§01·tested · /gpus/rtx-5090
Recipes tested on RTX 5090
48 community-tested setups — recipes whose author ran them on this exact card.
nvidia50 series32GB VRAM
- multimodalintermediate32GB+recipe
Qwen3.8-27B on RTX 5090: the full 262K context window on one card
- multimodalintermediate32GB+recipe
Muse Glimmer 30B on RTX 5090: K-Quant-Dynamic, vision and DFlash in llama.cpp
- videoadvanced24GB+recipe
LTX-2.5 on RTX 5090: 22B Audio-Video in 32 GB via ComfyUI's INT8 Path
- llmintermediate32GB+recipe
Nanbeige4.2-3B on RTX 5090: 256K context with a near-lossless q8_0 KV cache
- videoadvanced12GB+recipe
MiniMax H3 on RTX 5090: 32 GB video+audio, and what nvfp4 does not buy you
- multimodaladvanced24GB+recipe
Fara1.5-9B on RTX 5090: BF16 Computer-Use Agent on Blackwell sm_120
- multimodaladvanced24GB+recipe
Fara1.5-4B on RTX 5090: BF16 Computer-Use Agent on Blackwell sm_120 with llama.cpp
- multimodaladvanced32GB+recipe
Fara1.5-27B on RTX 5090: Q6_K Computer-Use Agent on Blackwell sm_120
- llmintermediate32GB+recipe
Phi-4 (14B) on RTX 5090: Local Private Assistant via llama.cpp / Ollama (32GB)
- llmintermediate32GB+recipe
Mistral Small 3.2 24B on RTX 5090: Local Private Assistant via llama.cpp / Ollama (32GB)
- llmintermediate32GB+recipe
Devstral Small 2 (24B) on RTX 5090: Local Agentic Coding via llama.cpp + OpenHands (32GB Quality Tier, near-lossless Q8_0)
- llmadvanced24GB+recipe
Laguna XS 2.1 on RTX 5090: Local Agentic Coding via Ollama / llama.cpp + OpenHands (32GB Tier)
- llmadvanced32GB+recipe
North Mini Code 1.0 on RTX 5090: Local Agentic Coding via llama.cpp + OpenHands (32GB Blackwell Tier)
- llmadvanced32GB+recipe
Ornith 1.0 35B on RTX 5090: Comfortable Local Agentic Coding via llama.cpp + OpenHands (32GB Tier)
- multimodalintermediate24GB+recipe
Qwen3.5-35B-A3B on RTX 5090: Blackwell MXFP4 MoE Chat at 165 tok/s
- multimodalintermediate20GB+recipe
Qwen3.5 27B on RTX 5090: Q4_K GGUF local chat via llama.cpp
- llmintermediate20GB+recipe
Gemma 4 31B on RTX 5090: dense 31B local chat at 61 tok/s, with Q5/Q6 quality headroom in 32 GB
- imagebeginner24GB+recipe
Flux.1 Dev on RTX 5090: ComfyUI Image Generation Guide
- llmintermediate19GB+recipe
Qwen3-30B-A3B on RTX 5090: 226 tok/s MoE Chat with Room to Spare
- llmadvanced32GB+recipe
Llama 3.1 70B on RTX 5090: Fitting a 70B Model on One 32 GB Card with IQ3 GGUF
- imageintermediate16GB+recipe
Krea 2 on RTX 5090 via ComfyUI: 8-Step Turbo + Resident Full-Quality Raw in 32GB
- videoadvanced31GB+recipe
Sulphur 2 on RTX 5090: Uncensored LTX-2.3 Video at fp8mixed, Native
- llmbeginner10GB+recipe
Llama 3.1 8B on RTX 5090: Local Chat via llama.cpp + Unsloth UD-Q4_K_XL GGUF
- 3dadvanced24GB+recipe
TRELLIS.2-4B on RTX 5090: First Consumer Card That Hits the 24 GB Floor for Image-to-3D
