Local LLM on NVIDIA RTX 5070

Is the NVIDIA RTX 5070 the right card for this?

At 12 GB this is an entry point rather than a destination: small and mid-sized models run comfortably, and anything above the 14B class does not fit at Q4 without offloading to system RAM. 672 GB/s of GDDR7 and a 250 W draw make it easy to slot into an existing machine. At about $550 the memory costs roughly $46 per gigabyte, which is the best ratio in this guide's consumer range. Catalogue throughput class: Q4 7B: 30-42, Q4 3B: 60-80.

One other card here carries 12 GB, so the two load the same weights and differ only in how fast they read them. At 672 GB/s this is the quickest of that group, so what you gain over the others is tokens per second, not a model you could not otherwise open.

At 2025 this is among the newest entries here, which cuts both ways. The memory technology is current, and the runtime support is the least settled of anything in this guide: check that the backend you intend to use has shipped a release naming this generation before buying for it.

What the price gap buys

The cheapest 12 GB option in this guide is the NVIDIA RTX 3060 12GB at about $250. This card costs $300 more and reads memory 87% faster, which is roughly the gain in tokens per second. It buys no additional model: both load exactly the same weights, so the question is only whether you are waiting on the output.

At 1 year old this is still a current retail part, so the price above is a shop price rather than a listing price and stock is the usual constraint rather than condition. Buying new also means the 250 W figure is the number your power supply has to meet from day one.

Where the NVIDIA RTX 5070 sits in this guide

Across the 20 cards catalogued here, the NVIDIA RTX 5070 ranks 13th on memory bandwidth at 672 GB/s, 6th cheapest at about $550, and 14th on bandwidth per watt at 2.7 GB/s per watt. Those three positions, not the capacity figure, are what separate it from other 12 GB parts.

GDDR7 is the current generation of board-mounted graphics memory, and it is the reason this card reaches 672 GB/s on a 12 GB bus where the previous generation needed a wider one. For token generation, which reads every weight once per token, that bandwidth translates almost linearly into speed.

This is a CUDA card, which in practice means every local runner in our tool directory supports it first. Ollama and llama.cpp pick it up without configuration; vLLM and TensorRT-LLM target it specifically. Driver and CUDA toolkit versions are the usual source of trouble, not the card.

Which catalogue models fit on NVIDIA RTX 5070?

Of the 46 models in our catalogue with per-size memory figures, 31 have at least one size that fits in 12 GB at Q4_K_M with room for context. The largest of each are below, biggest first.

Model	Largest size that fits	Weights at Q4_K_M
StarCoder2	15B	10 GB
LLaVA	13B	10 GB
DeepSeek Coder V2	16B	10 GB
Qwen 2.5	14B	9 GB
Qwen 2.5-Coder	14B	9 GB
Qwen 3	14B	9 GB
DeepSeek R1	14B	9 GB
Phi-4	14B	9 GB

A further 23 smaller models also fit; see the model catalogue.

The next sizes up are out of reach without offloading: GPT-OSS 20B at 14 GB, Mistral Small 24B at 15 GB, Magistral 24B at 15 GB.

Expected tokens per second on NVIDIA RTX 5070

Memory bandwidth, not compute, is the bottleneck for single-stream token generation. The NVIDIA RTX 5070's 672 GB/s puts the following ceiling on a batch of one at 2048 tokens of context.

Model	Quantization	Status	Approx tokens/sec
Llama 3.2 1B	Q4_K_M	Fits in VRAM	~121 tok/s
Llama 3.1 8B	Q4_K_M	Fits in VRAM	~40 tok/s
Llama 3.1 8B	Q8_0	Fits in VRAM	~30 tok/s
Mistral Nemo 12B	Q4_K_M	Fits in VRAM	~27 tok/s
Qwen 2.5 14B	Q4_K_M	Fits in VRAM	~24 tok/s
Qwen 3 32B	Q4_K_M	Needs CPU offload	~4 tok/s

These come from a bandwidth heuristic (tokens/sec ≈ bandwidth in GB/s × a per-model efficiency factor), not from a test rig in our office. Real numbers move with model architecture, batch size and KV-cache size. Treat them as an order of magnitude, and treat the "reported throughput class" row in the table above as the community-measured figure.

Build recommendations for NVIDIA RTX 5070

Best model to download first

Llama 3.1 8B or Mistral Nemo 12B. Start with ollama pull llama3.1:8b.

Recommended inference backend

Use Ollama for general use or Mullama if you need a drop-in Ollama alternative with native bindings for 6 languages. For production serving on multi-GPU setups, see vLLM.

Sources

[1] nvidia.com/rtx-5070 · VRAM figures for catalogue models from src/data/models.json (2026-06-29); card specifications from src/data/gpus.json (2026-06-29).