What GPU do I need to run RedHatAI/phi-4-FP8-dynamic?
14.7B parameters, published in F8_E4M3. View on Hugging Face
Set up phi-4-FP8-dynamicphi-4-FP8-dynamic is published by RedHatAI on Hugging Face, with 97,070 downloads and 3 likes to date. It's a Phi3ForCausalLM model built for text-generation, published natively in F8_E4M3.
VRAM required & cheapest live GPU fit
Required VRAM = weight size at each precision times a flat 1.2 overhead for activations and fragmentation. The KV-cache grows with context and is not in that factor; it is listed per model below where the architecture is published. Full formula and assumptions: methodology.
| Precision | Weight size | Required VRAM | Cheapest live fit | GPUs needed | Est. $/hr (full fit) |
|---|---|---|---|---|---|
| FP8 (native) | 13.7 GB | 16.4 GB | RTX 4000 SFF Ada | 1 | $0.198/hr |
| INT4 (quantized) | 6.8 GB | 8.2 GB | RTX 3060 | 1 | $0.110/hr |
A GPU is only matched to a row if its hardware supports that precision, and the primary recommendation is always a single-GPU fit when one exists.
INT4 requires a quantized checkpoint actually published for this model, check its Hugging Face page before relying on this row.
Cheapest way to run phi-4-FP8-dynamic at its published (F8_E4M3) precision: 1× RTX 4000 SFF Ada, at $0.198/hr per GPU ($0.198/hr total). Quantizing to FP8 or INT4 (rows above) can cost less, but requires a compatible quantized checkpoint to exist for this model.
phi-4-FP8-dynamic: KV cache by context length
The KV cache is the memory the attention layers hold for every token of context, on top of the weights. It is not part of the flat 1.2x overhead in the table above, grows with context length and with every concurrent request, and is why a long-context deployment needs more VRAM than the table shows.
Grouped-query attention: every layer caches keys and values for a few shared KV heads. Cached per token: 204,800 bytes at 16-bit.
| Context | KV cache, one sequence |
|---|---|
| 4K | 0.78 GB |
| 16K (model maximum) | 3.1 GB |
Computed from the layer, head and window counts in the model's own configuration (RedHatAI/phi-4-FP8-dynamic), at 16 bits per cached value. Each concurrent sequence needs its own cache, on top of the weights.
phi-4-FP8-dynamic: common questions
Does phi-4-FP8-dynamic fit on a 24 GB GPU?
Yes. At FP8 (native) it needs 16.4 GB of VRAM, so a 24 GB card holds it with 7.6 GB to spare. A 16 GB card is not enough for it at FP8 (native).
Is phi-4-FP8-dynamic already quantized?
Yes. It is published in FP8, one byte per parameter, so the 16.4 GB figure is already a quantized footprint rather than a full-precision one. Only INT4 goes below it, at 8.2 GB. A GPU without FP8 tensor cores cannot run it as published, which is why cards here are matched on precision support and not on VRAM alone.
What is the least VRAM phi-4-FP8-dynamic can run in?
8.2 GB, at INT4 (quantized), which fits a 12 GB card, against 16.4 GB at FP8 (native). That row assumes an INT4 checkpoint has actually been published for this model, so check its Hugging Face page before planning around it.
Does quantizing phi-4-FP8-dynamic lower the GPU bill?
Yes. At FP8 (native) the cheapest live fit is one RTX 4000 SFF Ada at $0.198/hr. At INT4 (quantized) it drops to one RTX 3060 at $0.110/hr, provided a quantized checkpoint exists for it.
Weight-to-VRAM math, the fit rules, and how live prices are normalized: full methodology.
More Phi-4 models
- phi-4 (14.7B, BF16)
- Phi-4-reasoning (14.7B, BF16)
- Phi-4-reasoning-plus (14.7B, BF16)
- Phi-4-mini-instruct (3.8B, BF16)
- Phi-4-mini-reasoning (3.8B, BF16)
Lineage
Based on phi-4.
Alternatives at this size
Other models for text-generation within about a third of phi-4-FP8-dynamic's 14.7B parameters, from other model lines.
- Qwen2.5-14B-Instruct (14.8B, BF16)
- gpt-neox-20b (20.7B, F16)
- DeepSeek-Coder-V2-Lite-Instruct (15.7B, BF16)
- LLaDA2.0-mini (16.3B, BF16)
- vllm-translategemma-12b-it (13.2B, BF16)
More on phi-4-FP8-dynamic
Fits on a 24 GB GPU at FP8: every model that fits in 24 GB.
Fits on a 12 GB GPU at INT4: every model that fits in 12 GB.
Related reading: H100 pricing and specs, How much VRAM you need for LLMs, Serving LLMs with vLLM, vLLM vs TensorRT-LLM vs SGLang, and Best GPU for LLM inference.
