Question 1

Can the CPU-only (DDR5 tuned / 4-channel) run DEEPSEEK-R1 distill 14B?

Accepted Answer

DEEPSEEK-R1 distill 14B needs about 9.75GB working set (weights 9GB + KV cache 0.75GB at 4k context). The CPU-only (DDR5 tuned / 4-channel) offers ~30GB usable — verdict: full. 9.8 GB needed of 30 GB usable — headroom for context

Question 2

How fast is DEEPSEEK-R1 distill 14B on the CPU-only (DDR5 tuned / 4-channel)?

Accepted Answer

~9–14 tok/s on the CPU-only (DDR5 tuned / 4-channel). computed roofline: 179.2 GB/s × 0.45–0.70 efficiency window / 9.0GB (Q4_K_M) — bands, never points.

Question 3

What should I run DEEPSEEK-R1 distill 14B on instead?

Accepted Answer

Tesla P100 (16GB, used) runs it fully ($135).

Can the CPU-only (DDR5 tuned / 4-channel) run DEEPSEEK-R1 distill 14B?

Run DEEPSEEK-R1 distill 14B on instead