Model Anatomy See what's inside open-weight AI models

Can I run Swift-Qwen3.8-27b on an RTX 4090?

Yes, with a GGUF version

The published weights are too big, but the Q5_K_M file from bartowski/ukisai_Swift-Qwen3.8-27b-GGUF (20.9 GB) fits, with room for ≈ 8K tokens of context.

Full analysis of Swift-Qwen3.8-27b Everything that runs on RTX 4090 Get the Q5_K_M file on Hugging Face
ollama run hf.co/bartowski/ukisai_Swift-Qwen3.8-27b-GGUF:Q5_K_M

Files and context on RTX 4090

Each file of Swift-Qwen3.8-27b (and of GGUF versions the site has analyzed), and how much context fits next to it, with the context cache in 16-bit or 8-bit.

FileSizeOn RTX 4090 Context (16-bit)Context (8-bit)
As published55.6 GBDoesn't fit——
F16ukisai/Swift-Qwen3.8-27B-GGUF54.7 GBDoesn't fit——
Q8_0ukisai/Swift-Qwen3.8-27B-GGUF29.1 GBDoesn't fit——
Q6_K_Lukisai/Swift-Qwen3.8-27B-GGUF25.2 GBDoesn't fit——
Q6_K_Sukisai/Swift-Qwen3.8-27B-GGUF23.1 GBDoesn't fit——
Q6_Kukisai/Swift-Qwen3.8-27B-GGUF22.9 GBDoesn't fit——
Q5_K_Mukisai/Swift-Qwen3.8-27B-GGUF20.2 GBTight19K37K
Q5_K_Sukisai/Swift-Qwen3.8-27B-GGUF19.8 GBTight25K50K
Q4_K_Lukisai/Swift-Qwen3.8-27B-GGUF19.0 GBTight36K72K
Q4_K_Mukisai/Swift-Qwen3.8-27B-GGUF18.0 GBTight51K102K
IQ4_NLukisai/Swift-Qwen3.8-27B-GGUF17.7 GBTight56K113K
Q4_1ukisai/Swift-Qwen3.8-27B-GGUF17.5 GBTight58K117K
Q4_K_Sukisai/Swift-Qwen3.8-27B-GGUF16.6 GBFits73K145K
Q4_0ukisai/Swift-Qwen3.8-27B-GGUF16.0 GBFits82K163K
IQ4_XSukisai/Swift-Qwen3.8-27B-GGUF15.7 GBFits86K172K
IQ3_Mukisai/Swift-Qwen3.8-27B-GGUF15.1 GBFits95K190K
Q3_K_Lukisai/Swift-Qwen3.8-27B-GGUF14.3 GBFits106K212K
Q3_K_Mukisai/Swift-Qwen3.8-27B-GGUF13.6 GBFits117K233K
IQ3_XSukisai/Swift-Qwen3.8-27B-GGUF13.0 GBFits126K251K
Q3_K_Sukisai/Swift-Qwen3.8-27B-GGUF13.0 GBFits127K253K
IQ3_XXSukisai/Swift-Qwen3.8-27B-GGUF12.5 GBFits133K256K
Q2_Kukisai/Swift-Qwen3.8-27B-GGUF11.0 GBFits155K256K
IQ2_Mukisai/Swift-Qwen3.8-27B-GGUF10.7 GBFits160K256K
IQ2_Sukisai/Swift-Qwen3.8-27B-GGUF9.90 GBFits172K256K
IQ2_XSukisai/Swift-Qwen3.8-27B-GGUF9.30 GBFits181K256K
IQ2_XXSukisai/Swift-Qwen3.8-27B-GGUF9.09 GBFits184K256K
BF16bartowski/ukisai_Swift-Qwen3.8-27b-GGUF54.7 GBDoesn't fit——
Q8_0bartowski/ukisai_Swift-Qwen3.8-27b-GGUF29.1 GBDoesn't fit——
Q6_K_Lbartowski/ukisai_Swift-Qwen3.8-27b-GGUF25.0 GBDoesn't fit——
Q6_Kbartowski/ukisai_Swift-Qwen3.8-27b-GGUF23.9 GBDoesn't fit——
Q6_K_Sbartowski/ukisai_Swift-Qwen3.8-27b-GGUF22.9 GBDoesn't fit——
Q5_K_Mbartowski/ukisai_Swift-Qwen3.8-27b-GGUF20.9 GBTight8K16K
Q5_K_Sbartowski/ukisai_Swift-Qwen3.8-27b-GGUF19.6 GBTight28K56K
Q4_K_Lbartowski/ukisai_Swift-Qwen3.8-27b-GGUF18.8 GBTight39K78K
Q4_1bartowski/ukisai_Swift-Qwen3.8-27b-GGUF17.8 GBTight54K108K
IQ4_NLbartowski/ukisai_Swift-Qwen3.8-27b-GGUF17.4 GBTight60K119K
Q4_K_Mbartowski/ukisai_Swift-Qwen3.8-27b-GGUF17.4 GBTight60K119K
Q4_K_Sbartowski/ukisai_Swift-Qwen3.8-27b-GGUF16.4 GBFits76K151K
Q4_0bartowski/ukisai_Swift-Qwen3.8-27b-GGUF16.3 GBFits76K152K
IQ4_XSbartowski/ukisai_Swift-Qwen3.8-27b-GGUF15.5 GBFits89K178K
IQ3_Mbartowski/ukisai_Swift-Qwen3.8-27b-GGUF14.9 GBFits98K196K
Q3_K_Lbartowski/ukisai_Swift-Qwen3.8-27b-GGUF14.1 GBFits109K218K
Q3_K_Mbartowski/ukisai_Swift-Qwen3.8-27b-GGUF13.4 GBFits120K240K
IQ3_XSbartowski/ukisai_Swift-Qwen3.8-27b-GGUF12.8 GBFits129K256K
Q3_K_Sbartowski/ukisai_Swift-Qwen3.8-27b-GGUF12.7 GBFits130K256K
IQ3_XXSbartowski/ukisai_Swift-Qwen3.8-27b-GGUF12.3 GBFits136K256K
Q2_Kbartowski/ukisai_Swift-Qwen3.8-27b-GGUF10.8 GBFits158K256K
IQ2_Mbartowski/ukisai_Swift-Qwen3.8-27b-GGUF10.5 GBFits163K256K
IQ2_Sbartowski/ukisai_Swift-Qwen3.8-27b-GGUF9.68 GBFits175K256K
IQ2_XSbartowski/ukisai_Swift-Qwen3.8-27b-GGUF9.09 GBFits184K256K
IQ2_XXSbartowski/ukisai_Swift-Qwen3.8-27b-GGUF8.88 GBFits187K256K

Swift-Qwen3.8-27b on other hardware

Popular models that run on RTX 4090

Assumes 90% of the card's memory is usable and at least 4K tokens of context for one conversation, with the context cache in 16-bit unless noted. Fits leaves at least 20% of that memory for context; Tight leaves less.