Best GPU for Qwen2.5 72B: VRAM Requirements
Qwen2.5 72B needs about 162.9 GB of GPU memory in BF16 and 42.9 GB at 4 bits. See which GPUs fit it for inference and fine-tuning.
Qwen2.5 72B needs about 162.9 GB of VRAM in FP16, 82.9 GB in FP8 and 42.9 GB at 4 bits for inference with an 8,192-token context. In FP16 it fits on one NVIDIA B300 288GB; a 4-bit build fits on one NVIDIA L40S 48GB.
Qwen2.5 72B is a 72.7B-parameter language model from Alibaba Cloud (Qwen team), released in September 2024, with a 32,768-token native context window that extends to 131,072 tokens with YaRN and published under the Qwen License. Its weights alone take 145.4 GB in BF16 and 36.4 GB at 4 bits per parameter; the tables below add the KV cache and runtime memory, then list the GPUs that hold each setup.
Recommended GPUs for Qwen2.5 72B
Best for inference
1 GPU × NVIDIA H200 141GB
FP8 or INT8 weights, 82.9 GB at 8,192 tokens.
Best for fine-tuning
1 GPU × NVIDIA B300 288GB
LoRA on BF16 weights, 189 GB.
Budget pick
1 GPU × NVIDIA RTX 6000 Ada 48GB
Smallest card class that holds a 4-bit build, 42.9 GB.
Qwen2.5 72B VRAM requirements by precision
One request with an 8,192-token context, KV cache in BF16, plus a 10 % margin for activations and runtime buffers.
| Precision | Weights | KV cache | Total for inference |
|---|---|---|---|
| BF16 / FP16 | 145.4 GB | 2.7 GB | 162.9 GB |
| FP8 | 72.7 GB | 2.7 GB | 82.9 GB |
| INT4 | 36.4 GB | 2.7 GB | 42.9 GB |
Qwen2.5 72B VRAM by task
Fine-tuning rows assume a short sequence, batch 1 and gradient checkpointing; longer sequences add activation memory.
| Task | GPU memory |
|---|---|
| Inference at the full 131,072-token context (FP8) | 127.2 GB |
| QLoRA (4-bit base) | 47.3 GB |
| LoRA (BF16 base) | 189 GB |
| Full fine-tuning (BF16, AdamW) | 1,279.5 GB |
GPUs that fit Qwen2.5 72B
Number of cards whose combined memory holds each setup, assuming the model is split across them with tensor or pipeline parallelism. A dash means more than eight cards.
| GPU | Memory | BF16 inference (162.9 GB) | FP8 inference (82.9 GB) | INT4 inference (42.9 GB) | QLoRA (47.3 GB) | LoRA (189 GB) |
|---|---|---|---|---|---|---|
| NVIDIA GeForce RTX 5080 16GB | 16 GB | — | 8 GPUs | 4 GPUs | 4 GPUs | — |
| NVIDIA GeForce RTX 4090 24GB | 24 GB | 8 GPUs | 4 GPUs | 2 GPUs | 2 GPUs | 8 GPUs |
| NVIDIA L4 24GB | 24 GB | 8 GPUs | 4 GPUs | 2 GPUs | 2 GPUs | 8 GPUs |
| NVIDIA GeForce RTX 5090 32GB | 32 GB | 8 GPUs | 4 GPUs | 2 GPUs | 2 GPUs | 8 GPUs |
| NVIDIA Tesla V100 32GB | 32 GB | 8 GPUs | 4 GPUs | 2 GPUs | 2 GPUs | 8 GPUs |
| NVIDIA A100 PCIe 40GB | 40 GB | 8 GPUs | 4 GPUs | 2 GPUs | 2 GPUs | 8 GPUs |
| NVIDIA L40S 48GB | 48 GB | 4 GPUs | 2 GPUs | 1 GPU | 1 GPU | 4 GPUs |
| NVIDIA Quadro RTX 8000 48GB | 48 GB | 4 GPUs | 2 GPUs | 1 GPU | 1 GPU | 4 GPUs |
| NVIDIA RTX 6000 Ada 48GB | 48 GB | 4 GPUs | 2 GPUs | 1 GPU | 1 GPU | 4 GPUs |
| NVIDIA RTX A6000 48GB | 48 GB | 4 GPUs | 2 GPUs | 1 GPU | 1 GPU | 4 GPUs |
| NVIDIA A100 PCIe 80GB | 80 GB | 4 GPUs | 2 GPUs | 1 GPU | 1 GPU | 4 GPUs |
