Best GPU for Phi-4 (14B): VRAM Requirements
Phi-4 (14B) needs about 34.2 GB of GPU memory in BF16 and 9.9 GB at 4 bits. See which GPUs fit it for inference and fine-tuning.
Phi-4 (14B) needs about 34.2 GB of VRAM in FP16, 18 GB in FP8 and 9.9 GB at 4 bits for inference with an 8,192-token context. In FP16 it fits on one NVIDIA A100 PCIe 40GB; a 4-bit build fits on one NVIDIA GeForce RTX 5080 16GB.
Phi-4 (14B) is a 14.7B-parameter language model from Microsoft, released in December 2024, with a 16,384-token context window and published under the MIT. Its weights alone take 29.4 GB in BF16 and 7.4 GB at 4 bits per parameter; the tables below add the KV cache and runtime memory, then list the GPUs that hold each setup.
Recommended GPUs for Phi-4 (14B)
Best for inference
1 GPU × NVIDIA L4 24GB
FP8 or INT8 weights, 18 GB at 8,192 tokens.
Best for fine-tuning
1 GPU × NVIDIA L40S 48GB
LoRA on BF16 weights, 38.2 GB.
Budget pick
1 GPU × NVIDIA GeForce RTX 5080 16GB
Smallest card class that holds a 4-bit build, 9.9 GB.
Phi-4 (14B) VRAM requirements by precision
One request with an 8,192-token context, KV cache in BF16, plus a 10 % margin for activations and runtime buffers.
| Precision | Weights | KV cache | Total for inference |
|---|---|---|---|
| BF16 / FP16 | 29.4 GB | 1.7 GB | 34.2 GB |
| FP8 | 14.7 GB | 1.7 GB | 18 GB |
| INT4 | 7.4 GB | 1.7 GB | 9.9 GB |
Phi-4 (14B) VRAM by task
Fine-tuning rows assume a short sequence, batch 1 and gradient checkpointing; longer sequences add activation memory.
| Task | GPU memory |
|---|---|
| Inference at the full 16,384-token context (FP8) | 19.9 GB |
| QLoRA (4-bit base) | 9.6 GB |
| LoRA (BF16 base) | 38.2 GB |
| Full fine-tuning (BF16, AdamW) | 258.7 GB |
GPUs that fit Phi-4 (14B)
Number of cards whose combined memory holds each setup, assuming the model is split across them with tensor or pipeline parallelism. A dash means more than eight cards.
| GPU | Memory | BF16 inference (34.2 GB) | FP8 inference (18 GB) | INT4 inference (9.9 GB) | QLoRA (9.6 GB) | LoRA (38.2 GB) |
|---|---|---|---|---|---|---|
| NVIDIA GeForce RTX 5080 16GB | 16 GB | 4 GPUs | 2 GPUs | 1 GPU | 1 GPU | 4 GPUs |
| NVIDIA GeForce RTX 4090 24GB | 24 GB | 2 GPUs | 1 GPU | 1 GPU | 1 GPU | 2 GPUs |
| NVIDIA L4 24GB | 24 GB | 2 GPUs | 1 GPU | 1 GPU | 1 GPU | 2 GPUs |
| NVIDIA GeForce RTX 5090 32GB | 32 GB | 2 GPUs | 1 GPU | 1 GPU | 1 GPU | 2 GPUs |
| NVIDIA Tesla V100 32GB | 32 GB | 2 GPUs | 1 GPU | 1 GPU | 1 GPU | 2 GPUs |
| NVIDIA A100 PCIe 40GB | 40 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
| NVIDIA L40S 48GB | 48 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
| NVIDIA Quadro RTX 8000 48GB | 48 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
| NVIDIA RTX 6000 Ada 48GB | 48 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
| NVIDIA RTX A6000 48GB | 48 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
| NVIDIA A100 PCIe 80GB | 80 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
| NVIDIA A100 SXM 80GB | 80 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
| NVIDIA H100 PCIe 80GB | 80 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
| NVIDIA H100 SXM 80GB | 80 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
| NVIDIA RTX PRO 6000 Blackwell 96GB | 96 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
| NVIDIA H200 141GB | 141 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
| NVIDIA B200 180GB | 180 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
| NVIDIA B300 288GB | 288 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
| NVIDIA GB300 NVL | 288 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
What is Phi-4 (14B)?
Phi-4 is a 14-billion-parameter dense model from Microsoft Research, trained largely on synthetic and curated data with a focus on reasoning and math. Its 16K-token context and MIT license suit compact, self-hosted assistants.
- Developer
- Microsoft
- Released
- December 2024
- Parameters
- 14.7B
- Context window
- 16,384 tokens
- License
- MIT
Source: Phi-4 model card (as of November 17, 2026).
Frequently asked questions
- How much VRAM does Phi-4 (14B) need?
- For inference with an 8,192-token context and one request, Phi-4 (14B) needs about 34.2 GB in BF16, 18 GB in FP8, 9.9 GB in INT4, including a 10 % runtime margin. Real usage depends on the inference engine and its settings.
- Can Phi-4 (14B) run on a single GPU?
- Yes. In BF16 it needs about 34.2 GB, which fits one NVIDIA A100 PCIe 40GB (40 GB).
- How much GPU memory does fine-tuning Phi-4 (14B) take?
- QLoRA (4-bit base): about 9.6 GB; LoRA (BF16 base): about 38.2 GB; Full fine-tuning (BF16, AdamW): about 258.7 GB. Longer training sequences and larger batches add activation memory on top.
- What is the best GPU for Phi-4 (14B)?
- Best for inference: 1 GPU × NVIDIA L4 24GB (FP8 or INT8 weights, 18 GB at 8,192 tokens). Best for fine-tuning: 1 GPU × NVIDIA L40S 48GB (LoRA on BF16 weights, 38.2 GB). Budget pick: 1 GPU × NVIDIA GeForce RTX 5080 16GB (Smallest card class that holds a 4-bit build, 9.9 GB). Check live offers before booking, as availability changes.
- Can Phi-4 (14B) run on an RTX 4090?
- Yes. In FP8 it needs about 18 GB, which fits one RTX 4090 (24 GB). FP16 needs about 34.2 GB, 2 of these cards.
- Can Phi-4 (14B) run on an RTX 5090?
- Yes. In FP8 it needs about 18 GB, which fits one RTX 5090 (32 GB). FP16 needs about 34.2 GB, 2 of these cards.
Estimates use the formulas of the LLM VRAM calculator (decimal GB). Real usage depends on the inference engine, quantization format and memory fragmentation: test on the target GPU before a long rental.
Explore the live market
Every offer on the live market comes from a reviewed supplier and is sold by EverXCloud Inc.