Qwen 2.5 VL 72B needs ~37.1 GB VRAM. AMD Instinct MI100 32GB has 32.0 GB. With Q2_K quantization, expect ~15 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
20.9 GB over capacity — needs offload or smaller quantization
Fit status
Too heavy
Decode
5.1 tok/s
TTFT
37631 ms
Safe context
4K
Memory
52.9 GB / 32.0 GB
Offload
40%
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 3.8 GB of extra host RAM just for the offloaded portion, before OS and other tools.
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Chat | F | Too heavy | 5.7 tok/s | 18583 ms | 4K |
| Coding | F | Too heavy | 5.1 tok/s | 37631 ms | 4K |
| Agentic Coding | F | Too heavy | 4.3 tok/s | 65915 ms | 4K |
| Reasoning | F | Too heavy | 5.1 tok/s | 44473 ms | 4K |
| RAG | F | Too heavy | 4.3 tok/s | 82393 ms | 4K |
Inference speed
Estimated decode speed (tokens/sec) for Qwen 2.5 VL 72B at Q4_K_M across popular GPUs and Apple Silicon, including multi-GPU rigs, using the fastest local runtime per device. Fastest is 2× RX 7900 XTX 24GB at ~17 tok/s. Speed is memory-bandwidth bound, so cards that fit the whole model in VRAM run far faster than ones that offload to system RAM.
| GPU / Mac | Memory | Quant | Speed (tok/s) | Fits? |
|---|---|---|---|---|
2× RX 7900 XTX 24GB | 48 GB | Q4_K_M | 16.7 | Heavy offload |
MacBook Pro M4 Max 128GB | 128 GB | Q4_K_M | 14.9 | Fits |
Mac Studio M3 Ultra 256GB | 256 GB | Q4_K_M | 13.8 | Fits |
Mac Studio M2 Ultra 128GB | 128 GB | Q4_K_M | 11.5 | Fits |
MacBook Pro M4 Max 64GB | 64 GB | Q4_K_M | 11.0 | Too big |
Mac Studio M1 Ultra 128GB | 128 GB | Q4_K_M | 10.9 | Fits |
 2× RTX 4090 24GB | 48 GB | Q4_K_M | 8.8 | Heavy offload |
| 2× RTX 3090 24GB | 48 GB | Q4_K_M | 8.1 | Heavy offload |
| 4× RTX 3060 12GB | 48 GB | Q4_K_M | 7.1 | Heavy offload |
| RTX 5090 32GB | 32 GB | Q4_K_M | 5.3 | Too big |
MacBook Pro M4 Pro 48GB | 48 GB | Q4_K_M | 5.1 | Too big |
MacBook Pro M3 Max 64GB | 64 GB | Q4_K_M | 4.4 | Too big |
MacBook Pro M1 Max 64GB | 64 GB | Q4_K_M | 4.0 | Too big |
RX 7900 XTX 24GB | 24 GB | Q4_K_M | 2.6 | Too big |
| RTX 4090 24GB | 24 GB | Q4_K_M | 2.0 | Too big |
| RTX 4080 Super 16GB | 16 GB | Q4_K_M | 2.0 | Too big |
| RTX 3090 24GB | 24 GB | Q4_K_M | 2.0 | Too big |
| RTX 4070 12GB | 12 GB | Q4_K_M | 2.0 | Too big |
| RTX 3060 12GB | 12 GB | Q4_K_M | 2.0 | Too big |
| RTX 4060 8GB | 8 GB | Q4_K_M | 2.0 | Too big |
Estimates for single-stream decoding at Q4_K_M; real tokens/sec varies with prompt length, context, batch size, and runtime build. Prompt processing (prefill) is faster than the decode figures shown here.
How Qwen 2.5 VL 72B (72B params) fits at each quantization level on AMD Instinct MI100 32GB (32.0 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 28.1 GB | Low | F0 |
Q3_K_S | 3 | 35.3 GB | Low | F0 |
NVFP4 | 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 |
Copy-paste commands to run Qwen 2.5 VL 72B on your machine.
Run
lms load Qwen2.5-VL-72B-Instruct && lms server startUpgrade-Optionen
Makes the model fit on the accelerator instead of staying completely out of reach.
Raises estimated decode speed by about 41%.
ca. $3,999 MSRP
Makes the model fit on the accelerator instead of staying completely out of reach.
Removes host-memory offload, which is usually the single biggest latency and throughput win.
ca. $8,000 MSRP
Makes the model fit on the accelerator instead of staying completely out of reach.
Removes host-memory offload, which is usually the single biggest latency and throughput win.
ca. $10,000 MSRP
NVIDIA H100 80GBGrößter SprungMakes the model fit on the accelerator instead of staying completely out of reach.
Removes host-memory offload, which is usually the single biggest latency and throughput win.
ca. $40,000 MSRP
Yes, AMD Instinct MI100 32GB can run Qwen 2.5 VL 72B at Q2_K quantization (Very compromised (needs ~3.8 GB host RAM)). The recommended Q4_K_M requires 52.9 GB which exceeds available memory, but at Q2_K it needs only 37.1 GB. Expected decode speed: 14.5 tok/s.
Qwen 2.5 VL 72B (72B parameters) requires approximately 52.9 GB at Q4_K_M quantization. On AMD Instinct MI100 32GB, it fits at Q2_K using 37.1 GB.
The recommended quantization is Q4_K_M, but on AMD Instinct MI100 32GB the best fitting quantization is Q2_K, which uses 37.1 GB.
On AMD Instinct MI100 32GB, Qwen 2.5 VL 72B achieves approximately 14.5 tokens per second decode speed with a time-to-first-token of 13383ms using Q2_K quantization.
For coding workloads, Qwen 2.5 VL 72B on AMD Instinct MI100 32GB receives a F grade with 5.1 tok/s and 4K context.
On AMD Instinct MI100 32GB, Qwen 2.5 VL 72B can safely use up to 4K tokens of context at Q2_K quantization. The model's official context limit is 33K, but available memory constrains the safe maximum.
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.
Paste this snippet into any page to show a live fit card.
<iframe src="https://willitrunai.com/embed/qwen-2.5-vl-72b-on-instinct-mi100-32gb" width="400" height="180" frameborder="0" style="border:none;border-radius:12px;overflow:hidden;" title="Will It Run AI — fit result"></iframe>
Preview: