24GB · AI Score 7.7/100 · first-party measured on 12 AI workloads
7.7 AI Score ✓ Measured
Every number on this page is first-party: NVIDIA GeForce RTX 3090 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 delivers about 144.9 tokens/sec. Stepping up to Qwen3 32B it holds roughly 37.95 tok/s. Llama 3.3 70B does not fit. It needs roughly 42GB and this card has 24GB. For image generation, SDXL runs at 3.73 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. NVIDIA GeForce RTX 3090 isn't a retail purchase for most people. It's rented by the hour. You can run this exact card on RunPod.
Still the used-market AI story. 1.34 tokens/watt is genuinely decent for Ampere, and the 24GB runs 8 of my 12 workloads including Qwen3 32B. Held 99% of its 350W rating on video gen without drama. If you want cheap VRAM, this is it. 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 | 206.67 tok/s | 2.7 GB peak154 W37°C1.34 tok/WQ4_K_M | ✓ Measured |
| Llama 3.1 8B | 144.9 tok/s | 4.9 GB peak179 W41°C0.81 tok/WQ4_K_M | ✓ Measured |
| Qwen2.5-Coder 14B | 78.79 tok/s | 8.7 GB peak178 W48°C0.44 tok/WQ4_K_M | ✓ Measured |
| Qwen3 32B | 37.95 tok/s | 18.7 GB peak136 W51°C0.28 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 | 7.46 images/min | 15.8 GB peak342 W57°C8 s/img | ✓ Measured |
| Z-Image Turbo | 3.83 images/min | 23.1 GB peak344 W62°C15.6 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) | 2.17 frames/s | 9.3 GB peak240 W60°C44.8 s/clip | ✓ Measured |
| Wan 2.2 5B (720p) | 0.22 frames/s | 16.9 GB peak314 W64°C221.8 s/clip | ✓ Measured |
| Architecture | Ampere (GA102) |
| CUDA cores | 10,496 |
| VRAM | 24GB GDDR6X |
| Memory bus | 384-bit |
| Memory bandwidth | 936.2 GB/s |
| Boost clock | 1,695 MHz |
| TDP | 350 W |
| Process | 8nm |
| Interface | PCIe 4.0 x16 |
| Release date | 2020-09-24 |
| Launch MSRP | $1,499 |
NVIDIA GeForce RTX 3090 scores 7.7/100, #33 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). #8 of 61 desktop cards in this vertical.
| GPU | Relative | % | AI Score |
|---|---|---|---|
| NVIDIA RTX 5000 Ada Generation | 143% | 11 | |
| NVIDIA GeForce RTX 4090 | 135% | 10.4 | |
| NVIDIA GeForce RTX 3090 Ti | 110% | 8.5 | |
| NVIDIA Titan RTX | 106% | 8.2 | |
| NVIDIA GeForce RTX 3090 | 100% | 7.7 | |
| AMD Radeon RX 7900 XTX | 87% | 6.7 | |
| AMD Radeon RX 7900 XT | 81% | 6.2 | |
| GeForce RTX 4080 Super | 70% | 5.4 | |
| GeForce RTX 5080 | 64% | 4.9 |
Same card, other workloads: NVIDIA GeForce RTX 3090 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 | 8.3 min | 37.34 Wh | all 2 stages measured |
| Full codebase review | 12.7 min | 37.63 Wh | measured |
| 60-second AI short film | 15.2 min | 57.1 Wh | all 3 stages measured |
| 10 short social clips | 42.4 min | 209.99 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 $1,499 to buy. The cheapest listed rate on Vast.ai is $0.069/hour, but that is the floor: we budget $0.083/hour, a 20% premium, because idle time, storage and unavailable cheap instances all land on the same bill. At that rate buying wins after 18,104 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 | $60 | 24.8 years |
| 8 hours a day, working on it | 2,920 | $242 | 6.2 years |
| 24/7, always-on agent | 8,760 | $725 | 2.1 years |
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.