Sunk Cost sunkcost.ai Data checked 2026-09-03

Can a NVIDIA RTX PRO 6000 Blackwell, 96GB run local LLMs?

Yes — 33 of the 39 open models on this site fit in its 95 GB of usable memory, the strongest being Qwen3.8 27B. Whether that saves you money is a different question, and the answer is usually no.

Price$18,000
Memory96 GB, about 95 GB of it addressable by the GPU at 1792 GB/s
Best model it runsQwen3.8 27B — Sonnet-class, 78 tok/s
Pay-back against the APIPays back in 227 years at 500k tokens a day

Run the numbers on this machine

What it runs

ModelSpeedClassGood atMemory
Qwen3.8 27BQ4_K_M 78 tok/s Sonnet-class 19 GB
Ling 3.0 flashQ4_K_M 80 tok/s Haiku-class 82 GB
Qwen3.6 27BQ4_K_M 77 tok/s Haiku-class 19 GB
Qwen3.6 35B-A3BQ4_K_M 221 tok/s Haiku-class 23 GB
Muse Glimmer 30BQ4_K_M 81 tok/s Haiku-class 18 GB
Gemma 4 26B-A4BQ4_K_M 162 tok/s Haiku-class 18 GB
Qwen3.5 122B-A10BUD-Q4_K_M 79 tok/s Haiku-class 79 GB
Gemma 4 31B itQ4_K_M 56 tok/s Haiku-class 26 GB
Granite 4.2 30BQ4_K_M 55 tok/s Haiku-class 26 GB
GLM-4.7-FlashQ4_K_M 157 tok/s Haiku-class 20 GB
Gemma 4 12BQ4_K_M 182 tok/s Below every hosted tier 8.0 GB
Qwen3.5 9BQ4_K_M 215 tok/s Below every hosted tier 6.8 GB

21 more fit; the calculator lists them all.

The specifics

Chip
RTX PRO 6000 Blackwell — Blackwell · 96 GB GDDR7 ECC
Memory bandwidth
1792 GB/s
Usable by the GPU
95 GB — Bandwidth: 28 Gbps × 512-bit bus ÷ 8 = 1,792 GB/s (Puget Systems; NVIDIA lists 1792 GB/sec). Usable memory is the VRAM less 1 GB: the margin llama.cpp's automatic fitting (--fit) leaves free on each device by default (common/common.h, fit_params_target = 1 GiB). Treated as nominal GB, like the other machines here.
Power under load
600 W (published) — 600 W is the maker's rated board power for the card alone. The rest of the PC adds more, and LLM decoding usually draws less than the full rating. One llama.cpp user saw about 390 W while benchmarking gpt-oss-120b on this card.
Availability
Newegg price on 2026-09-16; NVIDIA publishes no list price. Not the 300 W Max-Q version.
Sources
source 1, source 2, source 3, source 4, source 5, source 6