| NVIDIA A100 SXM 80GB | 80 GB | 4 GPUs | 2 GPUs | 1 GPU | 1 GPU | 4 GPUs |
| NVIDIA H100 PCIe 80GB | 80 GB | 4 GPUs | 2 GPUs | 1 GPU | 1 GPU | 4 GPUs |
| NVIDIA H100 SXM 80GB | 80 GB | 4 GPUs | 2 GPUs | 1 GPU | 1 GPU | 4 GPUs |
| NVIDIA RTX PRO 6000 Blackwell 96GB | 96 GB | 2 GPUs | 1 GPU | 1 GPU | 1 GPU | 2 GPUs |
| NVIDIA H200 141GB | 141 GB | 2 GPUs | 1 GPU | 1 GPU | 1 GPU | 2 GPUs |
| NVIDIA B200 180GB | 180 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 2 GPUs |
| NVIDIA B300 288GB | 288 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
| NVIDIA GB300 NVL | 288 GB | 1 GPU | 1 GPU | 1 GPU | 1 GPU | 1 GPU |
What is Qwen2.5 72B?
Qwen2.5 72B is the flagship dense model of the Qwen2.5 series, aimed at multilingual chat, long documents, coding and math. It is released under the Qwen License rather than Apache 2.0, which sets conditions for very large commercial deployments.
- Developer
- Alibaba Cloud (Qwen team)
- Released
- September 2024
- Parameters
- 72.7B
- Context window
- 32,768 tokens native, 131,072 with YaRN
- License
- Qwen License
The model card supports 131,072 tokens with YaRN scaling; the native window is 32,768 tokens.
Source: Qwen2.5 72B Instruct model card (as of November 17, 2026).
Frequently asked questions
- How much VRAM does Qwen2.5 72B need?
- For inference with an 8,192-token context and one request, Qwen2.5 72B needs about 162.9 GB in BF16, 82.9 GB in FP8, 42.9 GB in INT4, including a 10 % runtime margin. Real usage depends on the inference engine and its settings.
- Can Qwen2.5 72B run on a single GPU?
- Yes. In BF16 it needs about 162.9 GB, which fits one NVIDIA B300 288GB (288 GB).
- How much GPU memory does fine-tuning Qwen2.5 72B take?
- QLoRA (4-bit base): about 47.3 GB; LoRA (BF16 base): about 189 GB; Full fine-tuning (BF16, AdamW): about 1,279.5 GB. Longer training sequences and larger batches add activation memory on top.
- What is the best GPU for Qwen2.5 72B?
- Best for inference: 1 GPU × NVIDIA H200 141GB (FP8 or INT8 weights, 82.9 GB at 8,192 tokens). Best for fine-tuning: 1 GPU × NVIDIA B300 288GB (LoRA on BF16 weights, 189 GB). Budget pick: 1 GPU × NVIDIA RTX 6000 Ada 48GB (Smallest card class that holds a 4-bit build, 42.9 GB). Check live offers before booking, as availability changes.
- Can Qwen2.5 72B run on an RTX 4090?
- Not on one card: even a 4-bit build needs about 42.9 GB, so it takes 2 RTX 4090 cards (48 GB combined) with the model split across them.
- Can Qwen2.5 72B run on an RTX 5090?
- Not on one card: even a 4-bit build needs about 42.9 GB, so it takes 2 RTX 5090 cards (64 GB combined) with the model split across them.
Estimates use the formulas of the LLM VRAM calculator (decimal GB). Real usage depends on the inference engine, quantization format and memory fragmentation: test on the target GPU before a long rental.
Explore the live market
Every offer on the live market comes from a reviewed supplier and is sold by EverXCloud Inc.