 16GB at the tested precision, Qwen3 32B, Llama 3.3 70B, FLUX.1-dev, FLUX.1 Kontext and others. We publish those as hard gates rather than quietly dropping to a smaller quant.
Quietly one of the most efficient cards I measured, 1.77 tokens/watt, never over 65°C, ran at 96% of its rating on SDXL. The 16GB gets you 6 of 12 workloads; the FLUX-class models still say no. 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 0.6B | 677.37 tok/s | 67 W40°CQ4_K_M | ✓ Measured |
| Llama 3.2 1B | 602.28 tok/s | 88 W47°CQ4_K_M | ✓ Measured |
| MiniCPM5 2B | 312.75 tok/s | 108 W41°CQ4_K_M | ✓ Measured |
| LFM2.5 2.6B | 307.16 tok/s | 119 W42°CQ4_K_M | ✓ Measured |
| Granite 4.1 3B | 226.16 tok/s | 126 W43°CQ4_K_M | ✓ Measured |
| Nemotron 3 Nano 4B | 198 tok/s | 136 W43°CQ4_K_M | ✓ Measured |
| Qwen3 4B | 201.76 tok/s | 2.7 GB peak114 W48°C1.77 tok/WQ4_K_M | ✓ Measured |
| DeepSeek Coder 7B Instruct v1.5 | 143.42 tok/s | 159 W45°CQ4_K_M | ✓ Measured |
| Qwen2.5-7B | 134.28 tok/s | 179 W51°CQ4_K_M | ✓ Measured |
| Qwen2.5-Coder 7B | 134.36 tok/s | 183 W50°CQ4_K_M | ✓ Measured |
| Llama 3 8B | 127.3 tok/s | 162 W45°CQ4_K_M | ✓ Measured |
| Llama 3.1 8B | 127.09 tok/s | 4.9 GB peak166 W52°C0.76 tok/WQ4_K_M | ✓ Measured |
| Qwen3 8B | 123.68 tok/s | 171 W50°CQ4_K_M | ✓ Measured |
| Nemotron Nano 9B v2 | 91.27 tok/s | 164 W44°CQ4_K_M | ✓ Measured |
| Ornith 1.5 9B | 109.71 tok/s | 169 W44°CQ4_K_M | ✓ Measured |
| Gemma 4 12B | 78.94 tok/s | 175 W51°CQ4_K_M | ✓ Measured |
| Qwen2.5-Coder 14B | 69.93 tok/s | 8.6 GB peak188 W56°C0.37 tok/WQ4_K_M | ✓ Measured |
| Qwen3 14B | 70.88 tok/s | 182 W46°CQ4_K_M | ✓ Measured |
| gpt-oss-20b | 220.65 tok/s | 146 W49°CQ4_K_M | ✓ Measured |
| Gemma 4 26B A4B | ✕ Won't fit needs ~20 GB | VRAM-gated at this precision | Est. |
| Qwen3 30B A3B | ✕ Won't fit needs ~20 GB | VRAM-gated at this precision | Est. |
| Gemma 4 31B | ✕ Won't fit needs ~22 GB | VRAM-gated at this precision | Est. |
| 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 |
|---|---|---|---|
| Stable Diffusion 1.5 | 51.23 images/min | 216 W45°C | ✓ Measured |
| SDXL Turbo | 484.82 images/min | 78 W38°C | ✓ Measured |
| Z-Image Turbo | 3.07 images/min | 157 W51°C | ✓ Measured 3 hosts ±16% |
| Sana 1.6B | 27.92 images/min | 243 W49°C | ✓ Measured |
| Stable Diffusion XL | 10.16 images/min | 14.7 GB peak298 W65°C5.9 s/img | ✓ Measured |
| Playground v2.5 | 6.33 images/min | 262 W54°C | ✓ Measured |
| PixArt-Sigma XL | 14.6 images/min | 250 W52°C | ✓ 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 |
|---|---|---|---|
| Stable Video Diffusion | 1.44 clips/min | 285 W70°C | ✓ Measured 2 hosts ±1% |
| LTX-Video (image to video) | 1.87 clips/min | 206 W67°C | ✓ Measured 2 hosts ±2% · CPU offload |
| Wan 2.2 TI2V-5B (image to video) | 0.8 clips/min | 244 W69°C | ✓ Measured 2 hosts ±0% · CPU offload |
| Stable Video Diffusion XT | 0.81 clips/min | 300 W76°C | ✓ Measured CPU offload |
