48GB · AI Score 21.0/100 · first-party measured on 12 AI workloads
21 AI Score ✓ Measured
Every number on this page is first-party: NVIDIA RTX A6000 was run on our pinned 12-workload AI suite, with under 0.5% run-to-run variance. On Llama 3.1 8B (Q4_K_M) NVIDIA RTX A6000 delivers about 124.87 tokens/sec. Stepping up to Qwen3 32B it holds roughly 32.18 tok/s. The full Llama 3.3 70B still runs, at about 15.73 tok/s. For image generation, SDXL runs at 5.01 it/s, and FLUX.1-dev at 1.01 it/s. All 12 workloads fit in 48GB. There is no model in our suite this card has to turn down. NVIDIA RTX A6000 isn't a retail purchase for most people. It's rented by the hour. You can run this exact card on RunPod.
| Workload | Result | Telemetry | Data |
|---|---|---|---|
| MiniCPM5 2B | 270.65 tok/s | 159 W50°CQ4_K_M | ✓ Measured |
| LFM2.5 2.6B | 279.08 tok/s | 154 W48°CQ4_K_M | ✓ Measured |
| Granite 4.1 3B | 202.59 tok/s | 181 W55°CQ4_K_M | ✓ Measured |
| Nemotron 3 Nano 4B | 190.78 tok/s | 184 W52°CQ4_K_M | ✓ Measured |
| Qwen3 4B | 184.34 tok/s | 2.7 GB peak144 W48°C1.28 tok/WQ4_K_M | ✓ Measured |
| DeepSeek Coder 7B Instruct v1.5 | 140.44 tok/s | 196 W54°CQ4_K_M | ✓ Measured |
| Llama 3 8B | 124.54 tok/s | 195 W55°CQ4_K_M | ✓ Measured |
| Llama 3.1 8B | 124.87 tok/s | 4.7 GB peak169 W52°C0.74 tok/WQ4_K_M | ✓ Measured |
| Qwen3 8B | 120.24 tok/s | 195 W55°CQ4_K_M | ✓ Measured |
| Nemotron Nano 9B v2 | 90.62 tok/s | 209 W54°CQ4_K_M | ✓ Measured |
| Ornith 1.5 9B | 106.49 tok/s | 198 W54°CQ4_K_M | ✓ Measured |
| Gemma 4 12B | 75.82 tok/s | 206 W55°CQ4_K_M | ✓ Measured |
| Qwen2.5-Coder 14B | 68.51 tok/s | 8.7 GB peak171 W58°C0.4 tok/WQ4_K_M | ✓ Measured |
| Qwen3 14B | 70.24 tok/s | 212 W55°CQ4_K_M | ✓ Measured |
| Gemma 4 26B A4B | 138.08 tok/s | 156 W53°CQ4_K_M | ✓ Measured |
| Qwen3 30B A3B | 187.52 tok/s | 144 W53°CQ4_K_M | ✓ Measured |
| Qwen3 32B | 32.18 tok/s | 18.7 GB peak126 W60°C0.26 tok/WQ4_K_M | ✓ Measured |
| Llama 3.3 70B | 15.73 tok/s | 39.8 GB peak114 W63°C0.14 tok/WQ4_K_M | ✓ Measured |
| Workload | Result | Telemetry | Data |
|---|---|---|---|
| Stable Diffusion XL | 10.02 images/min | 15.8 GB peak297 W59°C6 s/img | ✓ Measured |
| Z-Image Turbo | 5.4 images/min | 25.6 GB peak296 W67°C11.1 s/img | ✓ Measured |
| Chroma1-HD | 0.62 images/min | 299 W88°C | ✓ Measured |
| FLUX.1 dev | 2.16 images/min | 36.5 GB peak298 W76°C27.7 s/img | ✓ Measured |
| Workload | Result | Telemetry | Data |
|---|---|---|---|
| FLUX.1 Kontext dev | 1.05 images/min | 35.3 GB peak296 W79°C57.7 s/img | ✓ Measured |
| Architecture | Ampere |
| CUDA cores | 10,752 |
| VRAM | 48GB GDDR6 |
| Memory bus | 384-bit |
| Memory bandwidth | 768 GB/s |
| Boost clock | 1,860 MHz |
| TDP | 300 W |
| Process | 8nm |
| Interface | PCIe 4.0 x16 |
| Release date | 2020-10-05 |
| Launch MSRP | $4,650 |
NVIDIA RTX A6000 scores 21.0/100, #20 of 102. It ran all 12 workloads. Every figure here is our own measurement.
100% = this card, AI & Machine Learning headline metric (AI Score). #4 of 20 workstation cards in this vertical.
| GPU | Relative | % | AI Score |
|---|---|---|---|
| NVIDIA RTX PRO 6000 Blackwell Workstation Edition | 253% | 53.1 | |
| NVIDIA RTX PRO 6000 Blackwell Max-Q Workstation Edition | 207% | 43.5 | |
| NVIDIA RTX PRO 5000 Blackwell | 150% | 31.6 | |
| NVIDIA RTX A6000 | 100% | 21 | |
| NVIDIA RTX PRO 4500 Blackwell | 61% | 12.9 | |
| NVIDIA RTX A5500 | 40% | 8.4 | |
| NVIDIA Quadro RTX 8000 | 39% | 8.2 | |
| NVIDIA RTX PRO 4000 Blackwell | 36% | 7.6 |
← All AI & Machine Learning GPU rankings
| Transistors | 28,300 million |
| Die size | 628.4 mm² |
| Process node | 8 nm |
| Fabricated by | Samsung |
| Transistor density | 45 million per mm² |
Denser than 81% 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 |
|---|---|---|---|
| 24-frame storyboard | 7.1 min | 23.42 Wh | all 2 stages measured |
| 6-panel comic page | 12.1 min | 42.76 Wh | all 3 stages measured |
| Full codebase review | 14.6 min | 41.58 Wh | measured |
| Character sheet, 12 poses | 15.2 min | 58.73 Wh | all 2 stages measured |
| Long-form article batch | 29.7 min | 56.52 Wh | measured |
| Product photo shoot | 45.5 min | 207.87 Wh | all 2 stages measured |
| Photo restoration batch | 1 h 37 min | 470.32 Wh | measured |
This card is $4,650 to buy. The cheapest listed rate on RunPod is $0.330/hour, but that is the floor: we budget $0.396/hour, a 20% premium, because idle time, storage and unavailable cheap instances all land on the same bill. At that rate buying wins after 11,742 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 | $289 | 16.1 years |
| 8 hours a day, working on it | 2,920 | $1,156 | 4.0 years |
| 24/7, always-on agent | 8,760 | $3,469 | 1.3 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.