Mac Studio M3 Ultra (512GB) running DeepSeek V3 671B locally
Hardware Corner benchmarked DeepSeek-V3-0324 (4-bit, MLX) on a 512GB M3 Ultra Mac Studio: ~21 tok/s at tiny context, falling to ~5.8 tok/s at ~16K tokens, with 466GB peak memory.
Solo2-10 peopleDocument search / chat with filesWriting assistant$3k+Sensitivity: Very high (legal/medical/financial)
Hardware
| Item | Approx. price (as of 2026-10-16) |
|---|---|
| On sale −22% Apple Mac Studio M4 Max (14C CPU/32C GPU, 36GB, 512GB) - Apple Certified Refurbished (Apple Refurbished, related current offer) | $1,949.00 |
| On sale −17% Apple Mac Studio M4 Max (1TB) (Micro Center, related current offer) | $3,149.99 |
| Apple Mac Studio (M3 Ultra, 80-core GPU, 512GB unified memory) | $10,000 (as stated by source) |
Source prices are as stated by the source. Retailer offers: prices checked 2026-10-16, may change. Some links may be affiliate links; we may earn a commission at no extra cost to you. This never changes what we list.
Software stack
- MLX (mlx-lm)
Models used
- DeepSeek-V3-0324 (4-bit)
Good at
- Runs a 671B model on one quiet desktop
- 21.05 tok/s generation at 69-token context
Weak at
- At 15,777 tokens: 5.79 tok/s generation and 227s of prompt processing
- Expensive; long documents mean long waits
Rough total cost
$10,000 for the 512GB configuration, as stated by the source.
Getting started
- Install mlx-lm on macOS.
- Download DeepSeek-V3-0324 4-bit (~352GB per other reports).
- Run generation with MLX; keep prompts short for usable speed.
Source & credit
How Fast is Mac Studio M3 Ultra Running the New DeepSeek V3 LLM? — Hardware Corner
Summarized from the source; quotes are the source author's words. Check the original before buying.