| CogVideoX-5B I2V | 0.22 clips/min | 246 W69°C | ✓ Measured CPU offload |
| Workload | Result | Telemetry | Data |
|---|---|---|---|
| Wan 2.1 1.3B | 0.38 frames/s | 271 W65°C128.2 s/clip | ✓ Measured 2 hosts ±1% · CPU offload |
| CogVideoX-2B | 0.31 frames/s | 294 W68°C157.1 s/clip | ✓ Measured 2 hosts ±1% · CPU offload |
| CogVideoX-5B | 0.11 frames/s | 270 W72°C442.3 s/clip | ✓ Measured CPU offload |
| LTX-Video (distilled) | 3.71 frames/s | 182 W52°C26.2 s/clip | ✓ Measured 3 hosts ±12% · CPU offload |
| Wan 2.2 5B (720p) | ✕ Won't fit needs ~18 GB | VRAM-gated at this precision | ✓ Measured |
| Architecture | Ada Lovelace (AD103) |
| CUDA cores | 9,728 |
| VRAM | 16GB GDDR6X |
| Memory bus | 256-bit |
| Memory bandwidth | 716.8 GB/s |
| Boost clock | 2,505 MHz |
| TDP | 320 W |
| Process | 4nm (TSMC 4N) |
| Interface | PCIe 4.0 x16 |
| Release date | 2022-11-16 |
| Launch MSRP | $1,199 |
NVIDIA GeForce RTX 4080 scores 4.7/100, #42 of 102. It ran 6 of 12; 6 exceeded its 16GB. Every figure here is our own measurement.
100% = this card, AI & Machine Learning headline metric (AI Score). #10 of 61 desktop cards in this vertical.
| GPU | Relative | % | AI Score |
|---|---|---|---|
| NVIDIA GeForce RTX 3090 Ti | 177% | 8.5 | |
| NVIDIA Titan RTX | 171% | 8.2 | |
| NVIDIA GeForce RTX 3090 | 160% | 7.7 | |
| GeForce RTX 5080 | 108% | 5.2 | |
| NVIDIA GeForce RTX 4080 | 100% | 4.8 | |
| GeForce RTX 4080 Super | 98% | 4.7 | |
| GeForce RTX 5070 Ti | 98% | 4.7 | |
| NVIDIA RTX 4000 (Ada Generation) | 92% | 4.4 | |
| NVIDIA GeForce RTX 4070 Ti Super | 90% | 4.3 |
Same card, other workloads: NVIDIA GeForce RTX 4080 Gaming benchmarks
← All AI & Machine Learning GPU rankings
| Transistors | 45,900 million |
| Die size | 378.6 mm² |
| Process node | 4 nm |
| Fabricated by | TSMC |
| Transistor density | 121.2 million per mm² |
Denser than 94% of the 746 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 | 14.3 min | 44.74 Wh | measured |
| Animate a batch of images | 25.1 min | 101.75 Wh | measured |
Can't run: Product photo shoot (needs FLUX.1 Kontext dev), Photo restoration batch (needs FLUX.1 Kontext dev), Restore and enlarge photos (needs FLUX.1 Kontext dev), Character sheet, 12 poses (needs FLUX.1 dev), Short social clips (needs Qwen3 32B), 60-second AI short film (needs Qwen3 32B), 24-frame storyboard (needs Qwen3 32B), 6-panel comic page (needs Qwen3 32B), Long-form article batch (needs Llama 3.3 70B).
This card is $1,199 to buy. The cheapest listed rate on Vast.ai is $0.228/hour, but that is the floor: we budget $0.274/hour, a 20% premium, because idle time, storage and unavailable cheap instances all land on the same bill. At that rate buying wins after 4,382 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 | $200 | 6.0 years |
| 8 hours a day, working on it | 2,920 | $799 | 1.5 years |
| 24/7, always-on agent | 8,760 | $2,397 | 6.0 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.