- 3dintermediate29GB+recipe
Hunyuan3D-2.1 on RTX 5090: Image-to-Mesh + PBR Texture in One Pass
- 3dadvanced16GB+recipe
TRELLIS image-large on RTX 5090: Image-to-3D Mesh Generation with the Blackwell Build Path
- 3dintermediate12GB+recipe
Waypoint 1.5 on RTX 5090: Real-Time Interactive World Model at 720p, 72 FPS
- imageintermediate24GB+recipe
ERNIE-Image-Turbo on RTX 5090: 8-Step Text-to-Image at BF16 in ComfyUI
- videointermediate28GB+recipe
LTX-2.3 on RTX 5090: First Card That Hits the 32 GB Floor
- ttsbeginner3GB+recipe
Kokoro TTS on RTX 5090: 82M-Parameter Text-to-Speech, 54 Voices, 30 GB Free for a Multi-Model Server
- multimodalbeginner20GB+recipe
Gemma 4 E4B on RTX 5090: Multimodal Inference via BF16 (with 24 GB of Headroom to Spend)
- imageintermediate21GB+recipe
Qwen-Image on RTX 5090: 20B Text-to-Image via ComfyUI FP8 (Blackwell Native Path)
- imagebeginner9GB+recipe
Flux.2 Klein 4B on RTX 5090: FP8 1.2-Second Generation, Blackwell-Native Speed Win
- imageintermediate18GB+recipe
LongCat-Image (base T2I) on RTX 5090: Bilingual 6B Text-to-Image via diffusers BF16 with 14 GB Headroom
- imageintermediate24GB+recipe
Chroma1-Base (V48) on RTX 5090: Uncensored 8.9B FLUX.1-Schnell De-Distillation via Diffusers BF16
- imageintermediate17GB+recipe
HiDream-O1-Image on RTX 5090: 2048×2048 Text-to-Image with MXFP8 Blackwell-Native Acceleration in ComfyUI
- llmintermediate29GB+recipe
Qwen3-32B on RTX 5090: Q6_K_XL GGUF via llama.cpp (with AWQ-INT4 + 128K context alternative)
- llmintermediate29GB+recipe
Gemma 4 26B A4B-it on RTX 5090: Q8_0 Quality Tier via ggml-org GGUF + llama.cpp
- llmbeginner10GB+recipe
DeepSeek-R1-Distill-Qwen-14B on RTX 5090: 128K Reasoning Context Unlocked
- llmintermediate17GB+recipe
Qwen3-14B on RTX 5090: FP8 via vLLM with Native Blackwell Acceleration
- llmbeginner6GB+recipe
Qwen3-8B on RTX 5090: Q4_K_M GGUF with 26 GB of Headroom for Colocation, BF16, or Full 131K Context
- llmbeginner16GB+recipe
gpt-oss 20B on RTX 5090: MXFP4 Chat at 298 tok/s via Ollama or vLLM
- videointermediate14GB+recipe
HunyuanVideo-1.5 on RTX 5090: 480p Step-Distilled I2V with an 8 GB Headroom Unlock
- videointermediate22GB+recipe
LightX2V on RTX 5090: 4-Step Text-to-Video with Distilled Wan2.1-14B (Blackwell-Native FP8 + Future NVFP4 Path)
- videointermediate24GB+recipe
Wan 2.2 T2V-A14B on RTX 5090: 720p text-to-video in ComfyUI with FP8 scaled weights (Blackwell)
- videointermediate20GB+recipe
CogVideoX 1.5 5B on RTX 5090: 1360x768 Text-to-Video, No-Sequential-Offload Path
- videointermediate22GB+recipe
Mochi 1 on RTX 5090: 85-frame 480p Text-to-Video with Diffusers
- videointermediate8GB+recipe
Wan 2.2 TI2V-5B on RTX 5090: 720p Text/Image-to-Video in ComfyUI