New: connect Claude & other AIs to GenAIList over MCP — research the catalog and contribute to the shared knowledge base. Learn how →
NVIDIA GeForce RTX 3070
Runs downloadable LLMs up to ~8.5B at Q4, ~73-87 tok/s on an 8B model.
What you can run
At 8k context, FP16 KV. Full-speed tok/s holds up to "max ctx"; beyond it the context spills to system RAM (CPU/PCIe offload) and drops to the degraded figure.
| Model size | FP16 | Q8 | Q4 |
|---|---|---|---|
| 3B | won't run |
109-129.9 t/s ≤44k ctx · then ~9.7-11.6 (slow) |
194.7-232 t/s ≤64k ctx · then ~17.4-20.7 (slow) |
| 7-8B | won't run | won't run |
73-87 t/s ≤11k ctx · then ~6.5-7.8 (slow) |
| 13B | won't run | won't run | won't run |
| 32B | won't run | won't run | won't run |
| 70B | won't run | won't run | won't run |
| 120B | won't run | won't run | won't run |
| 400B | won't run | won't run | won't run |
Estimate: tok/s ≈ bandwidth ÷ (params × bytes/param) × efficiency (NVIDIA). Cyan = fits ~7.4 GB usable. Beyond "ctx", KV spills to system RAM (CPU/PCIe offload) (rough degraded speed shown).
What else you need for a full build▶
Minimum supporting components for a desktop built around the NVIDIA GeForce RTX 3070. Guidance — exact needs depend on your case, other parts and how much you offload to system RAM.
Run popular models on the NVIDIA GeForce RTX 3070
Videos about the NVIDIA GeForce RTX 3070
running ai locally with 3070 on llama3.2 model
What AI Models can a NVIDIA 3070 Handle
More hardware like the NVIDIA GeForce RTX 3070
About the NVIDIA GeForce RTX 3070
NVIDIA GeForce RTX 3070 is a discrete GPU from NVIDIA on Ampere, 2020. For local genAI the numbers that matter are its 8 GB (how big a model + context fits) and 448 GB/s bandwidth (generation speed). Plan for a ~650W power supply.
Frequently asked questions
What size LLM can the NVIDIA GeForce RTX 3070 run?▶
With ~7.4 GB usable it runs up to ~8.5B at Q4 (~2.5B at FP16) at 8k context. Bigger spills into system RAM (CPU/PCIe offload) and slows sharply.
How many tokens/sec does the NVIDIA GeForce RTX 3070 generate?▶
Generation is bandwidth-bound. At 448 GB/s an 8B model at Q4 runs ~73-87 tok/s (estimate).
Why does long context slow it down?▶
The KV cache for the context must also fit fast memory; once weights+KV exceed 7.4 GB it spills to system RAM (CPU/PCIe offload) and throughput collapses. Each row shows the max usable context.
What else do I need to run the NVIDIA GeForce RTX 3070?▶
Besides the card: a ≥650W 80+ Gold power supply, at least 16GB of system RAM (2× the 8GB VRAM is ideal for CPU offload), a free PCIe x16 slot, any modern multi-core CPU, and a ≥1TB NVMe SSD for model weights.