text model · Mistral · Windows
Can I run Mistral Small 3 24B on Nvidia GeForce RTX 5090 (32GB)?
Yes. Mistral Small 3 24B runs on Nvidia GeForce RTX 5090 (32GB) at Q4_K_M (~16.3 GB of ~31 GB usable).
Runs at Q4_K_M using ~16.3 GB of ~31 GB usable. You have room for Q8_0 for higher quality.
That figure is at a 4k context and moves about ±15% as context length changes. Nvidia GeForce RTX 5090 (32GB) leaves ~14.7 GB of headroom, room to step up to Q8_0 for higher quality.
- Q4_K_M needed
- ~16.3 GB
- Usable on device
- ~31 GB
- Device memory
- 32 GB
- Best quant
- Q4_K_M
Which quant fits
Running cost · estimate
- Power draw
- ~575 W
- Electricity / 1M tokens
- ~$0.3
- Pays for itself after
- ~9,995M tok
At ~$0.15/kWh and the estimated ~81 tok/s, a million generated tokens costs about $0.3 in electricity. Versus a hosted API at ~$0.5 per million tokens, the ~$1,999 Nvidia GeForce RTX 5090 (32GB) pays for itself after roughly 9,995 million tokens, so local hardware is mostly a fixed cost, not a per-token one. 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 mistral-small3.2:24b llama-cli -hf bartowski/mistralai_Mistral-Small-3.2-24B-Instruct-2506-GGUF:Q4_K_M lms get bartowski/mistralai_Mistral-Small-3.2-24B-Instruct-2506-GGUF - Parameters
- 24B
- Q4_K_M size
- 14.33 GB
- Q8_0 size
- 25.05 GB
- Context
- 128k
- Ollama tag
- mistral-small3.2:24b
- Memory
- 32 GB vram
- Usable for weights
- ~31 GB
- Power draw
- ~575 W
- Best runtime
- vLLM (Linux) / Ollama (CUDA)
You could also run
Run Mistral Small 3 24B on other hardware
FAQ
Can Nvidia GeForce RTX 5090 (32GB) run Mistral Small 3 24B?
Yes. Mistral Small 3 24B runs on Nvidia GeForce RTX 5090 (32GB) at Q4_K_M (~16.3 GB of ~31 GB usable).
How much memory does Mistral Small 3 24B need?
Nvidia GeForce RTX 5090 (32GB) has room to spare. At Q4_K_M the weights are ~14.33 GB; with KV cache and runtime overhead, budget ~16.3 GB at a 4k context.
What is the best tool to run Mistral Small 3 24B 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/mistral-small-3-24b/nvidia-rtx-5090-32gb) 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.