Cheapest box to run Gemma-4-31B at Q8_0
Google’s Gemma-4-31B is 30.7B 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.
Gemma-4-31B at Q8_0 needs 35GB.
Cheapest Config that fits: Mac mini, M5 Pro, 48GB. Fits, 9GB headroom. Est. 6.1 tok/s.
Best offer today: $2,499 at Apple — in stock, checked 32 minutes ago.
Price history: 34 days recorded.
Dense. Only 10 of 60 layers are global attention; the sliding-window layers stop growing past 1024 tokens, so the K/V figure counts the global layers alone.
Offers — Mac mini, M5 Pro, 48GB, 512GB
- Applelist pricein stockchecked 32 minutes ago$2,499
- Mac mini, M6, 16GB$1,099Won't fit−22GB headroom
- Mac mini, M6, 24GB$1,299Won't fit−14GB headroom
- Mac mini, M6, 32GB$1,499Won't fit−6GB headroom
- Mac mini, M5 Pro, 24GB$1,699Won't fit−14GB headroom
- Mac mini, M5 Pro, 48GB$2,499Fits9GB headroomest. 6.1 tok/s
- Mac Studio, M5 Max, 36GB$2,499Won't fit−2GB headroom
- Mac mini, M5 Pro, 64GB$2,899Fits24GB headroomest. 6.1 tok/s
- GeForce RTX 5090, 32GB$2,999.99+hostWon't fit−4GB headroom
- Fits24GB headroomest. 10.9 tok/s
- Mac Studio, M5 Max, 48GB$3,099Fits9GB headroomest. 12.2 tok/s
- Mac Studio, M5 Max, 64GB$3,499Fits24GB headroomest. 12.2 tok/s
- Fits85GB headroomest. 5.4 tok/s
- Fits53GB headroomest. 16.3 tok/s
- Fits85GB headroomest. 10.9 tok/s
- Fits85GB headroomest. 5.4 tok/s
- Fits85GB headroomest. 12.2 tok/s
- Fits53GB headroomest. 23.9 tok/s
- Fits213GB headroomest. 16.3 tok/s
- Fits213GB headroomest. 23.9 tok/s
- Fits469GB headroomest. 23.9 tok/s
| Config | Memory | Price | $/GB | Fit | Headroom | Est. speed |
|---|---|---|---|---|---|---|
| Mac mini, M6 | 16GB | $1,099 | $69 | Won't fit | −22GB | — |
| Mac mini, M6 | 24GB | $1,299 | $54 | Won't fit | −14GB | — |
| Mac mini, M6 | 32GB | $1,499 | $47 | Won't fit | −6GB | — |
| Mac mini, M5 Pro | 24GB | $1,699 | $71 | Won't fit | −14GB | — |
| Mac mini, M5 Pro | 48GB | $2,499 | $52 | Fits | 9GB | 6.1 tok/s |
| Mac Studio, M5 Max | 36GB | $2,499 | $69 | Won't fit | −2GB | — |
| Mac mini, M5 Pro | 64GB | $2,899 | $45 | Fits | 24GB | 6.1 tok/s |
| GeForce RTX 5090 | 32GB | $2,999.99+host | $94 | Won't fit | −4GB | — |
| Mac Studio, M4 Max, refurb | 64GB | $3,049 | $48 | Fits | 24GB | 10.9 tok/s |
| Mac Studio, M5 Max | 48GB | $3,099 | $65 | Fits | 9GB | 12.2 tok/s |
| Mac Studio, M5 Max | 64GB | $3,499 | $55 | Fits | 24GB | 12.2 tok/s |
| Ascent GX10, GB10 Grace Blackwell | 128GB | $3,999.99 | $31 | Fits | 85GB | 5.4 tok/s |
| Mac Studio, M3 Ultra, refurb | 96GB | $4,329 | $45 | Fits | 53GB | 16.3 tok/s |
| Mac Studio, M4 Max, refurb | 128GB | $4,409 | $34 | Fits | 85GB | 10.9 tok/s |
| DGX Spark, GB10 Grace Blackwell | 128GB | $4,699 | $37 | Fits | 85GB | 5.4 tok/s |
| Mac Studio, M5 Max | 128GB | $5,099 | $40 | Fits | 85GB | 12.2 tok/s |
| Mac Studio, M5 Ultra | 96GB | $5,499 | $57 | Fits | 53GB | 23.9 tok/s |
| Mac Studio, M3 Ultra, refurb | 256GB | $7,729 | $30 | Fits | 213GB | 16.3 tok/s |
| Mac Studio, M5 Ultra | 256GB | $9,499 | $37 | Fits | 213GB | 23.9 tok/s |
| Mac Studio, M5 Ultra | 512GB | — | — | Fits | 469GB | 23.9 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 = 30.7B × 1.063 B/param (8.5 bits) + 0.168 GB/1K × 8K K/V + 1.2 GB runtime
speed = bandwidth × 0.65 ÷ (30.7B active × 1.063 B/param) = 32.6 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
- Qwen3-Coder-NextQ3_K_M, est. 105 tok/s
- Llama-3.3-70BQ3_K_M, est. 5.8 tok/s
- Qwen3.5-35B-A3BQ8_0, est. 48.2 tok/s
- Qwen3-32BQ8_0, est. 5.9 tok/s
- Gemma-4-31BQ8_0, est. 6.1 tok/s
- Qwen3-30B-A3BQ8_0, est. 43.8 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.