RTX 4070 Super 12GB vs RTX 5090 32GB for Local LLMs

The RTX 5090 32GB has about 256% more memory bandwidth (1,792 vs 504 GB/s). Batch-1 decode tracks bandwidth closely, so a model that fits both cards should decode roughly 256% faster on the RTX 5090 32GB. The RTX 5090 32GB fits more of the catalog at Q4/8K — 38 vs 18 of 61 models (32 vs 12 GB VRAM). Largest model that fits the RTX 5090 32GB but not the RTX 4070 Super 12GB: Mixtral-8x7B-Instruct-v0.1 (46.7B). MSRP is $599 for the RTX 4070 Super 12GB vs $1,999 for the RTX 5090 32GB. MSRP is a launch list price, not live retail — Amazon prices move; click through for the current price. In the Q4/8K head-to-head below (40 shared models), the RTX 5090 32GB posts the higher tok/s in 40 rows vs 0. Methodology.

Spec comparison

	RTX 4070 Super 12GB	RTX 5090 32GB
Vendor	NVIDIA	NVIDIA
VRAM	12 GB	32 GB
Memory type	GDDR6X	GDDR7
Memory bandwidth	504 GB/s	1,792 GB/s
TDP	220 W	575 W
MSRP (list)	$599	$1,999
Buy	Search Amazon (affiliate)	Search Amazon (affiliate)

MSRP is the launch list price, not live retail. Buy links are Amazon affiliate links — the price after you click is the live price.

Q4 / 8K head-to-head — top 15 of 40 shared models

Model	Params (B)	RTX 4070 Super 12GB tok/s	RTX 4070 Super 12GB fits	RTX 5090 32GB tok/s	RTX 5090 32GB fits	Winner
Qwen3-0.6B	0.8	400.9 est.	Yes	1425.5 est.	Yes	RTX 5090 32GB
gemma-3-1b-it	1.0	320.7 est.	Yes	1140.4 est.	Yes	RTX 5090 32GB
Llama-3.2-1B-Instruct	1.2	267.3 est.	Yes	950.3 est.	Yes	RTX 5090 32GB
Qwen3-1.7B	2.0	160.4 est.	Yes	570.2 est.	Yes	RTX 5090 32GB
Qwen3-Coder-Next	79.7	21.3 est.	No	538.6 est.	No	RTX 5090 32GB
Qwen3-Next-80B-A3B-Instruct	81.3	20.5 est.	No	518.3 est.	No	RTX 5090 32GB
Qwen3-30B-A3B	30.5	130.2 est.	No	380.1 est.	Yes	RTX 5090 32GB
Qwen3-Coder-30B-A3B-Instruct	30.5	130.2 est.	No	380.1 est.	Yes	RTX 5090 32GB
GLM-4.7-Flash	31.2	125.0 est.	No	380.1 est.	Yes	RTX 5090 32GB
Qwen3-30B-A3B-Instruct-2507	30.5	130.2 est.	No	380.1 est.	Yes	RTX 5090 32GB
Llama-3.2-3B-Instruct	3.2	100.2 est.	Yes	356.4 est.	Yes	RTX 5090 32GB
Phi-4-mini-instruct	3.8	84.4 est.	Yes	300.1 est.	Yes	RTX 5090 32GB
Phi-4-mini-reasoning	3.8	84.4 est.	Yes	300.1 est.	Yes	RTX 5090 32GB
gemma-3-4b-it	4.3	82.2 est.	Yes	292.4 est.	Yes	RTX 5090 32GB
Qwen3-4B-Instruct-2507	4.0	80.2 est.	Yes	285.1 est.	Yes	RTX 5090 32GB

est. = bandwidth-based estimate; measured = curated real benchmark that overrides the estimate. Fits = weights + KV(8K) + 1 GB overhead ≤ 95% of VRAM.

Fit coverage at Q4 / 8K

	RTX 4070 Super 12GB	RTX 5090 32GB
Catalog models that fit	18 of 61	38 of 61
Largest exclusive fit	—	Mixtral-8x7B-Instruct-v0.1 (46.7B)

RTX 4070 Super 12GB details → · RTX 5090 32GB details →