Cheapest box to run Phi-4 at Q8_0
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 Q8_0 needs 18GB.
Cheapest Config that fits: Mac mini, M6, 32GB. Fits, 11GB headroom. Est. 7.4 tok/s.
Best offer today: $1,499 at Apple — in stock, checked 9 minutes ago.
Price history: 50 days recorded.
Offers — Mac mini, M6, 32GB, 512GB
- Mac mini, M6, 16GB$1,099Won't fit−5GB headroom
- Mac mini, M6, 24GB$1,299Tight3GB headroomest. 7.4 tok/s
- Mac mini, M6, 32GB$1,499Fits11GB headroomest. 7.4 tok/s
- Mac mini, M5 Pro, 24GB$1,699Tight3GB headroomest. 13.4 tok/s
- Mac mini, M5 Pro, 48GB$2,499Fits26GB headroomest. 13.4 tok/s
- Mac Studio, M5 Max, 36GB$2,499Fits15GB headroomest. 20.1 tok/s
- Mac mini, M5 Pro, 64GB$2,899Fits41GB headroomest. 13.4 tok/s
- GeForce RTX 5090, 32GB$2,999.99+hostFits13GB headroomest. 78.3 tok/s
- Fits41GB headroomest. 23.9 tok/s
- Mac Studio, M5 Max, 48GB$3,099Fits26GB headroomest. 26.8 tok/s
- Mac Studio, M5 Max, 64GB$3,499Fits41GB headroomest. 26.8 tok/s
- Fits103GB headroomest. 11.9 tok/s
- Fits71GB headroomest. 35.8 tok/s
- Fits102GB headroomest. 23.9 tok/s
- Fits103GB headroomest. 11.9 tok/s
- Fits102GB headroomest. 26.8 tok/s
- Fits71GB headroomest. 52.4 tok/s
- Fits230GB headroomest. 35.8 tok/s
- Fits230GB headroomest. 52.4 tok/s
- Fits486GB headroomest. 52.4 tok/s
| Config | Memory | Price | Fit | Headroom | Est. speed |
|---|---|---|---|---|---|
| Mac mini, M6 | 16GB | $1,099 | Won't fit | −5GB | — |
| Mac mini, M6 | 24GB | $1,299 | Tight | 3GB | 7.4 tok/s |
| Mac mini, M6 | 32GB | $1,499 | Fits | 11GB | 7.4 tok/s |
| Mac mini, M5 Pro | 24GB | $1,699 | Tight | 3GB | 13.4 tok/s |
| Mac mini, M5 Pro | 48GB | $2,499 | Fits | 26GB | 13.4 tok/s |
| Mac Studio, M5 Max | 36GB | $2,499 | Fits | 15GB | 20.1 tok/s |
| Mac mini, M5 Pro | 64GB | $2,899 | Fits | 41GB | 13.4 tok/s |
| GeForce RTX 5090 | 32GB | $2,999.99+host | Fits | 13GB | 78.3 tok/s |
| Mac Studio, M4 Max, refurb | 64GB | $3,049 | Fits | 41GB | 23.9 tok/s |
| Mac Studio, M5 Max | 48GB | $3,099 | Fits | 26GB | 26.8 tok/s |
| Mac Studio, M5 Max | 64GB | $3,499 | Fits | 41GB | 26.8 tok/s |
| Ascent GX10, GB10 Grace Blackwell | 128GB | $3,999.99 | Fits | 103GB | 11.9 tok/s |
| Mac Studio, M3 Ultra, refurb | 96GB | $4,329 | Fits | 71GB | 35.8 tok/s |
| Mac Studio, M4 Max, refurb | 128GB | $4,409 | Fits | 102GB | 23.9 tok/s |
| DGX Spark, GB10 Grace Blackwell | 128GB | $4,699 | Fits | 103GB | 11.9 tok/s |
| Mac Studio, M5 Max | 128GB | $5,099 | Fits | 102GB | 26.8 tok/s |
| Mac Studio, M5 Ultra | 96GB | $5,499 | Fits | 71GB | 52.4 tok/s |
| Mac Studio, M3 Ultra, refurb | 256GB | $7,729 | Fits | 230GB | 35.8 tok/s |
| Mac Studio, M5 Ultra | 256GB | $9,499 | Fits | 230GB | 52.4 tok/s |
| Mac Studio, M5 Ultra | 512GB | — | Fits | 486GB | 52.4 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 × 1.063 B/param (8.5 bits) + 0.205 GB/1K × 8K K/V + 1.2 GB runtime
speed = bandwidth × 0.65 ÷ (14B active × 1.063 B/param) = 14.9 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 M6 32GB runs
- Qwen3.5-35B-A3BQ5_K_M, est. 40.0 tok/s
- Qwen3-32BQ5_K_M, est. 4.9 tok/s
- Gemma-4-31BQ6_K, est. 4.4 tok/s
- Qwen3-30B-A3BQ6_K, est. 31.4 tok/s
- Gemma-3-27BQ6_K, est. 5.0 tok/s
- Qwen3.6-27BQ6_K, est. 5.0 tok/s
Best quant that still fits at 8K context, largest model first. Full list on the Config page.
Bandwidth 170GB/s, usable memory 29GB of 32GB.