What can NVIDIA RTX PRO 6000 Blackwell run for chat?

Build: RTX PRO 6000 Blackwell + Threadripper PRO + 128GB

Memory: 96 GB VRAM + 128 GB system RAM
Runner: llama.cpp / Ollama (CUDA)

Runs comfortably
255 models

Ranked by fit for chat use case + predicted speed. Click a row for VRAM breakdown.

#1Qwen 3 0.6B
0.6B
qwen
Commercial OK
Quant: Q4_K_MContext: 8,192VRAM: 2.4 GBHeadroom: 93.6 GB
3215
tok/s
Estimated
Weights
0.30 GB
KV cache
0.30 GB
Activations
0.02 GB
Runtime
1.80 GB
#2Qwen 3 8B
8B
qwen
Commercial OK
Quant: Q8_0Context: 8,192VRAM: 14.4 GBHeadroom: 81.6 GB
ollama run qwen3:8b
137
tok/s
Estimated
Weights
8.20 GB
KV cache
4.00 GB
Activations
0.42 GB
Runtime
1.80 GB
#3Llama 3.2 3B Instruct
3B
llama
Commercial OK
Quant: Q8_0Context: 8,192VRAM: 6.9 GBHeadroom: 89.1 GB
ollama run llama3.2:3b
365
tok/s
Estimated
Weights
3.40 GB
KV cache
1.50 GB
Activations
0.18 GB
Runtime
1.80 GB
#4Qwen 3 1.7B
1.7B
qwen
Commercial OK
Quant: Q4_K_MContext: 8,192VRAM: 3.6 GBHeadroom: 92.4 GB
1135
tok/s
Estimated
Weights
0.90 GB
KV cache
0.85 GB
Activations
0.05 GB
Runtime
1.80 GB
#5Dolphin 3.0 8B
8B
dolphin
Commercial OK
Quant: Q4_K_MContext: 8,192VRAM: 11.1 GBHeadroom: 84.9 GB
ollama run dolphin3:8b
241
tok/s
Estimated
Weights
5.00 GB
KV cache
4.00 GB
Activations
0.26 GB
Runtime
1.80 GB
#6TinyLlama 1.1B Chat v1.0
1.1B
llama
Commercial OK
Quant: Q4_K_MContext: 2,048VRAM: 2.6 GBHeadroom: 93.4 GB
1754
tok/s
Estimated
Weights
0.60 GB
KV cache
0.14 GB
Activations
0.03 GB
Runtime
1.80 GB
#7Gemma 2 2B Instruct
2B
gemma
Commercial OK
Quant: Q4_K_MContext: 8,192VRAM: 4.0 GBHeadroom: 92.0 GB
965
tok/s
Estimated
Weights
1.10 GB
KV cache
1.00 GB
Activations
0.06 GB
Runtime
1.80 GB
#8Mistral 7B Instruct v0.3
7B
mistral
Commercial OK
Quant: Q5_K_MContext: 8,192VRAM: 10.7 GBHeadroom: 85.3 GB
ollama run mistral:7b
242
tok/s
Estimated
Weights
5.10 GB
KV cache
3.50 GB
Activations
0.26 GB
Runtime
1.80 GB
#9DeepSeek V2 Lite Chat
15.7B
deepseek
Commercial OK
Quant: Q4_K_MContext: 8,192VRAM: 18.7 GBHeadroom: 77.3 GB
804
tok/s
Estimated
Weights
8.60 GB
KV cache
7.85 GB
Activations
0.44 GB
Runtime
1.80 GB
#10Gemma 2 9B Instruct
9B
gemma
Commercial OK
Quant: Q8_0Context: 8,192VRAM: 16.6 GBHeadroom: 79.4 GB
ollama run gemma2:9b
122
tok/s
Estimated
Weights
9.80 GB
KV cache
4.50 GB
Activations
0.50 GB
Runtime
1.80 GB
Quant: Q4_K_MContext: 4,096VRAM: 7.6 GBHeadroom: 88.4 GB
276
tok/s
Estimated
Weights
3.90 GB
KV cache
1.75 GB
Activations
0.20 GB
Runtime
1.80 GB
#12Kumru 2B
2.4B
mistral
Commercial OK
Quant: Q4_K_MContext: 8,192VRAM: 4.6 GBHeadroom: 91.4 GB
ollama run alibayram/kumru:latest
804
tok/s
Estimated
Weights
1.50 GB
KV cache
1.20 GB
Activations
0.08 GB
Runtime
1.80 GB

Runs with tradeoffs
12 models

Tight VRAM, partial CPU offload, or context-limited.

