Local AI Deals

Cheapest box to run DeepSeek-V3 at Q8_0

DeepSeek’s DeepSeek-V3 is 671B parameters, 37B of them read per token. 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.

Fit finderspecs, not prices
Quant
Context

DeepSeek-V3 at Q8_0 needs 715GB.

Nothing on the Grid runs DeepSeek-V3 at Q8_0 today. The largest Config, Mac Studio, M5 Ultra, 512GB, is 211GB short. At it fits with 174GB headroom.

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 = 671B × 1.063 B/param (8.5 bits) + 0.070 GB/1K × 8K K/V + 1.2 GB runtime
speed = bandwidth × 0.5 ÷ (37B active × 1.063 B/param) = 39.3 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.