GPUs with 48 GB VRAM and more
Cards with at least 48 GB of video memory start at €0.46 an hour on VRAMfarm (L40). 48 GB holds a language model of about 72B parameters in 4-bit or 21B in FP16. For images and video: Wan 2.2 14B video and every popular image model at full quality.
48 GB is the step to 70B language models in 4-bit and the large Wan 2.2 video model. The RTX A6000, L40 and L40S are workstation and data-centre cards with this size.
Cards with 48 GB or more, cheapest first
| GPU | Type | Memory | TFLOPS FP32 | Where | Price per hour | ≈ per month | Action |
|---|---|---|---|---|---|---|---|
| L40 | data centre | 48 GB | ≈90.5 | €0.46/h | €331 per month | Rent | |
| RTX A6000 | workstation | 48 GB | ≈38.7 | €0.72/h | €518 per month | Rent | |
| L40S | data centre | 48 GB | ≈91.6 | +1 | €1.06/h | €763 per month | Rent |
| A100 | data centre | 80 GB | ≈19.5 | +2 | €1.46/h | €1,051 per month | Rent |
| RTX PRO 6000 | workstation | 96 GB | ≈125 | +8 | €2.12/h | €1,526 per month | Rent |
| H100 | data centre | 80 GB | ≈67 | +1 | €3.90/h | €2,808 per month | Rent |
| H200 | data centre | 141 GB | ≈67 | +2 | €5.43/h | €3,910 per month | Rent |
| B200 | data centre | 192 GB | ≈80 | €16.12/h | €11,606 per month | Rent |
To launch, your balance must cover one hour of the machine. You are billed by the minute for actual use.
What runs on 48 GB
Language models
Up to about 72B parameters in 4-bit (AWQ, GPTQ, GGUF Q4) or 21B in FP16, with room for a normal context. Long context and many parallel users need more. The VRAM calculator works it out for your model.
Images and video
Wan 2.2 14B video and every popular image model at full quality.
Training
QLoRA of 32B models, 70B with Unsloth and short sequences.
Typical jobs at this size
- Llama 3.3 70B in 4-bit
- Long video at full resolution
- QLoRA on 30B models
Frequently asked questions
What is the cheapest GPU with 48 GB or more?
The NVIDIA L40, from €0.46 an hour on VRAMfarm, billed by the minute.
Which LLM can I run on 48 GB?
About 72B parameters in 4-bit or 21B in FP16, with room for a normal context. For example: Qwen3 14B, Qwen3 32B, Llama 3.3 70B.
Can I combine the memory of several GPUs?
For language models, yes: vLLM and llama.cpp split a model across cards, so 2 × 24 GB can run what needs about 40 GB. For image and video models memory does not add up, one card has to hold the model.
How do I know 48 GB is enough?
Add up the weights (2 GB per billion parameters in FP16, 0.6 GB in 4-bit), the context cache and 10–15% for buffers, or use our VRAM calculator. If the model does not fit, it either fails with “CUDA out of memory” or runs much slower with offloading.
Rent 48 GB of VRAM
Top up from €10, pick a card from the list and start in a few minutes.