Laptop
RTX 4090 Laptop (16 GB VRAM, 32 GB RAM) — Local LLM Performance
RTX 4090 Laptop (16 GB VRAM, 32 GB RAM) is a Laptop running Windows or Linux, with 16 GB VRAM + 32 GB system RAM.
Specifications
- Memory for models
- 16 GB
- Memory bandwidth
- 576 GB/s
- Memory upgradeable
- Yes
- Architecture
- Ada Lovelace AD103
- Form factor
- Laptop
- Operating system
- Windows or Linux
- Released
- 2023
- Llama 3.1 8B Q4_K_M
- 56–107 tok/s
Worth knowing before you buy
- A mobile RTX 4090 carries 16 GB, not the desktop card's 24 GB, and is closer to a desktop 4080 in throughput — which is why it is modelled against that chip here.
- Sustained throughput depends on the chassis power limit; thin laptops throttle well below the quoted figures.
Models on the RTX 4090 laptop
- Yi 1.5 FamilyYes — comfortably
- VibeThinkerYes — comfortably
- Aya 3B (Tiny Aya)Yes — comfortably
- StarCoder 2Yes
- SmolLM3Yes — comfortably
- SmolLM2Yes — comfortably
- Qwen3-CoderYes, but it is tight
- Qwen3.8Yes
- Qwen 3.6Yes
- Qwen 3.5Yes — comfortably
- Qwen 3Yes
- Qwen 2.5 VLYes — comfortably
- Qwen 2.5 FamilyYes
- Poolside Laguna XS 2.1Yes
- Phi-4 MiniYes — comfortably
- Phi-4 FamilyYes
- Phi 3.5 FamilyYes — comfortably
- OLMo 2Yes, but it is tight
- North Mini CodeYes, but it is tight
- Nex-N2Yes
- Nex-N2.5Yes
- Nemotron Cascade 2Yes
- Nemotron 3 Nano OmniYes, but it is tight
- Mistral Small 3.1Yes
- Mistral Small 3.2Yes
- Mistral FamilyYes, but it is tight
- MinistralYes — comfortably
- Ministral 3Yes
- MiniCPM-VYes — comfortably
- Magistral SmallYes
- Llama 3.2 VisionYes
- Llama 3.2 FamilyYes
- Llama 3.1 FamilyYes — comfortably
- LFM2.5Yes, but it is tight
- InternLM 3Yes, but it is tight
- IBM Granite 4.0Yes
- IBM Granite 4.2Yes — comfortably
- IBM Granite 4.1Yes — comfortably
- Granite 3.0Yes — comfortably
- GPT-OSSYes, but it is tight
- GLM-5 / GLM-5.1No
- GLM-4.7 / GLM-Z1Yes — comfortably
- Gemma 4Yes
- Gemma 3nYes — comfortably
- Gemma 3Yes, but it is tight
- Gemma 2 FamilyYes
- Falcon 3Yes
- EXAONE 3.5Yes — comfortably
- DevstralYes
- DeepSeek R1Yes
- DeepSeek-OCRYes — comfortably
- Cosmos 3Yes, but it is tight
- Cogito v1Yes
- CodestralYes
- Bonsai 27BYes — comfortably
- BitNet b1.58Yes — comfortably
- Aya ExpanseYes — comfortably