Cheapest box to run Phi-4 at BF16
Microsoft’s Phi-4 is 14B parameters, dense. The number below is computed from that and each vendor’s published specs, not asserted. Switch the quant or the context and it recomputes in place.
Phi-4 at BF16 needs 31GB.
Cheapest Config that fits: Mac mini, M5 Pro, 48GB. Fits, 13GB headroom. Est. 7.1 tok/s.
Best offer today: $2,499 at Apple — in stock, checked 38 hours ago.
Collecting price history — 3 of 90 days. No timing verdict yet.
Offers — Mac mini, M5 Pro, 48GB, 512GB
- Applelist pricein stockchecked 38 hours ago$2,499
- Mac mini, M6, 16GB$1,099Won't fit−18GB headroom
- Mac mini, M6, 24GB$1,299Won't fit−10GB headroom
- Mac mini, M6, 32GB$1,499Won't fit−2GB headroom
- Mac mini, M5 Pro, 24GB$1,699Won't fit−10GB headroom
- Mac mini, M5 Pro, 48GB$2,499Fits13GB headroomest. 7.1 tok/s
- Mac Studio, M5 Max, 36GB$2,499Tight2GB headroomest. 14.3 tok/s
- Mac mini, M5 Pro, 64GB$2,899Fits28GB headroomest. 7.1 tok/s
- GeForce RTX 5090, 32GB$2,999.99+hostWon't fit−0GB headroom
- Fits28GB headroomest. 12.7 tok/s
- Mac Studio, M5 Max, 48GB$3,099Fits13GB headroomest. 14.3 tok/s
- Mac Studio, M5 Max, 64GB$3,499Fits28GB headroomest. 14.3 tok/s
- Fits89GB headroomest. 6.3 tok/s
- Fits57GB headroomest. 19.0 tok/s
- Fits89GB headroomest. 12.7 tok/s
- Fits89GB headroomest. 6.3 tok/s
- Fits89GB headroomest. 14.3 tok/s
- Fits57GB headroomest. 27.9 tok/s
- Fits217GB headroomest. 19.0 tok/s
- Fits217GB headroomest. 27.9 tok/s
- Fits473GB headroomest. 27.9 tok/s
| Config | Memory | Price | $/GB | Fit | Headroom | Est. speed |
|---|---|---|---|---|---|---|
| Mac mini, M6 | 16GB | $1,099 | $69 | Won't fit | −18GB | — |
| Mac mini, M6 | 24GB | $1,299 | $54 | Won't fit | −10GB | — |
| Mac mini, M6 | 32GB | $1,499 | $47 | Won't fit | −2GB | — |
| Mac mini, M5 Pro | 24GB | $1,699 | $71 | Won't fit | −10GB | — |
| Mac mini, M5 Pro | 48GB | $2,499 | $52 | Fits | 13GB | 7.1 tok/s |
| Mac Studio, M5 Max | 36GB | $2,499 | $69 | Tight | 2GB | 14.3 tok/s |
| Mac mini, M5 Pro | 64GB | $2,899 | $45 | Fits | 28GB | 7.1 tok/s |
| GeForce RTX 5090 | 32GB | $2,999.99+host | $94 | Won't fit | −0GB | — |
| Mac Studio, M4 Max, refurb | 64GB | $3,049 | $48 | Fits | 28GB | 12.7 tok/s |
| Mac Studio, M5 Max | 48GB | $3,099 | $65 | Fits | 13GB | 14.3 tok/s |
| Mac Studio, M5 Max | 64GB | $3,499 | $55 | Fits | 28GB | 14.3 tok/s |
| Ascent GX10, GB10 Grace Blackwell | 128GB | $3,999.99 | $31 | Fits | 89GB | 6.3 tok/s |
| Mac Studio, M3 Ultra, refurb | 96GB | $4,329 | $45 | Fits | 57GB | 19.0 tok/s |
| Mac Studio, M4 Max, refurb | 128GB | $4,409 | $34 | Fits | 89GB | 12.7 tok/s |
| DGX Spark, GB10 Grace Blackwell | 128GB | $4,699 | $37 | Fits | 89GB | 6.3 tok/s |
| Mac Studio, M5 Max | 128GB | $5,099 | $40 | Fits | 89GB | 14.3 tok/s |
| Mac Studio, M5 Ultra | 96GB | $5,499 | $57 | Fits | 57GB | 27.9 tok/s |
| Mac Studio, M3 Ultra, refurb | 256GB | $7,729 | $30 | Fits | 217GB | 19.0 tok/s |
| Mac Studio, M5 Ultra | 256GB | $9,499 | $37 | Fits | 217GB | 27.9 tok/s |
| Mac Studio, M5 Ultra | 512GB | — | — | Fits | 473GB | 27.9 tok/s |
GeForce RTX 5090 is a card, not a computer: every price shown for it is the cheapest listing we found plus $1,000 for the minimum host machine.
memory = 14B × 2.000 B/param (16 bits) + 0.205 GB/1K × 8K K/V + 1.2 GB runtime
speed = bandwidth × 0.65 ÷ (14B active × 2.000 B/param) = 28.0 GB read per token
usable = memory − operating system reserve (8% on a Mac, 6% on a DGX Spark, 4% on a card)
Bandwidth is the vendor’s published figure. Every speed is an estimate and says so. Methodology.
What else Mac mini M5 Pro 48GB runs
- Llama-3.3-70BQ3_K_M, est. 5.8 tok/s
- Qwen3-32BQ8_0, est. 5.9 tok/s
- Qwen3-30B-A3BQ8_0, est. 43.8 tok/s
- Gemma-3-27BQ8_0, est. 7.0 tok/s
- Devstral-Small-2507Q8_0, est. 7.8 tok/s
- gpt-oss-20bBF16, est. 21.3 tok/s
Best quant that still fits at 8K context, largest model first. Full list on the Config page.
Bandwidth 307GB/s, usable memory 44GB of 48GB.