text model · Mistral · Windows
Can I run Devstral 2 123B on AMD Ryzen AI Halo (128GB)?
Yes. Devstral 2 123B runs on AMD Ryzen AI Halo (128GB) at Q4_K_M (~73.3 GB of ~96 GB usable).
Runs at Q4_K_M using ~73.3 GB of ~96 GB usable.
That figure is at a 4k context and moves about ±15% as context length changes. AMD Ryzen AI Halo (128GB) leaves ~22.7 GB of headroom.
- Q4_K_M needed
- ~73.3 GB
- Usable on device
- ~96 GB
- Device memory
- 128 GB
- Best quant
- Q4_K_M
Which quant fits
Running cost · estimate
- Power draw
- ~120 W
- Electricity / 1M tokens
- ~$2.5
At ~$0.15/kWh and the estimated ~2 tok/s, a million generated tokens costs about $2.5 in electricity. TDP is the peak draw, so this is an upper bound. Assumptions.
Run it
Pick your tool. All 3 load the same Q4_K_M weights.
ollama run devstral-2:123b llama-cli -hf unsloth/Devstral-2-123B-Instruct-2512-GGUF:Q4_K_M lms get unsloth/Devstral-2-123B-Instruct-2512-GGUF - Parameters
- 123B
- Q4_K_M size
- 69.75 GB
- Q8_0 size
- 123.73 GB
- Context
- 256k
- Ollama tag
- devstral-2:123b
- Memory
- 128 GB unified
- Usable for weights
- ~96 GB
- Power draw
- ~120 W
- Best runtime
- llama.cpp (Vulkan/ROCm) / LM Studio
You could also run
Run Devstral 2 123B on other hardware
FAQ
Can AMD Ryzen AI Halo (128GB) run Devstral 2 123B?
Yes. Devstral 2 123B runs on AMD Ryzen AI Halo (128GB) at Q4_K_M (~73.3 GB of ~96 GB usable).
How much memory does Devstral 2 123B need?
AMD Ryzen AI Halo (128GB) has room to spare. At Q4_K_M the weights are ~69.75 GB; with KV cache and runtime overhead, budget ~73.3 GB at a 4k context.
What is the best tool to run Devstral 2 123B on Windows?
LM Studio for a simple setup; Ollama (CUDA) for the most speed. AMD GPUs run via Vulkan/ROCm at roughly half CUDA throughput. NVIDIA is the smooth path on Windows.
Embed this
A live badge for your model card or README, updated as the data is. [](https://localmodel.run/can-i-run/devstral-2-123b/amd-ryzen-ai-halo-128gb) Sources
Weights are measured from GGUF files; KV cache and overhead are computed, so totals can vary ~15% with context and runtime. Any tok/s is a bandwidth estimate. See methodology.