Qwen 2.5 72B VRAM Requirements
Qwen 2.5 72B needs about 44.5 GB of VRAM at Q4_K_M with an 8K context: 41.0 GB for weights, 2.5 GB for the KV cache and 0.9 GB of runtime overhead. At Q8_0 it needs 75.4 GB and at full FP16 precision 139 GB. The smallest GPU that runs it fully is the Apple M5 Pro 64GB.
44.5 GB VRAM
Qwen 2.5 72B at Q4_K_M with 8K context and the KV cache in FP16.
- Weights 41.0 GB
- KV cache 2.5 GB
- Runtime overhead 0.9 GB
Which GPUs can run it
- RTX 3060 12GBToo small
- RTX 4060 Ti 16GB34% on GPU~1.3 tok/s
- RTX 3090 24GB51% on GPU~1.8 tok/s
- Radeon RX 7900 XTX 24GB51% on GPU~1.7 tok/s
- RTX 4090 24GB51% on GPU~1.8 tok/s
- RTX 5090 32GB68% on GPU~2.7 tok/s
- RTX 6000 Ada 48GBFits~14 tok/s
- A100 80GBFits~29 tok/s
- H100 80GB SXMFits~46 tok/s
- Ryzen AI Max+ 395 128GBFits~3.4 tok/s
- Apple M4 Max 128GBFits~7.8 tok/s
- Apple M5 Max 128GBFits~8.7 tok/s
- RTX PRO 6000 Blackwell 96GBFits~25 tok/s
- NVIDIA DGX Spark 128GBFits~3.6 tok/s
- Apple M3 Ultra 512GBFits~12 tok/s
Multi-GPU: about 3× 24 GB cards (RTX 3090 / 4090) with the model split across them.
Qwen 2.5 72B VRAM by quantization
Total memory with an FP16 KV cache. "Smallest tier" is the smallest standard VRAM size with a 6% safety margin at 8K.
| Quantization | Bits/weight | Weights | Total at 8K | Total at 32K | Smallest tier |
|---|---|---|---|---|---|
| Q4_K_M | 4.85 | 41.0 GB | 44.5 GB | 52.0 GB | 48 GB |
| Q5_K_M | 5.69 | 48.2 GB | 51.6 GB | 59.1 GB | 64 GB |
| Q6_K | 6.59 | 55.8 GB | 59.2 GB | 66.7 GB | 64 GB |
| Q8_0 | 8.50 | 71.9 GB | 75.4 GB | 82.9 GB | 96 GB |
| FP16 / BF16 | 16.00 | 135 GB | 139 GB | 146 GB | Multi-GPU / server |
| Q3_K_M | 3.91 | 33.1 GB | 36.5 GB | 44.0 GB | 48 GB |
| Q2_K | 3.00 | 25.4 GB | 28.8 GB | 36.3 GB | 32 GB |
VRAM by context length
At Q4_K_M. The KV cache grows linearly with context: 0.31 GB per 1,000 tokens in FP16.
| Context | KV cache (FP16) | KV cache (Q8_0) | Total, FP16 cache | Total, Q8_0 cache |
|---|---|---|---|---|
| 4K tokens | 1.3 GB | 0.7 GB | 43.2 GB | 42.6 GB |
| 8K tokens | 2.5 GB | 1.3 GB | 44.5 GB | 43.3 GB |
| 32K tokens | 10.0 GB | 5.3 GB | 52.0 GB | 47.3 GB |
GPUs that can run Qwen 2.5 72B
At Q4_K_M with 8K context. "Best quant" is the highest-quality format that fits fully; "max context" is at Q4_K_M.
| GPU | Usable memory | Status | Best quant | Max context | Speed |
|---|---|---|---|---|---|
| Apple M5 Pro 64GB | 48.0 GB | Fits | Q4_K_M | 10K | ~4.4 tok/s |
| Apple M1 Max 64GB | 48.0 GB | Fits | Q4_K_M | 10K | ~5.7 tok/s |
| Apple M4 Max 64GB | 48.0 GB | Fits | Q4_K_M | 10K | ~7.8 tok/s |
| RTX A6000 48GB | 48.0 GB | Fits | Q4_K_M | 10K | ~11 tok/s |
| L40S 48GB | 48.0 GB | Fits | Q4_K_M | 10K | ~12 tok/s |
| RTX 6000 Ada 48GB | 48.0 GB | Fits | Q4_K_M | 10K | ~14 tok/s |
| A100 80GB | 80.0 GB | Fits | Q6_K | 32K | ~29 tok/s |
| H100 80GB SXM | 80.0 GB | Fits | Q6_K | 32K | ~46 tok/s |
| Ryzen AI Max+ 395 128GB | 96.0 GB | Fits | Q8_0 | 32K | ~3.4 tok/s |
| Apple M3 Max 128GB | 96.0 GB | Fits | Q8_0 | 32K | ~5.7 tok/s |
| Apple M4 Max 128GB | 96.0 GB | Fits | Q8_0 | 32K | ~7.8 tok/s |
| Apple M5 Max 128GB | 96.0 GB | Fits | Q8_0 | 32K | ~8.7 tok/s |
| RTX PRO 6000 Blackwell 96GB | 96.0 GB | Fits | Q8_0 | 32K | ~25 tok/s |
| NVIDIA DGX Spark 128GB | 112 GB | Fits | Q8_0 | 32K | ~3.6 tok/s |
| A100 40GB | 40.0 GB | 85% on GPU | Q3_K_M | – | ~4.9 tok/s |
| RTX 5090 32GB | 32.0 GB | 68% on GPU | Q2_K | – | ~2.7 tok/s |
| Radeon AI PRO R9700 32GB | 32.0 GB | 68% on GPU | Q2_K | – | ~2.3 tok/s |
| RTX 3090 24GB | 24.0 GB | 51% on GPU | – | – | ~1.8 tok/s |
| RTX 3090 Ti 24GB | 24.0 GB | 51% on GPU | – | – | ~1.8 tok/s |
| RTX 4090 24GB | 24.0 GB | 51% on GPU | – | – | ~1.8 tok/s |
Multi-GPU: about 3× 24 GB cards (RTX 3090 or 4090) run it at Q4_K_M, using tensor or layer splitting in llama.cpp, vLLM or ExLlamaV2.
About Qwen 2.5 72B
Qwen's largest dense 2.5 model, with the same layer and head layout as Llama 70B. Memory needs are slightly higher than Llama 3.3 70B because of the larger vocabulary and wider feed-forward layers.
| Spec | Value |
|---|---|
| Parameters | 72.7B |
| Layers | 80 |
| Hidden size | 8192 |
| Attention heads / KV heads | 64 / 8 |
| Head dimension | 128 |
| Vocabulary | 152,064 |
| Max context (used here) | 32K tokens |
| KV cache per token (FP16) | 320 KB |
| Released | 2024-09 |
| License | Qwen License |
Qwen 2.5 72B VRAM: common questions
Is Qwen 2.5 72B harder to run than Llama 3.3 70B?
Barely. It is about 1 GB larger at Q4_K_M and has the same KV cache size per token.
How much VRAM does Qwen 2.5 72B need?
About 44.5 GB at Q4_K_M with an 8K context, 75.4 GB at Q8_0 and 139 GB at FP16. Longer contexts add 0.31 GB per 1,000 tokens with an FP16 cache.
What is the cheapest GPU that can run Qwen 2.5 72B?
The smallest listed device that fits it fully at Q4_K_M with 8K context is the Apple M5 Pro 64GB (48.0 GB usable). With 24 GB cards you need about 3 of them.