Cheapest box to run Gemma-3n-E4B at Q5_K_M
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 Q5_K_M needs 7GB.
Cheapest Config that fits: Mac mini, M6, 16GB. Fits, 6GB headroom. Est. 39.0 tok/s.
Best offer today: $1,099 at Apple — in stock, checked 38 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, 16GB, 512GB
- Applelist pricein stockchecked 38 hours ago$1,099
- Mac mini, M6, 16GB$1,099Fits6GB headroomest. 39.0 tok/s
- Mac mini, M6, 24GB$1,299Fits14GB headroomest. 39.0 tok/s
- Mac mini, M6, 32GB$1,499Fits22GB headroomest. 39.0 tok/s
- Mac mini, M5 Pro, 24GB$1,699Fits14GB headroomest. 70.4 tok/s
- Mac mini, M5 Pro, 48GB$2,499Fits38GB headroomest. 70.4 tok/s
- Mac Studio, M5 Max, 36GB$2,499Fits26GB headroomest. 141 tok/s
- Mac mini, M5 Pro, 64GB$2,899Fits52GB headroomest. 70.4 tok/s
- GeForce RTX 5090, 32GB$2,999.99+hostFits24GB headroomest. 411 tok/s
- Fits52GB headroomest. 125 tok/s
- Mac Studio, M5 Max, 48GB$3,099Fits38GB headroomest. 141 tok/s
- Mac Studio, M5 Max, 64GB$3,499Fits52GB headroomest. 141 tok/s
- Fits114GB headroomest. 62.6 tok/s
- Fits82GB headroomest. 188 tok/s
- Fits113GB headroomest. 125 tok/s
- Fits114GB headroomest. 62.6 tok/s
- Fits113GB headroomest. 141 tok/s
- Fits82GB headroomest. 275 tok/s
- Fits241GB headroomest. 188 tok/s
- Fits241GB headroomest. 275 tok/s
- Fits497GB headroomest. 275 tok/s
| Config | Memory | Price | $/GB | Fit | Headroom | Est. speed |
|---|---|---|---|---|---|---|
| Mac mini, M6 | 16GB | $1,099 | $69 | Fits | 6GB | 39.0 tok/s |
| Mac mini, M6 | 24GB | $1,299 | $54 | Fits | 14GB | 39.0 tok/s |
| Mac mini, M6 | 32GB | $1,499 | $47 | Fits | 22GB | 39.0 tok/s |
| Mac mini, M5 Pro | 24GB | $1,699 | $71 | Fits | 14GB | 70.4 tok/s |
| Mac mini, M5 Pro | 48GB | $2,499 | $52 | Fits | 38GB | 70.4 tok/s |
| Mac Studio, M5 Max | 36GB | $2,499 | $69 | Fits | 26GB | 141 tok/s |
| Mac mini, M5 Pro | 64GB | $2,899 | $45 | Fits | 52GB | 70.4 tok/s |
| GeForce RTX 5090 | 32GB | $2,999.99+host | $94 | Fits | 24GB | 411 tok/s |
| Mac Studio, M4 Max, refurb | 64GB | $3,049 | $48 | Fits | 52GB | 125 tok/s |
| Mac Studio, M5 Max | 48GB | $3,099 | $65 | Fits | 38GB | 141 tok/s |
| Mac Studio, M5 Max | 64GB | $3,499 | $55 | Fits | 52GB | 141 tok/s |
| Ascent GX10, GB10 Grace Blackwell | 128GB | $3,999.99 | $31 | Fits | 114GB | 62.6 tok/s |
| Mac Studio, M3 Ultra, refurb | 96GB | $4,329 | $45 | Fits | 82GB | 188 tok/s |
| Mac Studio, M4 Max, refurb | 128GB | $4,409 | $34 | Fits | 113GB | 125 tok/s |
| DGX Spark, GB10 Grace Blackwell | 128GB | $4,699 | $37 | Fits | 114GB | 62.6 tok/s |
| Mac Studio, M5 Max | 128GB | $5,099 | $40 | Fits | 113GB | 141 tok/s |
| Mac Studio, M5 Ultra | 96GB | $5,499 | $57 | Fits | 82GB | 275 tok/s |
| Mac Studio, M3 Ultra, refurb | 256GB | $7,729 | $30 | Fits | 241GB | 188 tok/s |
| Mac Studio, M5 Ultra | 256GB | $9,499 | $37 | Fits | 241GB | 275 tok/s |
| Mac Studio, M5 Ultra | 512GB | — | — | Fits | 497GB | 275 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 × 0.709 B/param (5.67 bits) + 0.072 GB/1K × 8K K/V + 1.2 GB runtime
speed = bandwidth × 0.65 ÷ (4B active × 0.709 B/param) = 2.8 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 16GB runs
- gpt-oss-20bQ3_K_M, est. 48.3 tok/s
- Phi-4Q5_K_M, est. 11.1 tok/s
- Qwen3-VL-8B-InstructQ8_0, est. 13.0 tok/s
- Gemma-3n-E4BQ8_0, est. 26.0 tok/s
- Qwen3-4BBF16, est. 13.8 tok/s
Best quant that still fits at 8K context, largest model first. Full list on the Config page.
Bandwidth 170GB/s, usable memory 13GB of 16GB.