Local AI Deals

Cheapest box to run DeepSeek-V4-Pro at Q8_0

DeepSeek’s DeepSeek-V4-Pro is 1600B parameters, 49B 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-V4-Pro at Q8_0 needs 1701GB.

Nothing on the Grid runs DeepSeek-V4-Pro at Q8_0 today. The largest Config, Mac Studio, M5 Ultra, 512GB, is 1197GB short.

The same compressed sparse attention. 1.6 trillion parameters means no Config we track holds it, and we list it so the arithmetic says so plainly.

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 = 1600B × 1.063 B/param (8.5 bits) + 0.009 GB/1K × 8K K/V + 1.2 GB runtime
speed = bandwidth × 0.5 ÷ (49B active × 1.063 B/param) = 52.1 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.