← AI Hashrate

L40S 48GB vs RTX 5090 32GB for Local LLMs

The RTX 5090 32GB has about 107% more memory bandwidth (1,792 vs 864 GB/s). Batch-1 decode tracks bandwidth closely, so a model that fits both cards should decode roughly 107% faster on the RTX 5090 32GB. The L40S 48GB fits more of the catalog at Q4/8K — 39 vs 38 of 61 models (48 vs 32 GB VRAM). Largest model that fits the L40S 48GB but not the RTX 5090 32GB: Qwen3-Coder-Next (79.7B). MSRP is $8,000 for the L40S 48GB 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 (47 shared models), the RTX 5090 32GB posts the higher tok/s in 39 rows vs 8. Methodology.

Spec comparison

L40S 48GBRTX 5090 32GB
VendorNVIDIANVIDIA
VRAM48 GB32 GB
Memory typeGDDR6GDDR7
Memory bandwidth864 GB/s1,792 GB/s
TDP350 W575 W
MSRP (list)$8,000$1,999
BuySearch 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 47 shared models

ModelParams (B)L40S 48GB tok/sL40S 48GB fitsRTX 5090 32GB tok/sRTX 5090 32GB fitsWinner
Qwen3-0.6B0.8687.3 est.Yes1425.5 est.YesRTX 5090 32GB
gemma-3-1b-it1.0549.8 est.Yes1140.4 est.YesRTX 5090 32GB
Llama-3.2-1B-Instruct1.2458.2 est.Yes950.3 est.YesRTX 5090 32GB
Qwen3-1.7B2.0274.9 est.Yes570.2 est.YesRTX 5090 32GB
Qwen3-Next-80B-A3B-Instruct81.3562.3 est.No518.3 est.NoL40S 48GB
Qwen3-Coder-Next79.7183.3 est.Yes538.6 est.NoRTX 5090 32GB
Qwen3-30B-A3B30.5183.3 est.Yes380.1 est.YesRTX 5090 32GB
Qwen3-Coder-30B-A3B-Instruct30.5183.3 est.Yes380.1 est.YesRTX 5090 32GB
GLM-4.7-Flash31.2183.3 est.Yes380.1 est.YesRTX 5090 32GB
Qwen3-30B-A3B-Instruct-250730.5183.3 est.Yes380.1 est.YesRTX 5090 32GB
Llama-3.2-3B-Instruct3.2171.8 est.Yes356.4 est.YesRTX 5090 32GB
Phi-4-mini-instruct3.8144.7 est.Yes300.1 est.YesRTX 5090 32GB
Phi-4-mini-reasoning3.8144.7 est.Yes300.1 est.YesRTX 5090 32GB
gemma-3-4b-it4.3141.0 est.Yes292.4 est.YesRTX 5090 32GB
Qwen3-4B-Instruct-25074.0137.5 est.Yes285.1 est.YesRTX 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

L40S 48GBRTX 5090 32GB
Catalog models that fit39 of 6138 of 61
Largest exclusive fitQwen3-Coder-Next (79.7B)—

L40S 48GB details → · RTX 5090 32GB details →