141 GB VRAM (+45)4800 GB/s (+800)
C
Adds memory headroom for longer context windows and future model growth.62.7 tok/s decode
Adds memory headroom for longer context windows and future model growth.
~$30,000 MSRP
Can Qwen3.5 122B A10B run on NVIDIA GH200 96GB?
YES — Tight Fit
C52Usable
○Estimated from fit model
Qwen3.5 122B A10B needs ~84.9 GB VRAM. NVIDIA GH200 96GB has 96.0 GB. With Q3_K_M quantization, expect ~50 tok/s.
Runtime: OllamaCapacity: TightBandwidth: HighStack: BasicBottleneck: Balanced
Share:
Operating mode
Choose the run profile you care about
Interactive favors responsiveness, while light API and scale-out lean harder on serving readiness. The fit stays the same, but the recommendation lens changes.
Current mode
Balanced
Balanced for general local use. Keeps the ranking neutral across personal and serving workflows.
Select quantization to explore
F16 (Maximum quality) — 275.2 GB, exceeds 96.0 GB available
275.2 GB required96.0 GB available
287% VRAM needed
179.2 GB over capacity — needs offload or smaller quantization
Fit status
Too heavy
Decode
2.7 tok/s
TTFT
71136 ms
Safe context
4K
Memory
275.2 GB / 96.0 GB
Offload
70%
Memory breakdown
Weights250.1 GB
KV Cache14.3 GB
Runtime1.2 GB
Headroom9.6 GB
See how fast it feels
With memory offload — actual speed may be lowerSee how fast it feelsQwen3.5 122B A10B on NVIDIA GH200 96GB
1st promptCold start — includes initialization
>What is local AI inference?
Local AI inference means running an AI language model directly on your own hardware — your laptop, desktop, or server — instead of sending requests to a remote cloud API.
When you run inference locally the model weights are loaded into your GPU or unified memory. Each token you generate requires reading those weights from memory, so memory bandwidth is the main bottleneck for decode speed.
Key benefits of running locally:
- Full privacy: your prompts never leave your machine
- No per-token cost or rate limits
- Works offline once the model is downloaded
- Latency depends only on your hardware
2nd promptWarm — model ready, no init overhead
>How much VRAM do I need?
It depends on the model size and quantization level. A rough rule of thumb:
Model size Q4 (4-bit) Q8 (8-bit) FP16
7B params ~4.3 GB ~7.5 GB ~14 GB
13B params ~7.9 GB ~13.9 GB ~26 GB
70B params ~42.7 GB ~74.9 GB ~140 GB
Most people use 4-bit quantization (Q4_K_M) which gives 90-95% of full quality at a fraction of the memory. A 24 GB GPU can comfortably run most 7B-13B models.
Estimated: 2.7 tok/s decode · 71.1s TTFT (warm) · 7 tok/s prefill
What limits this setup
This setup is broadly balanced for this model.
No major red flags
This recommendation has enough memory headroom and acceptable estimated speed for the selected workload.
Best improvement path
Performance by workload
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Chat | B | Runs well | 50.4 tok/s | 2095 ms | 28K |
| Coding | C | Tight fit | 50.4 tok/s | 3841 ms | 28K |
| Agentic Coding | C | Runs with offload (needs ~1.9 GB host RAM) | 40.7 tok/s | 6911 ms | 28K |
| Reasoning | C | Tight fit | 50.4 tok/s | 4539 ms | 28K |
| RAG | C | Runs with offload (needs ~1.9 GB host RAM) | 40.7 tok/s | 8639 ms |
Quantization options
How Qwen3.5 122B A10B (122B params) fits at each quantization level on NVIDIA GH200 96GB (96.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 47.6 GB | Low | C48 |
Q3_K_S | 3 | 59.8 GB | Low | C48 |
NVFP4 | 4 |
Get started
Copy-paste commands to run Qwen3.5 122B A10B on your machine.
Run
docker run --rm -it ghcr.io/ggerganov/llama.cpp:full \
--hf-repo "unsloth/Qwen3.5-122B-A10B-GGUF" \
--hf-file "Qwen3.5-122B-A10B-GGUF-Q3_K_M.gguf" \
-c 4096 -ngl 99Upgrade options
Hardware that runs Qwen3.5 122B A10B well
141 GB VRAM (+45)4800 GB/s (+800)
C
Adds memory headroom for longer context windows and future model growth.62.7 tok/s decode
Adds memory headroom for longer context windows and future model growth.
~$30,000 MSRP
180 GB VRAM (+84)8000 GB/s (+4000)
C
Raises estimated decode speed by about 107%.104.5 tok/s decode
Raises estimated decode speed by about 107%.
Adds memory headroom for longer context windows and future model growth.
~$30,000 MSRP