24GB · AI Score 8.5/100 · first-party measured on 12 AI workloads
8.5 AI Score ✓ Measured
Every number on this page is first-party: NVIDIA GeForce RTX 3090 Ti 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 3090 Ti delivers about 161.02 tokens/sec. Stepping up to Qwen3 32B it holds roughly 42.18 tok/s. Llama 3.3 70B does not fit. It needs roughly 42GB and this card has 24GB. For image generation, SDXL runs at 4.21 it/s, while FLUX.1-dev won't fit at BF16 (needs ~26GB). 4 of the 12 workloads won't fit on 24GB at the tested precision, Llama 3.3 70B, FLUX.1-dev, FLUX.1 Kontext, Qwen-Image-Edit. We publish those as hard gates rather than quietly dropping to a smaller quant.
A flat 450W under AI load, the logs barely moved off the rating. But the 24GB is why you'd want it: only 4 of 12 workloads gated, and it stayed at a civil 68°C even on video generation. Power bill aside, it just runs things. 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 | 227.54 tok/s | 3 GB peak214 W63°C1.07 tok/WQ4_K_M | ✓ Measured |
| Llama 3.1 8B | 161.02 tok/s | 4.7 GB peak248 W65°C0.65 tok/WQ4_K_M | ✓ Measured |
| Qwen2.5-Coder 14B | 88.25 tok/s | 8.7 GB peak264 W64°C0.33 tok/WQ4_K_M | ✓ Measured |
| Qwen3 32B | 42.18 tok/s | 18.7 GB peak235 W63°C0.18 tok/WQ4_K_M | ✓ Measured |
| Llama 3.3 70B | ✕ Won't fit needs ~46 GB | VRAM-gated at this precision | ✓ Measured |
| Workload | Result | Telemetry | Data |
|---|---|---|---|
| Stable Diffusion XL | 8.42 images/min | 15.8 GB peak412 W65°C7.1 s/img | ✓ Measured |
| Z-Image Turbo | 4.35 images/min | 23.1 GB peak387 W66°C13.9 s/img | ✓ 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) | 3.2 frames/s | 9.3 GB peak296 W63°C30.3 s/clip | ✓ Measured |
| Wan 2.2 5B (720p) | 0.27 frames/s | 16.7 GB peak380 W68°C179.5 s/clip | ✓ Measured |
| Architecture | Ampere (GA102) |
| CUDA cores | 10,752 |
| VRAM | 24GB GDDR6X |
| Memory bus | 384-bit |
| Memory bandwidth | 1008 GB/s |
| Boost clock | 1,860 MHz |
| TDP | 450 W |
| Process | 8nm (Samsung 8N) |
| Interface | PCIe 4.0 x16 |
| Release date | 2022-03-29 |
| Launch MSRP | $1,999 |
NVIDIA GeForce RTX 3090 Ti scores 8.5/100, #30 of 102. It ran 8 of 12; 4 exceeded its 24GB. Every figure here is our own measurement.
100% = this card, AI & Machine Learning headline metric (AI Score). #6 of 61 desktop cards in this vertical.
| GPU | Relative | % | AI Score |
|---|---|---|---|
| NVIDIA GeForce RTX 5090 | 262% | 22.3 | |
| NVIDIA RTX 5880 Ada Generation | 192% | 16.3 | |
| NVIDIA RTX 5000 Ada Generation | 129% | 11 | |
| NVIDIA GeForce RTX 4090 | 122% | 10.4 | |
| NVIDIA GeForce RTX 3090 Ti | 100% | 8.5 | |
| NVIDIA Titan RTX | 96% | 8.2 | |
| NVIDIA GeForce RTX 3090 | 91% | 7.7 | |
| AMD Radeon RX 7900 XTX | 79% | 6.7 | |
| AMD Radeon RX 7900 XT | 73% | 6.2 |
Same card, other workloads: NVIDIA GeForce RTX 3090 Ti Gaming benchmarks
← All AI & Machine Learning GPU rankings
| Transistors | 28,300 million |
| Die size | 628.4 mm² |
| Fabricated by | Samsung |
| Transistor density | 45 million per mm² |
Denser than 49% 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 |
|---|---|---|---|
| 24-frame storyboard | 6.3 min | 37.71 Wh | all 2 stages measured |
| 60-second AI short film | 9.9 min | 50.99 Wh | all 3 stages measured |
| Full codebase review | 11.3 min | 49.88 Wh | measured |
| 10 short social clips | 33.1 min | 207.05 Wh | all 3 stages measured |
Can't run: 40-product photo shoot (needs FLUX.1 Kontext dev), 6-panel comic page (needs FLUX.1 dev), 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), 100-photo restore and enlarge (needs FLUX.1 Kontext dev).
This card is $1,999 to buy. The cheapest listed rate on RunPod is $0.270/hour, but that is the floor: we budget $0.324/hour, a 20% premium, because idle time, storage and unavailable cheap instances all land on the same bill. At that rate buying wins after 6,170 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 | $237 | 8.5 years |
| 8 hours a day, working on it | 2,920 | $946 | 2.1 years |
| 24/7, always-on agent | 8,760 | $2,838 | 8.5 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.