text model · Yi · Windows
Can I run Yi 1.5 34B on Nvidia GeForce RTX 4090 (24GB)?
Yes. Yi 1.5 34B runs on Nvidia GeForce RTX 4090 (24GB) at Q4_K_M (~21.4 GB of ~23 GB usable).
Fits at Q4_K_M (~21.4 GB of ~23 GB usable) but with little headroom. Close other apps; a smaller context frees a few hundred MB.
That figure is at a 4k context and moves about ±15% as context length changes. Nvidia GeForce RTX 4090 (24GB) leaves ~1.6 GB of headroom.
- Q4_K_M needed
- ~21.4 GB
- Usable on device
- ~23 GB
- Device memory
- 24 GB
- Best quant
- Q4_K_M
Which quant fits
Running cost · estimate
- Power draw
- ~450 W
- Electricity / 1M tokens
- ~$0.55
At ~$0.15/kWh and the estimated ~34 tok/s, a million generated tokens costs about $0.55 in electricity. TDP is the peak draw, so this is an upper bound. Assumptions.
Run it
Pick your tool. All 2 load the same Q4_K_M weights.
llama-cli -hf bartowski/Yi-1.5-34B-Chat-GGUF:Q4_K_M lms get bartowski/Yi-1.5-34B-Chat-GGUF How to run it
On Windows use LM Studio (Best GUI on Windows, auto-detects CUDA/Vulkan backends.).
- Parameters
- 34B
- Q4_K_M size
- 19.24 GB
- Q8_0 size
- 34.03 GB
- Context
- 32k
- Memory
- 24 GB vram
- Usable for weights
- ~23 GB
- Power draw
- ~450 W
- Best runtime
- vLLM (Linux) / Ollama (CUDA)
You could also run
Run Yi 1.5 34B on other hardware
FAQ
Can Nvidia GeForce RTX 4090 (24GB) run Yi 1.5 34B?
Yes. Yi 1.5 34B runs on Nvidia GeForce RTX 4090 (24GB) at Q4_K_M (~21.4 GB of ~23 GB usable).
How much memory does Yi 1.5 34B need?
It is a tight fit on Nvidia GeForce RTX 4090 (24GB). At Q4_K_M the weights are ~19.24 GB; with KV cache and runtime overhead, budget ~21.4 GB at a 4k context.
What is the best tool to run Yi 1.5 34B 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/yi-1.5-34b/nvidia-rtx-4090-24gb) 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.