DeepSeek V2.5 236B
236B
deepseek
Commercial OK
Quant: Q4_K_MContext: 2,048VRAM: 172.0 GBHeadroom: 0.8 GB
  • Partial CPU offload: ~44% of layers run on CPU
  • CPU is the bottleneck — upgrading RAM bandwidth helps more than VRAM here
29
tok/s
Estimated
Weights
134.00 GB
KV cache
29.50 GB
Activations
6.70 GB
Runtime
1.80 GB
WizardLM-2 8x22B
141B
wizard
Commercial OK
Quant: Q4_K_MContext: 8,192VRAM: 160.5 GBHeadroom: 12.3 GB
  • Partial CPU offload: ~40% of layers run on CPU
  • CPU is the bottleneck — upgrading RAM bandwidth helps more than VRAM here
5
tok/s
Estimated
Weights
84.00 GB
KV cache
70.50 GB
Activations
4.21 GB
Runtime
1.80 GB
DeepSeek Coder V2 236B
236B
deepseek
Commercial OK
Quant: Q4_K_MContext: 2,048VRAM: 172.0 GBHeadroom: 0.8 GB
  • Partial CPU offload: ~44% of layers run on CPU
  • CPU is the bottleneck — upgrading RAM bandwidth helps more than VRAM here
29
tok/s
Estimated
Weights
134.00 GB
KV cache
29.50 GB
Activations
6.70 GB
Runtime
1.80 GB
GLM-5 Pro
144B
glm
Quant: AWQ-INT4Context: 8,192VRAM: 159.9 GBHeadroom: 12.9 GB
  • Partial CPU offload: ~40% of layers run on CPU
  • CPU is the bottleneck — upgrading RAM bandwidth helps more than VRAM here
25
tok/s
Estimated
Weights
82.00 GB
KV cache
72.00 GB
Activations
4.11 GB
Runtime
1.80 GB
Quant: Q4_K_MContext: 2,048VRAM: 167.6 GBHeadroom: 5.2 GB
  • Partial CPU offload: ~43% of layers run on CPU
  • CPU is the bottleneck — upgrading RAM bandwidth helps more than VRAM here
3
tok/s
Estimated
Weights
129.80 GB
KV cache
29.50 GB
Activations
6.49 GB
Runtime
1.80 GB
DBRX Base
132B
dbrx
Commercial OK
Quant: Q4_K_MContext: 8,192VRAM: 146.6 GBHeadroom: 26.2 GB
  • Partial CPU offload: ~34% of layers run on CPU
  • CPU is the bottleneck — upgrading RAM bandwidth helps more than VRAM here
20
tok/s
Estimated
Weights
75.00 GB
KV cache
66.00 GB
Activations
3.76 GB
Runtime
1.80 GB
Nemotron 3 Super (120B-A12B)
120B
other
Commercial OK
Quant: Q4_K_MContext: 2,048VRAM: 92.4 GBHeadroom: 3.6 GB
  • Tight VRAM fit — only 3.6 GB headroom left for context growth
ollama run nemotron3:super
16
tok/s
Estimated
Weights
72.00 GB
KV cache
15.00 GB
Activations
3.60 GB
Runtime
1.80 GB
Quant: Q4_K_MContext: 2,048VRAM: 93.8 GBHeadroom: 2.2 GB
  • Tight VRAM fit — only 2.2 GB headroom left for context growth
ollama run mistral-large:123b
16
tok/s
Estimated
Weights
73.00 GB
KV cache
15.38 GB
Activations
3.65 GB
Runtime
1.80 GB

What if you upgraded?

Hypothetical scenarios. We re-ran the compatibility engine for each.

+32 GB system RAM

~$80–150

Doubles your CPU-offload working set. Helps when models don't quite fit in VRAM.

Unlocks: 9 new comfortable, 14 new tradeoff

  • Gemma 3 270M
  • SmolLM2 135M Instruct
  • SmolLM2 360M Instruct
  • VBART Large (Turkish Summarization)

Upgrade to NVIDIA H200 NVL (PCIe)

~$32000

141 GB VRAM (vs your 96 GB) plus a bandwidth jump from ~1792 GB/s to ~4800 GB/s.

Unlocks: 14 new comfortable

  • Gemma 3 270M
  • SmolLM2 135M Instruct
  • Mixtral 8x22B Instruct
  • WizardLM-2 8x22B

Add a second NVIDIA RTX PRO 6000 Blackwell

~$8999

Tensor parallelism splits the model across both cards, effectively doubling VRAM. Bandwidth doesn't double — runs ~1.5× the single-card speed in practice.

Unlocks: 23 new comfortable

  • Qwen 3 235B-A22B
  • GLM-5
  • Gemma 3 270M
  • SmolLM2 135M Instruct

Some links above are affiliate links. We may earn a commission at no extra cost to you. How we make money.

Won't run
top 5 popular models

Need more memory than you have. Shown for orientation.

DeepSeek V4 Pro (1.6T MoE)
1600B
deepseek
Commercial OK

Even with CPU offload, needs more memory than your VRAM (96 GB) + 60% of system RAM (77 GB) combined.

Qwen 3.5 235B-A17B (MoE)
397B
qwen
Commercial OK

Even with CPU offload, needs more memory than your VRAM (96 GB) + 60% of system RAM (77 GB) combined.

Qwen 3 235B-A22B
235B
qwen
Commercial OK

Even with CPU offload, needs more memory than your VRAM (96 GB) + 60% of system RAM (77 GB) combined.

DeepSeek R1 (671B reasoning)
671B
deepseek
Commercial OK

Even with CPU offload, needs more memory than your VRAM (96 GB) + 60% of system RAM (77 GB) combined.

DeepSeek V4 Flash (284B MoE)
284B
deepseek
Commercial OK

Even with CPU offload, needs more memory than your VRAM (96 GB) + 60% of system RAM (77 GB) combined.

How to read these numbers

Measured here
Measured here - RunLocalAI ran this exact combo on owner hardware with public evidence.

Source-backed
Source-backed / community - a reproduced public source supports the speed, but it is not labeled as owner-measured.

Extrapolated
Extrapolated - predicted from a measured benchmark on similar-bandwidth hardware.

Estimated
Estimated - formula based on VRAM bandwidth and model architecture; not a benchmark row.

RunLocalAI Will-It-Run Framework →

Want a specific benchmark we don't have? Email Contact support and we'll prioritize it.