16GB · AI Score 5.4/100 · first-party measured on 12 AI workloads
4.7 AI Score Includes estimates
Every number on this page is first-party: GeForce RTX 4080 Super 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) GeForce RTX 4080 Super delivers about 126.29 tokens/sec. Llama 3.3 70B does not fit. It needs roughly 42GB and this card has 16GB. For image generation, SDXL runs at 4.56 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.
Being straight with you: two of my runs on this card (Z-Image and LTX video) came back 2-3x slower than the plain 4080 with every LLM number identical. That's a borderline-VRAM offload problem, not the card. Those runs are flagged and excluded from the score until I re-run them. What's solid: 1.72 tokens/watt and it only used 88% of its power rating. 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 |
|---|---|---|---|
| MiniCPM5 2B | 311.94 tok/s | 122 W49°CQ4_K_M | ✓ Measured |
| LFM2.5 2.6B | 308.63 tok/s | 120 W50°CQ4_K_M | ✓ Measured |
| Granite 4.1 3B | 227.06 tok/s | 137 W53°CQ4_K_M | ✓ Measured |
| Nemotron 3 Nano 4B | 200.69 tok/s | 135 W51°CQ4_K_M | ✓ Measured |
| Qwen3 4B | 197.53 tok/s | 2.7 GB peak115 W45°C1.72 tok/WQ4_K_M | ✓ Measured |
| DeepSeek Coder 7B Instruct v1.5 | 146.98 tok/s | 171 W54°CQ4_K_M | ✓ Measured |
| Llama 3 8B | 130.53 tok/s | 167 W56°CQ4_K_M | ✓ Measured |
| Llama 3.1 8B | 126.29 tok/s | 4.7 GB peak130 W48°C0.98 tok/WQ4_K_M | ✓ Measured |
| Qwen3 8B | 126.39 tok/s | 165 W54°CQ4_K_M | ✓ Measured |
| Nemotron Nano 9B v2 | 93.43 tok/s | 167 W59°CQ4_K_M | ✓ Measured |
| Ornith 1.5 9B | 111.6 tok/s | 164 W58°CQ4_K_M | ✓ Measured |
| Gemma 4 12B | 80.57 tok/s | 178 W55°CQ4_K_M | ✓ Measured |
| Qwen2.5-Coder 14B | 70.55 tok/s | 8.7 GB peak140 W51°C0.51 tok/WQ4_K_M | ✓ Measured |
| Qwen3 14B | 72.96 tok/s | 192 W59°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 | 51.96 images/min | 184 W50°C | ✓ Measured |
| SDXL Turbo | 410.08 images/min | 67 W43°C | ✓ Measured |
| Stable Diffusion XL | 11.54 images/min | 313 W69°C | ✓ Measured |
| Z-Image Turbo | 2.3 images/min | 162 W65°C | ✓ Measured 3 hosts ±5% |
| Sana 1.6B | 26.98 images/min | 217 W60°C | ✓ Measured |
| Playground v2.5 | 6.74 images/min | 233 W62°C | ✓ Measured |
| PixArt-Sigma XL | 14.59 images/min | 225 W63°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 |
|---|---|---|---|
| LTX-Video (distilled) | 2.85 frames/s | 158 W64°C34 s/clip | ✓ Measured 3 hosts ±5% · CPU offload |
| Wan 2.2 5B (720p) | ✕ Won't fit needs ~18 GB | VRAM-gated at this precision | ✓ Measured |
| Architecture | Ada Lovelace |
| CUDA cores | 10,240 |
| VRAM | 16GB GDDR6X |
| Memory bus | 256-bit |
| Memory bandwidth | 736 GB/s |
| Boost clock | 2,505 MHz |
| TDP | 320 W |
| Process | 4nm |
| Interface | PCIe 4.0 x16 |
| Release date | 2024-01-31 |
| Launch MSRP | $999 |
GeForce RTX 4080 Super scores 5.4/100, #43 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). #11 of 61 desktop cards in this vertical.
| GPU | Relative | % | AI Score |
|---|---|---|---|
| NVIDIA Titan RTX | 174% | 8.2 | |
| NVIDIA GeForce RTX 3090 | 164% | 7.7 | |
| GeForce RTX 5080 | 111% | 5.2 | |
| NVIDIA GeForce RTX 4080 | 102% | 4.8 | |
| GeForce RTX 4080 Super | 100% | 4.7 | |
| GeForce RTX 5070 Ti | 100% | 4.7 | |
| NVIDIA RTX 4000 (Ada Generation) | 94% | 4.4 | |
| NVIDIA GeForce RTX 4070 Ti Super | 91% | 4.3 | |
| AMD Radeon RX 7900 XTX | 79% | 3.7 |
Same card, other workloads: GeForce RTX 4080 Super 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 | 14.2 min | 33.05 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 $999 to buy. The cheapest listed rate on Vast.ai is $0.222/hour, but that is the floor: we budget $0.266/hour, a 20% premium, because idle time, storage and unavailable cheap instances all land on the same bill. At that rate buying wins after 3,750 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 | $194 | 5.1 years |
| 8 hours a day, working on it | 2,920 | $778 | 1.3 years |
| 24/7, always-on agent | 8,760 | $2,334 | 5.1 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.