text model · Mistral · Windows
Can I run Devstral 2 123B on Nvidia GeForce RTX 3060 Ti (8GB)?
No. Devstral 2 123B needs ~73.3 GB even at Q4_K_M, but Nvidia GeForce RTX 3060 Ti (8GB) only has ~7 GB usable.
Needs ~73.3 GB even at Q4_K_M, but only ~7 GB is usable.
That figure is at a 4k context and moves about ±15% as context length changes.
The gap is about 66.3 GB: Devstral 2 123B needs roughly 73.3 GB at Q4_K_M and Nvidia GeForce RTX 3060 Ti (8GB) leaves only about 7 GB usable for a model. The lightest tracked hardware that runs Devstral 2 123B is the Apple M4 Max (128GB) at 128 GB. See Devstral 2 123B on Apple M4 Max (128GB).
Too big for Nvidia GeForce RTX 3060 Ti (8GB)'s VRAM, but you could offload some layers to system RAM and still run it at roughly ~0.5 tok/s, far slower than a model that fits. How this is estimated.
- Q4_K_M needed
- ~73.3 GB
- Usable on device
- ~7 GB
- Device memory
- 8 GB
Which quant fits
- Parameters
- 123B
- Q4_K_M size
- 69.75 GB
- Q8_0 size
- 123.73 GB
- Context
- 256k
- Ollama tag
- devstral-2:123b
- Memory
- 8 GB vram
- Usable for weights
- ~7 GB
- Power draw
- ~200 W
- Best runtime
- Ollama (CUDA) / llama.cpp CUDA
What you can run instead
Run Devstral 2 123B on other hardware
FAQ
Can Nvidia GeForce RTX 3060 Ti (8GB) run Devstral 2 123B?
No. Devstral 2 123B needs ~73.3 GB even at Q4_K_M, but Nvidia GeForce RTX 3060 Ti (8GB) only has ~7 GB usable.
How much memory does Devstral 2 123B need?
Nvidia GeForce RTX 3060 Ti (8GB) does not have enough memory. 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/nvidia-rtx-3060-ti-8gb) 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.