GPU group

LLM fine-tuning.

GPUs with 80GB+ VRAM — comfortable for LoRA / QLoRA on 70B-class models.

GPU	Tier	Architecture	VRAM	Providers	AI models	From
Nvidia H200	datacenter	Hopper	141GB	10 →	—	$0.46/hr	Compare →
Nvidia A100	datacenter	Ampere	80GB	14 →	—	$0.48/hr	Compare →
Nvidia H100	datacenter	Hopper	80GB	20 →	—	$0.80/hr	Compare →
Nvidia RTX PRO 6000 Blackwell	workstation	Blackwell	96GB	1 →	—	$1.53/hr	Compare →
AMD AMD MI300X	datacenter	CDNA 3	192GB	3 →	—	$1.95/hr	Compare →
Nvidia GH200	datacenter	Hopper	96GB	1 →	—	$2.29/hr	Compare →
Nvidia B200	datacenter	Blackwell	192GB	8 →	—	$2.79/hr	Compare →
Nvidia B300	datacenter	Blackwell	288GB	2 →	—	$5.38/hr	Compare →