Hardware Requirements

DeepFellow adapts to your hardware – there are no strict minimum requirements. The optimal configuration depends on your use case, chosen models, and team size. Follow our recommendations below to maximize DeepFellow's performance within your infrastructure.

DeepFellow supports NVIDIA GPUs (CUDA) and Intel discrete GPUs (Vulkan). It also runs in CPU-only mode when no supported GPU is present, though performance will be substantially lower for larger models.

Enterprise Scale – High-Performance Inference

For teams with access to enterprise-grade GPUs (H100, H200, and beyond), DeepFellow scales seamlessly. Deploy multiple GPUs within a single machine, or distribute workloads across multiple machines using DeepFellow Infra for true horizontal scaling, see Installation.

Consumer & Mid-Range Deployments

DeepFellow runs efficiently on consumer-grade hardware, making it accessible for smaller teams and individual developers. Whether you're prototyping with lightweight models or running production workloads on mid-tier GPUs, DeepFellow delivers consistent performance across a wide range of configurations.

Here are some example configurations, suitable for both smaller models and more demanding projects.

Configuration	CPU	RAM	VRAM	GPU	Performance
1	6/12	32GB	16GB	RTX 5060 Ti	Enough for a single user, or a small group (assuming people will not be using the model at the same time). Enough for continuously running tasks, but the performance will be rather weak.
2	8/16	32GB	16GB	RTX 5070 Ti	2x more efficient than the first setup. Still quite limited while used at the same time by a group of users, it will slow down in such a case.
3	8/16	64GB	16GB	RTX 5080	2.5x more efficient than the first setup, suitable for a small to medium-sized group.
4	8/16	64GB	32GB	RTX 5090	4x more efficient than the first setup. More than enough for a single user, appropriate for a group of more or less 10 users. Very good performance on continuously running tasks.

Model Recommendations by VRAM

Once you install a service in DeepFellow Infra, choose a model that fits your available VRAM:

Model	8GB	12GB	16GB	24GB	32GB	64GB	80GB
qwen3:8b	✅	✅	✅	✅	✅	✅	✅
deepseek-r1:8b	✅	✅	✅	✅	✅	✅	✅
gemma3:4b	✅	✅	✅	✅	✅	✅	✅
gemma3:12b	❌	✅	✅	✅	✅	✅	✅
qwen3:14b	❌	✅	✅	✅	✅	✅	✅
deepseek-r1:14b	❌	✅	✅	✅	✅	✅	✅
gpt-oss:20b	❌	❌	❌	✅	✅	✅	✅
gemma3:27b	❌	❌	❌	✅	✅	✅	✅
qwen3:32b	❌	❌	❌	✅	✅	✅	✅
deepseek-r1:32b	❌	❌	❌	❌	✅	✅	✅
deepseek-r1:70b	❌	❌	❌	❌	❌	✅	✅

These recommendations apply to models that use Q4_K_M quantization.

Enterprise Scale – High-Performance Inference

Consumer & Mid-Range Deployments

Model Recommendations by VRAM

On this page