Phi-4 hardware requirements
What is Phi-4?
Phi-4 is a 14B dense model from Microsoft Research, released in December 2024. It was trained on a blend of synthetic "textbook-like" data, filtered web data and academic books, with a focus on reasoning and logic such as math and coding. Microsoft aims it mainly at English, its context window is 16k tokens, and the MIT license allows commercial use.
About the model
- Parameters
- 14.7B
- Attention
- Full attention
- Max context
- 16,384 tokens
- Released
- 2024-12-12
- Model card
- Phi-4 on Hugging Face
GGUF files
| Quant | File size | Quality | Published by |
|---|---|---|---|
| Q4_K_Mbaseline | 9.05 GB | Good balance | Community GGUF · lmstudio-community |
| Q6_K | 12.03 GB | Near-lossless | Community GGUF · lmstudio-community |
| Q8_0 | 15.58 GB | Near-lossless | Community GGUF · lmstudio-community |
Memory needed by quant and context
| Quant | 4k | 8k | 16k |
|---|---|---|---|
| Q4_K_M | 10.1 | 11.0 | 12.8 |
| Q6_K | 13.0 | 13.9 | 15.6 |
| Q8_0 | 16.4 | 17.3 | 19.0 |
Which GPUs can run Phi-4?
How to read the verdicts
- Runs great
- Fully on the GPU at 20 tok/s or more
- Runs well
- Fully on the GPU at 8–20 tok/s; MoE experts in system RAM at 20 tok/s or more; or at least 90% on the GPU at 8 tok/s or more
- Runs slowly
- 2–8 tok/s; CPU-only; less than 90% on the GPU; or MoE experts in system RAM below 20 tok/s
- Won't run
- Does not fit, or under 2 tok/s
Dense models split between GPU and CPU slow down sharply — the CPU side sets the pace. MoE models that keep only their experts in system RAM degrade far more gently.
Filter this list by budget, type and context in the calculator
Scroll sideways to see every column.
| Hardware | Verdict | Speed | Quant | Memory | Runs as | Context | Notes | Price |
|---|---|---|---|---|---|---|---|---|
| GeForce RTX 5090 | Runs great | est. 116.9 tok/s102.9–131.0calibrated estimate ±12% | Q4_K_M | 11.6 / 32.0 GB | Full GPU | up to 16k | — | $4,200 street (as of 2026-08-09) |
| NVIDIA RTX PRO 6000 Blackwell | Runs great | est. 116.9 tok/s102.9–131.0calibrated estimate ±12% | Q4_K_M | 11.6 / 96.0 GB | Full GPU | up to 16k | — | — |
| NVIDIA RTX PRO 6000 Blackwell Max-Q | Runs great | est. 116.9 tok/s102.9–131.0calibrated estimate ±12% | Q4_K_M | 11.6 / 96.0 GB | Full GPU | up to 16k | — | — |
| Radeon RX 7900 XTX | Runs great | est. 68.0 tok/s59.8–76.2calibrated estimate ±12% | Q4_K_M | 11.6 / 24.0 GB | Full GPU | up to 16k | — | $999 launch MSRP |
| GeForce RTX 3090 Ti | Runs great | est. 65.8 tok/s57.9–73.7calibrated estimate ±12% | Q4_K_M | 11.6 / 24.0 GB | Full GPU | up to 16k | — | $1,999 launch MSRP |
| GeForce RTX 4090 | Runs great | est. 65.8 tok/s57.9–73.7calibrated estimate ±12% | Q4_K_M | 11.6 / 24.0 GB | Full GPU | up to 16k | — | $2,755 street (as of 2026-08-09) |
| Apple M5 Ultra · 96 GB | Runs great | est. 65.2 tok/s52.2–78.3calibrated estimate ±20% | Q4_K_M | 11.3 / 67.2 GB | Unified memory | up to 16k | — | — |
| GeForce RTX 5080 | Runs great | est. 62.6 tok/s55.1–70.2calibrated estimate ±12% | Q4_K_M | 11.6 / 16.0 GB | Full GPU | up to 16k | — | $1,256 street (as of 2026-08-09) |
