NVIDIA · Ampere

What runs on a RTX 3090

24 GB at 936 GB/s. The largest model that fits at Q4_K_M is Qwen3 32B, at roughly 29.5 tokens per second.

Model

Hardware

Quantisation

Context 8K

KV cache precision

Runtime

Batch

GPUs

Monitor plugged into this GPU

Runs comfortably

6.4 GB of 21.8 GB · 30%

022 GB

Weights 4.6 GB

KV cache 1.0 GB

Runtime overhead 0.8 GB

Llama 3.1 8B at Q4_K_M leaves 15.3 GB spare on a RTX 3090. There is room to raise the context length or move up a quantisation level.

Generation112tok/s

Prompt processing1857tok/s

Max context128Ktokens

KV per 1K tokens0GB

Every quantisation of Llama 3.1 8B on a RTX 3090

Highlighted row is the highest quality that still fits at 8K context.

Quantisation	bpw	Weights	Total	Verdict	Max ctx	tok/s
FP16 / BF16	16.00	15.0 GB	16.8 GB	Runs comfortably	48K	37.6
INT8 / W8A8	8.50	7.7 GB	9.5 GB	Runs comfortably	106K	70.4
Q8_0 (GGUF)	8.50	7.7 GB	9.5 GB	Runs comfortably	106K	70.4
FP8 (E4M3)	8.00	7.5 GB	9.3 GB	Runs comfortably	108K	72.4
Q6_K	6.56	6.1 GB	8.0 GB	Runs comfortably	118K	86.9
AWQ 4-bit	4.25	5.4 GB	7.2 GB	Runs comfortably	124K	97.3
GPTQ 4-bit	4.25	5.4 GB	7.2 GB	Runs comfortably	124K	97.3
MXFP4	4.25	5.4 GB	7.2 GB	Runs comfortably	124K	97.3
Q5_K_M	5.67	5.3 GB	7.1 GB	Runs comfortably	125K	99.1
Q5_K_S	5.52	5.2 GB	7.0 GB	Runs comfortably	126K	101
Q4_K_M	4.85	4.6 GB	6.4 GB	Runs comfortably	128K	112
Q4_K_S	4.58	4.4 GB	6.2 GB	Runs comfortably	128K	117
Q4_0	4.55	4.4 GB	6.2 GB	Runs comfortably	128K	118
IQ4_XS	4.25	4.1 GB	5.9 GB	Runs comfortably	128K	124
Q3_K_M	3.91	3.8 GB	5.7 GB	Runs comfortably	128K	131
IQ3_M	3.70	3.7 GB	5.5 GB	Runs comfortably	128K	137
IQ3_XXS	3.06	3.2 GB	5.0 GB	Runs comfortably	128K	155
Q2_K	2.63	2.8 GB	4.6 GB	Runs comfortably	128K	171
IQ2_XXS	2.06	2.3 GB	4.2 GB	Runs comfortably	128K	198
IQ1_M	1.75	2.1 GB	3.9 GB	Runs comfortably	128K	216

Q4_K_M at 8K context. Bandwidth sets the speed; capacity sets the ceiling.

Model	Params	Weights	Verdict	Max ctx	tok/s
DeepSeek-R1 671B-A37B	671B	379.0 GB	Won't fit	—	1.96
Qwen3 235B-A22B	235.1B	132.8 GB	Won't fit	—	3.36
gpt-oss 120B-A5.1B	116.8B	66.0 GB	Won't fit	—	17.1
Llama 4 Scout 109B-A17B	109B	61.7 GB	Won't fit	—	5.39
Qwen2.5 72B	72.7B	41.2 GB	Won't fit	—	1.59
Llama 3.1 70B	70.5B	40.0 GB	Won't fit	—	1.67
Mixtral 8x7B	46.7B	26.4 GB	Won't fit	—	16.4
Command R 35B	35.0B	20.1 GB	Won't fit	640	3.12
Yi-1.5 34B	34.4B	19.5 GB	Won't fit	6K	19.6
Qwen3 32B	32.8B	18.6 GB	Fits, but tight	9K	29.5
Qwen2.5-Coder 32B	32.8B	18.6 GB	Fits, but tight	9K	29.5
DeepSeek-R1-Distill-Qwen 32B	32.8B	18.6 GB	Fits, but tight	9K	29.5
Qwen3 30B-A3B	30.5B	17.3 GB	Runs comfortably	39K	117
Gemma 3 27B	27.4B	15.6 GB	Runs comfortably	57K	35.4
Mistral Small 24B	23.6B	13.4 GB	Runs comfortably	32K	41.3
gpt-oss 20B-A3.6B	20.9B	11.9 GB	Runs comfortably	128K	183
Qwen3 14B	14.8B	8.5 GB	Runs comfortably	80K	63.5
Phi-4 14B	14.7B	8.4 GB	Runs comfortably	16K	63.0
Mistral NeMo 12B	12.3B	7.0 GB	Runs comfortably	89K	75.3
Gemma 3 12B	12.2B	7.0 GB	Runs comfortably	128K	76.1
GLM-4 9B	9.4B	5.4 GB	Runs comfortably	128K	103
Qwen3 8B	8.2B	4.7 GB	Runs comfortably	115K	108
Llama 3.1 8B	8.0B	4.6 GB	Runs comfortably	128K	112
DeepSeek-R1-Distill-Llama 8B	8.0B	4.6 GB	Runs comfortably	128K	112
Qwen2.5 7B	7.6B	4.4 GB	Runs comfortably	128K	125
Mistral 7B v0.3	7.3B	4.1 GB	Runs comfortably	32K	124
Gemma 3 4B	4.3B	2.5 GB	Runs comfortably	128K	209
Qwen3 4B	4.0B	2.3 GB	Runs comfortably	128K	196
Llama 3.2 3B	3.2B	1.8 GB	Runs comfortably	128K	246
Llama 3.2 1B	1.2B	0.7 GB	Runs comfortably	128K	650