Can I run Huihui-Qwen3.8-27B-abliterated-GGUF on an RTX 5090?

Yes, just

Its Q8_K_L file (27.3 GB) fits in the 28.8 GB an RTX 5090 can use, with room for ≈ 21K tokens of context.

Full analysis of Huihui-Qwen3.8-27B-abliterated-GGUF Everything that runs on RTX 5090 Get the Q8_K_L file on Hugging Face

ollama run hf.co/huihui-ai/Huihui-Qwen3.8-27B-abliterated-GGUF:Q8_K_L

Files and context on RTX 5090

Each file of Huihui-Qwen3.8-27B-abliterated-GGUF (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.

File	Size	On RTX 5090	Context (16-bit)	Context (8-bit)
BF16	54.7 GB	Doesn't fit	—	—
Huihui-Qwen3.8-27B-abliterated-bf16	54.7 GB	Doesn't fit	—	—
Huihui-Qwen3.8-27B-abliterated-Q8_0_L	38.8 GB	Doesn't fit	—	—
UD-Q8_K_XL	31.5 GB	Doesn't fit	—	—
Q8_0	29.0 GB	Doesn't fit	—	—
Q8_K_L	27.3 GB	Tight	21K	42K
Q6_K_L	24.9 GB	Tight	55K	110K
UD-Q6_K_XL	24.8 GB	Tight	57K	114K
Huihui-Qwen3.8-27B-abliterated-UD-DW-Q6_K_L	23.9 GB	Tight	70K	140K
Q5_K_L	22.9 GB	Fits	86K	172K
Q6_K	22.4 GB	Fits	93K	185K
Q4_K_L	20.9 GB	Fits	115K	230K
UD-Q5_K_XL	20.7 GB	Fits	119K	238K
Q5_K	19.5 GB	Fits	136K	256K
Q3_K_L	18.7 GB	Fits	148K	256K
UD-Q4_K_XL	17.4 GB	Fits	168K	256K
Q2_K_L	17.2 GB	Fits	170K	256K
Q4_K	16.8 GB	Fits	176K	256K
Q4_K_M	16.6 GB	Fits	180K	256K
Q4_K_S	15.6 GB	Fits	194K	256K
UD-IQ4_XS	14.4 GB	Fits	212K	256K
Q3_K	13.5 GB	Fits	226K	256K
UD-Q3_K_XL	13.3 GB	Fits	228K	256K
Huihui-Qwen3.8-27B-abliterated-UD-DW-IQ3_S	12.3 GB	Fits	243K	256K
IQ3_S	12.1 GB	Fits	246K	256K
UD-IQ3_S	12.0 GB	Fits	249K	256K
Huihui-Qwen3.8-27B-abliterated-GSQ-RCO-IQ3_S	11.8 GB	Fits	251K	256K
Huihui-Qwen3.8-27B-abliterated-UD-DW-IQ3_XXS	11.4 GB	Fits	256K	256K
UD-IQ3_XXS	11.0 GB	Fits	256K	256K
Q2_K	10.9 GB	Fits	256K	256K
IQ3_XXS	10.4 GB	Fits	256K	256K
Huihui-Qwen3.8-27B-abliterated-GSQ-RCO-IQ3_XXS	10.1 GB	Fits	256K	256K
UD-Q2_K_XL	10.0 GB	Fits	256K	256K
UD-IQ2_S	8.78 GB	Fits	256K	256K
Huihui-Qwen3.8-27B-abliterated-UD-IQ2_S	8.42 GB	Fits	256K	256K

Huihui-Qwen3.8-27B-abliterated-GGUF on other hardware

Popular models that run on RTX 5090

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.