| GeForce RTX 3090 | Runs great | est. 61.1 tok/s53.7–68.4calibrated estimate ±12% | Q4_K_M | 11.6 / 24.0 GB | Full GPU | up to 16k | — | $1,050 street (as of 2026-08-09) |
| GeForce RTX 3080 12GB | Runs great | est. 59.5 tok/s52.4–66.7calibrated estimate ±12% | Q4_K_M | 11.6 / 12.0 GB | Full GPU | up to 8k | — | $799 launch MSRP |
| GeForce RTX 5070 Ti | Runs great | est. 58.5 tok/s51.4–65.5calibrated estimate ±12% | Q4_K_M | 11.6 / 16.0 GB | Full GPU | up to 16k | — | $949 street (as of 2026-08-09) |
| GeForce RTX 5080 Laptop | Runs great | est. 58.5 tok/s51.4–65.5calibrated estimate ±12% | Q4_K_M | 11.6 / 16.0 GB | Full GPU | up to 16k | — | — |
| GeForce RTX 5090 Laptop | Runs great | est. 58.5 tok/s51.4–65.5calibrated estimate ±12% | Q4_K_M | 11.6 / 24.0 GB | Full GPU | up to 16k | — | — |
| Radeon RX 7900 XT | Runs great | est. 56.7 tok/s49.9–63.5calibrated estimate ±12% | Q4_K_M | 11.6 / 20.0 GB | Full GPU | up to 16k | — | $899 launch MSRP |
| GeForce RTX 4080 Super | Runs great | est. 48.0 tok/s42.3–53.8calibrated estimate ±12% | Q4_K_M | 11.6 / 16.0 GB | Full GPU | up to 16k | — | $999 launch MSRP |
| GeForce RTX 4080 | Runs great | est. 46.8 tok/s41.2–52.4calibrated estimate ±12% | Q4_K_M | 11.6 / 16.0 GB | Full GPU | up to 16k | — | $1,199 launch MSRP |
| Radeon RX 9070 | Runs great | est. 45.7 tok/s40.2–51.2calibrated estimate ±12% | Q4_K_M | 11.6 / 16.0 GB | Full GPU | up to 16k | — | $639 street (as of 2026-08-09) |
| Radeon RX 9070 XT | Runs great | est. 45.7 tok/s40.2–51.2calibrated estimate ±12% | Q4_K_M | 11.6 / 16.0 GB | Full GPU | up to 16k | — | $689 street (as of 2026-08-09) |
| Radeon RX 7800 XT | Runs great | est. 44.2 tok/s38.9–49.5calibrated estimate ±12% | Q4_K_M | 11.6 / 16.0 GB | Full GPU | up to 16k | — | $499 launch MSRP |
| GeForce RTX 4070 Ti Super | Runs great | est. 43.8 tok/s38.6–49.1calibrated estimate ±12% | Q4_K_M | 11.6 / 16.0 GB | Full GPU | up to 16k | — | $799 launch MSRP |
| GeForce RTX 5070 | Runs great | est. 43.8 tok/s38.6–49.1calibrated estimate ±12% | Q4_K_M | 11.6 / 12.0 GB | Full GPU | up to 8k | — | $629 street (as of 2026-08-09) |
| GeForce RTX 5070 Ti Laptop | Runs great | est. 43.8 tok/s38.6–49.1calibrated estimate ±12% | Q4_K_M | 11.6 / 12.0 GB | Full GPU | up to 8k | — | — |
| Apple M3 Ultra · 96 GB | Runs great | est. 43.5 tok/s34.8–52.2calibrated estimate ±20% | Q4_K_M | 11.3 / 67.2 GB | Unified memory | up to 16k | — | — |
| Apple M5 Max (40-core GPU) · 48 GB | Runs great | est. 43.5 tok/s34.8–52.2calibrated estimate ±20% | Q4_K_M | 11.3 / 33.6 GB | Unified memory | up to 16k | — | — |
| Apple M2 Ultra · 64 GB | Runs great | est. 42.5 tok/s34.0–51.0calibrated estimate ±20% | Q4_K_M | 11.3 / 44.8 GB | Unified memory | up to 16k | — | — |
| Apple M4 Max (40-core GPU) · 48 GB | Runs great | est. 38.7 tok/s30.9–46.4calibrated estimate ±20% | Q4_K_M | 11.3 / 33.6 GB | Unified memory | up to 16k | — | — |
| GeForce RTX 4090 Laptop | Runs great | est. 37.6 tok/s33.1–42.1calibrated estimate ±12% | Q4_K_M | 11.6 / 16.0 GB | Full GPU | up to 16k | — | — |
| GeForce RTX 4070 | Runs great | est. 32.9 tok/s28.9–36.8calibrated estimate ±12% | Q4_K_M | 11.6 / 12.0 GB | Full GPU | up to 8k | — | $599 launch MSRP |
