Cheapest box to run Llama-3.3-70B at Q6_K
Meta’s Llama-3.3-70B is 70B 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.
Llama-3.3-70B at Q6_K needs 61GB.
Cheapest Config that fits: DGX Spark, GB10 Grace Blackwell, 128GB. Fits, 59GB headroom. Est. 3.1 tok/s.
Best offer today: $4,699 at NVIDIA — in stock, checked 9 hours ago.
Collecting price history — 1 of 90 days. No timing verdict yet.
Offers — DGX Spark, GB10 Grace Blackwell, 128GB, 4TB
- NVIDIAlist pricein stockchecked 9 hours ago$4,699
- Mac mini, M6, 16GB$1,099Won't fit−48GB headroom
- Mac mini, M6, 24GB$1,299Won't fit−40GB headroom
- Mac mini, M6, 32GB$1,499Won't fit−32GB headroom
- Mac mini, M5 Pro, 24GB$1,699Won't fit−40GB headroom
- Mac mini, M5 Pro, 48GB$2,499Won't fit−17GB headroom
- Mac Studio, M5 Max, 36GB$2,499Won't fit−28GB headroom
- Mac mini, M5 Pro, 64GB$2,899Won't fit−2GB headroom
- Mac Studio, M5 Max, 48GB$3,099Won't fit−17GB headroom
- Mac Studio, M5 Max, 64GB$3,499Won't fit−2GB headroom
- Fits59GB headroomest. 3.1 tok/s
- Fits59GB headroomest. 7.0 tok/s
- Fits27GB headroomest. 13.6 tok/s
- GeForce RTX 5090, 32GB$5,699.99+hostWon't fit−31GB headroom
- Fits187GB headroomest. 13.6 tok/s
- Fits443GB headroomest. 13.6 tok/s
| Config | Memory | Price | $/GB | Fit | Headroom | Est. speed |
|---|---|---|---|---|---|---|
| Mac mini, M6 | 16GB | $1,099 | $69 | Won't fit | −48GB | — |
| Mac mini, M6 | 24GB | $1,299 | $54 | Won't fit | −40GB | — |
| Mac mini, M6 | 32GB | $1,499 | $47 | Won't fit | −32GB | — |
| Mac mini, M5 Pro | 24GB | $1,699 | $71 | Won't fit | −40GB | — |
| Mac mini, M5 Pro | 48GB | $2,499 | $52 | Won't fit | −17GB | — |
| Mac Studio, M5 Max | 36GB | $2,499 | $69 | Won't fit | −28GB | — |
| Mac mini, M5 Pro | 64GB | $2,899 | $45 | Won't fit | −2GB | — |
| Mac Studio, M5 Max | 48GB | $3,099 | $65 | Won't fit | −17GB | — |
| Mac Studio, M5 Max | 64GB | $3,499 | $55 | Won't fit | −2GB | — |
| DGX Spark, GB10 Grace Blackwell | 128GB | $4,699 | $37 | Fits | 59GB | 3.1 tok/s |
| Mac Studio, M5 Max | 128GB | $5,099 | $40 | Fits | 59GB | 7.0 tok/s |
| Mac Studio, M5 Ultra | 96GB | $5,499 | $57 | Fits | 27GB | 13.6 tok/s |
| GeForce RTX 5090 | 32GB | $5,699.99+host | $178 | Won't fit | −31GB | — |
| Mac Studio, M5 Ultra | 256GB | $9,499 | $37 | Fits | 187GB | 13.6 tok/s |
| Mac Studio, M5 Ultra | 512GB | — | — | Fits | 443GB | 13.6 tok/s |
GeForce RTX 5090 is a card, not a computer: its price is the US street median plus $1,000 for the minimum host machine. NVIDIA's $1,999 MSRP is not a price anyone pays.
memory = 70B × 0.820 B/param (6.56 bits) + 0.328 GB/1K × 8K K/V + 1.2 GB runtime
speed = bandwidth × 0.65 ÷ (70B active × 0.820 B/param) = 57.4 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 DGX Spark GB10 Grace Blackwell 128GB runs
- Qwen3-235B-A22BQ3_K_M, est. 12.7 tok/s
- Mistral-Large-2411Q6_K, est. 1.8 tok/s
- gpt-oss-120bQ6_K, est. 32.6 tok/s
- Llama-4-Scout-17B-16EQ8_0, est. 7.6 tok/s
- Llama-3.3-70BQ8_0, est. 2.4 tok/s
- Qwen3-32BBF16, est. 2.8 tok/s
Best quant that still fits at 8K context, largest model first. Full list on the Config page.
Bandwidth 273GB/s, usable memory 120GB of 128GB.