Qwen 2.5 72B needs ~54.5 GB VRAM. Radeon PRO W7900 DS 48GB has 48.0 GB. With Q4_K_M quantization, expect ~7 tok/s.
Operating mode
Interactive favors responsiveness, while light API and scale-out lean harder on serving readiness. The fit stays the same, but the recommendation lens changes.
Current mode
Balanced
Balanced for general local use. Keeps the ranking neutral across personal and serving workflows.
Select quantization to explore
6.5 GB over capacity — needs offload or smaller quantization
Fit status
Very compromised (needs ~5.2 GB host RAM)
Decode
7.2 tok/s
TTFT
26721 ms
Safe context
4K
Memory
54.5 GB / 48.0 GB
Offload
10%
It fits through host-memory offload, and offload is the main reason performance drops.
CPU or host-memory offload is active
About 10% of the working set spills out of accelerator memory, which usually hurts latency and sustained decode throughput.
Very little memory headroom
You can run the model, but there is not much room left for longer context, bigger batches, extra apps, or future model updates.
Remove offload with more accelerator memory
Prioritize a GPU or unified-memory tier that fits the whole model natively. Removing offload usually helps more than small compute gains.
Buy headroom, not only minimum fit
A slightly larger memory tier gives you safer context growth and makes the recommendation more future-proof.
Increase host RAM if you keep offloading
This setup may need roughly {ram} GB of extra host RAM just for the offloaded portion, before OS and other tools.
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Chat | B | Very compromised | 7.3 tok/s | 14393 ms | 4K |
| Coding | B | Very compromised | 6.7 tok/s | 29060 ms | 4K |
| Agentic Coding | F | Too heavy | 5.6 tok/s | 50635 ms | 4K |
| Reasoning | B | Very compromised | 6.7 tok/s | 34343 ms | 4K |
| RAG | F | Too heavy | 5.6 tok/s | 63294 ms | 4K |
How Qwen 2.5 72B (72B params) fits at each quantization level on Radeon PRO W7900 DS 48GB (48.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 28.1 GB | Low | A79 |
Q3_K_SBest for your GPU | 3 | 35.3 GB | Low | A79 |
Copy-paste commands to run Qwen 2.5 72B on your machine.
Run
ollama run qwen2.5:72bUpgrade options
Removes host-memory offload, which is usually the single biggest latency and throughput win.
Raises estimated decode speed by about 1908%.
~$8,000 MSRP
Removes host-memory offload, which is usually the single biggest latency and throughput win.
Raises estimated decode speed by about 283%.
~$10,000 MSRP
Removes host-memory offload, which is usually the single biggest latency and throughput win.
Raises estimated decode speed by about 1176%.
~$12,000 MSRP
Yes, Radeon PRO W7900 DS 48GB can run Qwen 2.5 72B with a B grade (Very compromised). Expected decode speed: 6.7 tok/s.
Qwen 2.5 72B (72B parameters) requires approximately 54.5 GB of memory with Q4_K_M quantization.
The recommended quantization for Qwen 2.5 72B is Q4_K_M, which balances quality and memory efficiency.
On Radeon PRO W7900 DS 48GB, Qwen 2.5 72B achieves approximately 6.7 tokens per second decode speed with a time-to-first-token of 29060ms using Q4_K_M quantization.
For coding workloads, Qwen 2.5 72B on Radeon PRO W7900 DS 48GB receives a B grade with 6.7 tok/s and 4K context.
On Radeon PRO W7900 DS 48GB, Qwen 2.5 72B can safely use up to 4K tokens of context. The model's official context limit is 131K, but available memory constrains the safe maximum.
Paste this snippet into any page to show a live fit card.
<iframe src="https://willitrunai.com/embed/qwen-2.5-72b-on-radeon-pro-w7900-ds-48gb" width="400" height="180" frameborder="0" style="border:none;border-radius:12px;overflow:hidden;" title="Will It Run AI — fit result"></iframe>
Preview:
| 4 |
40.3 GB |
| Medium |
| F0 |
Q4_K_M | 4 | 43.9 GB | Medium | F0 |
Q5_K_M | 5 | 51.8 GB | High | F0 |
Q6_K | 6 | 59.0 GB | High | F0 |
Q8_0 | 8 | 77.0 GB | Very High | F0 |
F16 | 16 | 147.6 GB | Maximum | F0 |
Remove offload with more accelerator memory. Prioritize a GPU or unified-memory tier that fits the whole model natively. Removing offload usually helps more than small compute gains.