| GeForce RTX 4070 Super | Runs great | est. 32.9 tok/s28.9–36.8calibrated estimate ±12% | Q4_K_M | 11.6 / 12.0 GB | Full GPU | up to 8k | — | $599 launch MSRP |
| Apple M5 Max (32-core GPU) · 36 GB | Runs great | est. 32.6 tok/s26.1–39.1calibrated estimate ±20% | Q4_K_M | 11.3 / 25.2 GB | Unified memory | up to 16k | — | — |
| GeForce RTX 5060 Ti 16GB | Runs great | est. 29.2 tok/s25.7–32.7calibrated estimate ±12% | Q4_K_M | 11.6 / 16.0 GB | Full GPU | up to 16k | — | $569 street (as of 2026-08-09) |
| Apple M4 Max (32-core GPU) · 36 GB | Runs great | est. 29.0 tok/s23.2–34.9calibrated estimate ±20% | Q4_K_M | 11.3 / 25.2 GB | Unified memory | up to 16k | — | — |
| GeForce RTX 4080 Laptop | Runs great | est. 28.2 tok/s24.8–31.6calibrated estimate ±12% | Q4_K_M | 11.6 / 12.0 GB | Full GPU | up to 8k | — | — |
| Intel Arc B580 | Runs great | est. 25.1 tok/s20.1–30.1calibrated estimate ±20% | Q4_K_M | 11.6 / 12.0 GB | Full GPU | up to 8k | — | $290 street (as of 2026-08-09) |
| GeForce RTX 3060 12GB | Runs great | est. 23.5 tok/s20.7–26.3calibrated estimate ±12% | Q4_K_M | 11.6 / 12.0 GB | Full GPU | up to 8k | — | $250 street (as of 2026-08-09) |
| Radeon RX 9060 XT 16GB | Runs great | est. 22.8 tok/s20.1–25.5calibrated estimate ±12% | Q4_K_M | 11.6 / 16.0 GB | Full GPU | up to 8k | — | $449 street (as of 2026-08-09) |
| Apple M5 Pro · 24 GB | Runs great | est. 21.7 tok/s17.4–26.1calibrated estimate ±20% | Q4_K_M | 11.3 / 16.8 GB | Unified memory | up to 8k | — | — |
| GeForce RTX 2080 Ti | Runs well | est. 21.6 tok/s17.3–25.9calibrated estimate ±20% | Q4_K_M | 11.0 / 11.0 GB + 0.9 GB RAM | Partial offload | up to 8k |
| $999 launch MSRP |
| NVIDIA DGX Spark · 128 GB | Runs well | est. 19.8 tok/s15.9–23.8calibrated estimate ±20% | Q4_K_M | 11.3 / 116.9 GB | Unified memory | up to 16k | — | $3,999 launch MSRP (128 GB) |
| Ryzen AI Max+ 395 (Strix Halo) · 32 GB | Runs well | est. 19.6 tok/s15.7–23.5calibrated estimate ±20% | Q4_K_M | 11.3 / 22.4 GB | Unified memory | up to 16k | — | $1,999 launch MSRP (128 GB) |
| Apple M4 Pro · 24 GB | Runs well | est. 19.3 tok/s15.5–23.2calibrated estimate ±20% | Q4_K_M | 11.3 / 16.8 GB | Unified memory | up to 16k | — | — |
| GeForce GTX 1080 Ti | Runs well | est. 19.1 tok/s13.4–24.8theoretical estimate ±30% | Q4_K_M | 11.0 / 11.0 GB + 0.9 GB RAM | Partial offload | up to 8k | — | $699 launch MSRP |
| GeForce RTX 4060 Ti 16GB | Runs well | est. 18.8 tok/s16.5–21.0calibrated estimate ±12% | Q4_K_M | 11.6 / 16.0 GB | Full GPU | up to 16k | — | $499 launch MSRP |
| Apple M6 · 24 GB | Runs well | est. 12.1 tok/s9.7–14.5calibrated estimate ±20% | Q4_K_M | 11.3 / 16.8 GB | Unified memory | up to 16k | — | — |
| Apple M5 · 24 GB | Runs well | est. 10.9 tok/s8.7–13.1calibrated estimate ±20% | Q4_K_M | 11.3 / 16.8 GB | Unified memory | up to 16k | — | — |
| Apple M4 · 24 GB | Runs well | est. 8.5 tok/s6.8–10.2calibrated estimate ±20% | Q4_K_M | 11.3 / 16.8 GB | Unified memory | up to 8k | — | — |
| GeForce RTX 3080 10GB | Runs slowly | est. 15.1 tok/s12.1–18.1calibrated estimate ±20% | Q4_K_M | 10.0 / 10.0 GB + 1.9 GB RAM | Partial offload | up to 16k |
| $699 launch MSRP |
