← Interactive LLM VRAM Calculator — try your own context length & quantization

How much VRAM to run Qwen2.5 72B? (FP16/Q8/Q6/Q4 requirements)

Architecture: 80 layers, GQA with 8 KV heads. Figures are approximations for weights + an 8k-token FP16 KV cache + ~1 GB runtime overhead, based on typical published GGUF sizes. Actual usage varies by runtime and settings.

QuantizationWeightsTotal VRAM (8k context)Fits on (example)
FP16
full precision (16-bit)
~145 GB~149 GB2×A100 80 GB (160 GB)
Q8_0
8-bit
~77 GB~81 GB2×RTX A6000 (96 GB)
Q6_K
~6.6-bit
~64 GB~68 GBA100 / H100 (80 GB)
Q5_K_M
~5.7-bit
~54 GB~58 GB2×RTX 5090 (64 GB)
Q4_K_M
~4.8-bit
~47 GB~51 GB2×RTX 5090 (64 GB)

How much VRAM do I need to run Qwen2.5 72B?

You need roughly ~51 GB of VRAM to run Qwen2.5 72B at Q4_K_M with an 8k-token context — that fits on 2×RTX 5090 (64 GB). At Q8_0 plan for about ~81 GB, and full FP16 needs around ~149 GB. Qwen2.5 72B GGUF files are slightly larger than typical for the parameter count — Q4_K_M is ~47 GB of weights alone, so it does not quite fit in 48 GB with 8k context fully on-GPU.

What affects these numbers?

Three things: the quantization (bytes per parameter), the KV cache (grows linearly with context length — the table assumes 8k tokens), and runtime overhead. Longer contexts need more memory; quantizing the KV cache (e.g. q8_0 cache in llama.cpp) roughly halves that part. Use the interactive calculator to plug in your exact context length and quantization.