How much VRAM does Phi-4 need?

About 11.2 GB at Q4_K_M, 18.9 GB at Q8_0 and 34.7 GB at FP16, estimated for a ~4k context.

How fast is Phi-4 on an RTX 3070 Ti?

We do not estimate tokens/sec when the model does not fit. CPU offload is possible but slow.

What context length can I use?

The weights already do not fit, so longer context is not the limiting factor.

Home Can I run it?Phi-4Phi-4 on an RTX 3070 Ti

No. Phi-4 at Q4_K_M needs about 11.2 GB, and the GeForce RTX 3070 Ti has 8.0 GB usable at a ~4k context.

Can I run Phi-4 on an RTX 3070 Ti?

Phi-4 at Q4_K_M needs about 11.2 GB, and the GeForce RTX 3070 Ti has 8.0 GB usable. That figure is 8 GB of VRAM on the spec sheet, of which 8.0 GB is usable for weights and cache, at 608 GB/s memory bandwidth. Microsoft’s Phi-4, a dense 14B-class model often compared with other 12B–14B instruct weights. These are estimates from our formula, not measured benchmarks.

Quant that fits: None of Q4, Q8 or FP16 needs 11.2 GB
Memory needed: 11.2 GB at Q4_K_M, ~4k context
Usable VRAM: 8.0 GB of 8 GB VRAM
Est. tokens/sec: n/a does not fit

phi-4 · GeForce RTX 3070 Ti · Everything this machine runs · Q4 vs Q8 vs FP16

Memory at each quant

Quant	Need vs 8.0 GB	Verdict
Q4_K_M	11.2 GB	Q4Won’t fit
Q8_0	18.9 GB	Q8Won’t fit
FP16	34.7 GB	FP16Won’t fit

Estimated tokens per second

No speed estimate: Phi-4 does not fit in the 8.0 GB this machine can use, so we do not publish a tokens/sec figure for CPU offload.

Context headroom

Even at a short ~4k context, the weights do not fit. Extending context would only increase the gap.

Context headroom for Phi-4 on an RTX 3070 Ti at Q4_K_M
Context	KV cache (FP16)	Total at Q4_K_M	Verdict
4k	0.8 GB	11.2 GB	Won’t fit
8k	1.7 GB	12.1 GB	Won’t fit
16k	3.4 GB	13.7 GB	Won’t fit
32k	6.7 GB	17.1 GB	Won’t fit
64k	13.4 GB	23.8 GB	Won’t fit
128k	26.8 GB	37.2 GB	Won’t fit

From the model’s config. Try other sizes in the VRAM calculator or see the context length guide.

What to do instead

Run Qwen3-8B instead (6.7 GB at Q4). It is a different model, not a drop-in, but it does fit this machine.
To hold Phi-4 at Q4, you need about 11.2 GB. The GeForce RTX 3060 12GB has 12.0 GB usable ($329 list price).
Our estimate is only 3.2 GB over this machine’s usable memory, so a tighter quant or a shorter context may work in practice. We still mark it as a no under the published formula.
CPU offload can load the weights anyway, but it is slow. We do not publish an offload tokens/sec figure.
Full hardware list: phi-4. Specs for this machine: GeForce RTX 3070 Ti.

Other hardware for Phi-4

Other models on an RTX 3070 Ti

Qwen3-32B on an RTX 3070 TiWon’t fit
Qwen3-30B-A3B on an RTX 3070 TiWon’t fit
Qwen3.8-27B on an RTX 3070 TiWon’t fit
Gemma 3 27B on an RTX 3070 TiWon’t fit
Gemma 4 31B on an RTX 3070 TiWon’t fit
gpt-oss-20b on an RTX 3070 TiWon’t fit

Quick answers

Can I run Phi-4 on an RTX 3070 Ti?: No. Phi-4 at Q4_K_M needs about 11.2 GB, and the GeForce RTX 3070 Ti has 8.0 GB usable.
How much VRAM does Phi-4 need?: About 11.2 GB at Q4_K_M, 18.9 GB at Q8_0 and 34.7 GB at FP16, estimated for a ~4k context.
How fast is Phi-4 on an RTX 3070 Ti?: We do not estimate tokens/sec when the model does not fit. CPU offload is possible but slow.
What context length can I use?: The weights already do not fit, so longer context is not the limiting factor.