Radeon PRO W7900
AMD
CompatibleRuns Qwen2.5 72B Instruct Q4_K_M entirely in memory
Model · Alibaba Qwen
Qwen2.5 72B Instruct: VRAM and hardware requirements
At Q4_K_M with an 8K context, Qwen2.5 72B Instruct needs about 47.1 GB: a GPU with 80 GB or more runs it entirely in VRAM.
- Parameters72.7B
- Max context128K
- LicenseQwen License
Specifications
- Organization
- Alibaba Qwen
- Architecture
- qwen2
- Released
- September 2024
- Layers
- 80
- Hidden size
- 8192
- Attention heads
- 64
- KV heads
- 8
- Head dimension
- 128
- Vocabulary
- 152,064
- KV cache per 1K tokens (f16)
- 0.313 GB
- Repository
- Qwen/Qwen2.5-72B-Instruct
Official specData updated 24 days agoSource: Hugging Face config.json + model card
Quantizations and memory
| Quantization | File size | Memory needed | Minimum GPU memory | Quality |
|---|---|---|---|---|
| BF16 | Not available | ~146 GB | – | Reference (unquantized) |
| Q8_0 | 72.2 GB | ~79.1 GB | 96 GB | Near-lossless |
| Q6_K | 55.7 GB | ~61.9 GB | 80 GB | Near-lossless |
| Q5_K_M | 48.1 GB | ~54.2 GB | 80 GB | Very good |
| Q4_K_MMost used | 41 GB | ~47.1 GB | 80 GB | Good — the usual choice |
| IQ4_XS | 37 GB | ~43.2 GB | 48 GB | Good — the usual choice |
| Q3_K_M | 33 GB | ~39.1 GB | 48 GB | Noticeable loss |
| Q2_K | 25.5 GB | ~31.7 GB | 32 GB | Significant loss |
File sizes: not available until imported from the model repository. Memory = weights (estimated from bits per weight) + KV cache + runtime overhead + safety margin. (BF16 ≈ 16 bpw, Q8_0 ≈ 8.5 bpw, Q6_K ≈ 6.57 bpw, Q5_K_M ≈ 5.7 bpw, Q4_K_M ≈ 4.9 bpw, IQ4_XS ≈ 4.46 bpw, Q3_K_M ≈ 4 bpw, Q2_K ≈ 3.17 bpw)
Q4, Q5 or Q8: how to choose a quantization
GPUs that run Qwen2.5 72B Instruct
| RTX 509032 GB · 64 GB RAM~46.6 GB · 1.3–2.5 tok/s estimate | Possible but slow |
|---|---|
| RTX 409024 GB · 64 GB RAM~46.6 GB · 0.8–1.6 tok/s estimate | Not recommended |
| RTX 309024 GB · 64 GB RAM~46.6 GB · 0.8–1.6 tok/s estimate | Not recommended |
| RX 7900 XTX24 GB · 64 GB RAM~46.6 GB · 0.8–1.6 tok/s estimate | Not recommended |
| RTX 5060 Ti 16 GB16 GB · 64 GB RAM~46.6 GB · 0.6–1.2 tok/s estimate | Not recommended |
| RTX 5070 Ti16 GB · 64 GB RAM~46.6 GB · 0.6–1.2 tok/s estimate | Not recommended |
| RTX 508016 GB · 64 GB RAM~46.6 GB · 0.6–1.2 tok/s estimate | Not recommended |
| RTX 4060 Ti 16 GB16 GB · 64 GB RAM~46.6 GB · 0.6–1.2 tok/s estimate | Not recommended |
| RTX 4070 Ti SUPER16 GB · 64 GB RAM~46.6 GB · 0.6–1.2 tok/s estimate | Not recommended |
| RX 9070 XT16 GB · 64 GB RAM~46.6 GB · 0.6–1.2 tok/s estimate | Not recommended |
| RTX 3060 12 GB12 GB · 64 GB RAM~46.6 GB · 0.6–1.1 tok/s estimate | Not recommended |
| RTX 407012 GB · 64 GB RAM~46.6 GB · 0.6–1.1 tok/s estimate | Not recommended |
| RTX 507012 GB · 64 GB RAM~46.6 GB · 0.6–1.1 tok/s estimate | Not recommended |
| RTX 40608 GB · 64 GB RAM~46.6 GB · 0.5–1 tok/s estimate | Not recommended |
Macs that run Qwen2.5 72B Instruct
| M3 Ultra (80-core GPU) · 512 GBMac Studio~46.4 GB · 11–17 tok/s estimate | Excellent |
|---|---|
| M4 Max (40-core GPU) · 128 GBMacBook Pro 14, MacBook Pro 16~46.4 GB · 7.5–11 tok/s estimate | Compatible |
| M3 Max (40-core GPU) · 128 GBMacBook Pro 14, MacBook Pro 16~46.4 GB · 5.5–8 tok/s estimate | Compatible |
| M4 Pro (20-core GPU) · 64 GBMac mini, MacBook Pro 14~46.4 GB · 4–5.5 tok/s estimate | Compatible |
| M4 Max (32-core GPU) · 36 GBMacBook Pro 14, MacBook Pro 16~46.4 GB · No data | Incompatible |
| M4 · 32 GBMacBook Air, MacBook Pro 14~46.4 GB · No data | Incompatible |
| M5 · 32 GBMacBook Pro 14~46.4 GB · No data | Incompatible |
| M3 · 24 GBMacBook Air, MacBook Pro 14~46.4 GB · No data | Incompatible |
EstimateData updated 24 days agoSource: Run LLM Locally engine
Limits in plain words
The KV cache grows with context: at 32K it adds about 10 GB on top of the weights.
With under 8 GB of VRAM, use a smaller quantization or a smaller model of the same family.
Where to buy
Prices and links follow the market you choose.
AMD Ryzen AI Max+ 395
Unified memory PC
CompatibleRuns Qwen2.5 72B Instruct Q4_K_M entirely in memory
GeForce RTX 5090
Lowest memory tier
ExcellentRuns Qwen2.5 72B Instruct Q4_K_M entirely in memory
No current offer imported for this market: the link opens a marketplace search.
Frequently asked questions
How much VRAM does Qwen2.5 72B Instruct need?
About 47.1 GB at Q4_K_M with an 8K context, including the KV cache and runtime overhead. At Q8_0 it needs about 79.1 GB. A longer context adds more: about 10 GB of KV cache at 32K.
Can Qwen2.5 72B Instruct run on a Mac?
Yes on a Mac whose GPU can use at least 46.4 GB, e.g. M4 Pro (20-core GPU) 64 GB. macOS lets the GPU use about two thirds to three quarters of unified memory by default.
How fast is Qwen2.5 72B Instruct?
On a RTX 4090, we estimate 0.8–1.6 tok/s at Q4_K_M. This is an estimate from memory bandwidth (confidence low) until benchmarks are imported.