Cheapest box to run Gemma-3n-E4B at BF16
Google’s Gemma-3n-E4B is 6.8B 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-3n-E4B at BF16 needs 15GB.
Cheapest Config that fits: Mac mini, M6, 24GB. Fits, 6GB headroom. Est. 13.8 tok/s.
Best offer today: $1,299 at Best Buy — in stock, checked 37 hours ago.
Collecting price history — 3 of 90 days. No timing verdict yet.
Google names this model for its 4B effective parameters, but 6.8B raw weights are what you load, so the memory arithmetic uses 6.8B. The K/V figure accounts for layer sharing and sliding-window attention and is an estimate.
Offers — Mac mini, M6, 24GB, 512GB
- Mac mini, M6, 16GB$1,099Won't fit−2GB headroom
- Mac mini, M6, 24GB$1,299Fits6GB headroomest. 13.8 tok/s
- Mac mini, M6, 32GB$1,499Fits14GB headroomest. 13.8 tok/s
- Mac mini, M5 Pro, 24GB$1,699Fits6GB headroomest. 24.9 tok/s
- Mac mini, M5 Pro, 48GB$2,499Fits29GB headroomest. 24.9 tok/s
- Mac Studio, M5 Max, 36GB$2,499Fits18GB headroomest. 49.9 tok/s
- Mac mini, M5 Pro, 64GB$2,899Fits44GB headroomest. 24.9 tok/s
- GeForce RTX 5090, 32GB$2,999.99+hostFits15GB headroomest. 146 tok/s
- Fits44GB headroomest. 44.4 tok/s
- Mac Studio, M5 Max, 48GB$3,099Fits29GB headroomest. 49.9 tok/s
- Mac Studio, M5 Max, 64GB$3,499Fits44GB headroomest. 49.9 tok/s
- Fits105GB headroomest. 22.2 tok/s
- Fits73GB headroomest. 66.5 tok/s
- Fits105GB headroomest. 44.4 tok/s
- Fits105GB headroomest. 22.2 tok/s
- Fits105GB headroomest. 49.9 tok/s
- Fits73GB headroomest. 97.5 tok/s
- Fits233GB headroomest. 66.5 tok/s
- Fits233GB headroomest. 97.5 tok/s
- Fits489GB headroomest. 97.5 tok/s
| Config | Memory | Price | $/GB | Fit | Headroom | Est. speed |
|---|---|---|---|---|---|---|
| Mac mini, M6 | 16GB | $1,099 | $69 | Won't fit | −2GB | — |
| Mac mini, M6 | 24GB | $1,299 | $54 | Fits | 6GB | 13.8 tok/s |
| Mac mini, M6 | 32GB | $1,499 | $47 | Fits | 14GB | 13.8 tok/s |
| Mac mini, M5 Pro | 24GB | $1,699 | $71 | Fits | 6GB | 24.9 tok/s |
| Mac mini, M5 Pro | 48GB | $2,499 | $52 | Fits | 29GB | 24.9 tok/s |
| Mac Studio, M5 Max | 36GB | $2,499 | $69 | Fits | 18GB | 49.9 tok/s |
| Mac mini, M5 Pro | 64GB | $2,899 | $45 | Fits | 44GB | 24.9 tok/s |
| GeForce RTX 5090 | 32GB | $2,999.99+host | $94 | Fits | 15GB | 146 tok/s |
| Mac Studio, M4 Max, refurb | 64GB | $3,049 | $48 | Fits | 44GB | 44.4 tok/s |
| Mac Studio, M5 Max | 48GB | $3,099 | $65 | Fits | 29GB | 49.9 tok/s |
| Mac Studio, M5 Max | 64GB | $3,499 | $55 | Fits | 44GB | 49.9 tok/s |
| Ascent GX10, GB10 Grace Blackwell | 128GB | $3,999.99 | $31 | Fits | 105GB | 22.2 tok/s |
| Mac Studio, M3 Ultra, refurb | 96GB | $4,329 | $45 | Fits | 73GB | 66.5 tok/s |
| Mac Studio, M4 Max, refurb | 128GB | $4,409 | $34 | Fits | 105GB | 44.4 tok/s |
| DGX Spark, GB10 Grace Blackwell | 128GB | $4,699 | $37 | Fits | 105GB | 22.2 tok/s |
| Mac Studio, M5 Max | 128GB | $5,099 | $40 | Fits | 105GB | 49.9 tok/s |
| Mac Studio, M5 Ultra | 96GB | $5,499 | $57 | Fits | 73GB | 97.5 tok/s |
| Mac Studio, M3 Ultra, refurb | 256GB | $7,729 | $30 | Fits | 233GB | 66.5 tok/s |
| Mac Studio, M5 Ultra | 256GB | $9,499 | $37 | Fits | 233GB | 97.5 tok/s |
| Mac Studio, M5 Ultra | 512GB | — | — | Fits | 489GB | 97.5 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 = 6.8B × 2.000 B/param (16 bits) + 0.072 GB/1K × 8K K/V + 1.2 GB runtime
speed = bandwidth × 0.65 ÷ (4B active × 2.000 B/param) = 8.0 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 M6 24GB runs
- Qwen3-32BQ3_K_M, est. 7.1 tok/s
- Qwen3-30B-A3BQ4_K_M, est. 42.7 tok/s
- Gemma-3-27BQ4_K_M, est. 6.8 tok/s
- Devstral-Small-2507Q5_K_M, est. 6.5 tok/s
- gpt-oss-20bQ6_K, est. 28.8 tok/s
- Phi-4Q8_0, est. 7.4 tok/s
Best quant that still fits at 8K context, largest model first. Full list on the Config page.
Bandwidth 170GB/s, usable memory 21GB of 24GB.