| GeForce RTX 3070 | Runs slowly | est. 8.1 tok/s6.5–9.8calibrated estimate ±20% | Q4_K_M | 8.0 / 8.0 GB + 3.9 GB RAM | Partial offload | up to 16k |
| $499 launch MSRP |
| GeForce RTX 5060 | Runs slowly | est. 8.1 tok/s6.5–9.8calibrated estimate ±20% | Q4_K_M | 8.0 / 8.0 GB + 3.9 GB RAM | Partial offload | up to 16k |
| $339 street (as of 2026-08-09) |
| GeForce RTX 5060 Ti 8GB | Runs slowly | est. 8.1 tok/s6.5–9.8calibrated estimate ±20% | Q4_K_M | 8.0 / 8.0 GB + 3.9 GB RAM | Partial offload | up to 16k |
| $429 street (as of 2026-08-09) |
| GeForce RTX 4060 Ti 8GB | Runs slowly | est. 7.5 tok/s6.0–9.0calibrated estimate ±20% | Q4_K_M | 8.0 / 8.0 GB + 3.9 GB RAM | Partial offload | up to 16k | — | $399 launch MSRP |
| GeForce RTX 4060 | Runs slowly | est. 7.4 tok/s5.9–8.9calibrated estimate ±20% | Q4_K_M | 8.0 / 8.0 GB + 3.9 GB RAM | Partial offload | up to 16k | — | $299 launch MSRP |
| GeForce RTX 4060 Laptop | Runs slowly | est. 7.3 tok/s5.8–8.7calibrated estimate ±20% | Q4_K_M | 8.0 / 8.0 GB + 3.9 GB RAM | Partial offload | up to 16k | — | — |
| GeForce RTX 4070 Laptop | Runs slowly | est. 7.3 tok/s5.8–8.7calibrated estimate ±20% | Q4_K_M | 8.0 / 8.0 GB + 3.9 GB RAM | Partial offload | up to 16k | — | — |
| GeForce RTX 3050 8GB | Runs slowly | est. 7.0 tok/s5.6–8.4calibrated estimate ±20% | Q4_K_M | 8.0 / 8.0 GB + 3.9 GB RAM | Partial offload | up to 16k | — | $249 launch MSRP |
| DDR5-6000 dual-channel CPU · 16 GB | Runs slowly | est. 4.5 tok/s3.6–5.4calibrated estimate ±20% | Q4_K_M | 10.7 GB RAM | CPU only | up to 8k | — | — |
| DDR5-5600 dual-channel CPU · 16 GB | Runs slowly | est. 4.2 tok/s3.3–5.0calibrated estimate ±20% | Q4_K_M | 10.7 GB RAM | CPU only | up to 8k | — | — |
| DDR4-3200 dual-channel CPU · 16 GB | Runs slowly | est. 2.4 tok/s1.9–2.9calibrated estimate ±20% | Q4_K_M | 10.7 GB RAM | CPU only | up to 8k | — | — |
Cheapest GPUs that run it
- Runs greatGeForce RTX 3060 12GB$250 street (as of 2026-08-09), est. 23.5 tok/s (20.7–26.3, calibrated estimate ±12%).
Measured results for Phi-4
No public measurements for this model yet.
Frequently asked questions
How much VRAM does Phi-4 need?
At Q4_K_M the file is 9.05 GB; with 8k context a graphics card needs about 11.0 GB (weights 8.76 GB, KV cache 1.7 GB, compute buffer 0.6 GB). Part of the file stays off the graphics card: llama.cpp keeps the input embedding table in system RAM. A Mac counts the whole file, so it needs a little more.
What is the cheapest GPU that runs Phi-4 well?
Cheapest that runs it great: GeForce RTX 3060 12GB at $250 (street price as of 2026-08-09), est. 23.5 tok/s (20.7–26.3, calibrated estimate ±12%).
Can Phi-4 run on an 8 GB, 16 GB or 24 GB GPU?
At Q4_K_M with 8k context: GeForce RTX 4060 — Runs slowly, est. 7.4 tok/s (5.9–8.9, calibrated estimate ±20%); GeForce RTX 5060 Ti 16GB — Runs great, est. 29.2 tok/s (25.7–32.7, calibrated estimate ±12%); GeForce RTX 3090 — Runs great, est. 61.1 tok/s (53.7–68.4, calibrated estimate ±12%).
Speeds are estimates from memory bandwidth, calibrated against public benchmarks, and each one comes with an error band and a confidence label. Real results vary with drivers, backend, context length and thermals.