← AI Hashrate 中文

Arc Pro B70 32GB vs L40S 48GB for Local LLMs

The L40S 48GB has about 42% more memory bandwidth (864 vs 608 GB/s). Batch-1 decode tracks bandwidth closely, so a model that fits both cards should decode roughly 42% faster on the L40S 48GB. The L40S 48GB fits more of the catalog at Q4/8K — 41 vs 39 of 69 models (48 vs 32 GB VRAM). Largest model that fits the L40S 48GB but not the Arc Pro B70 32GB: Qwen3-Next-80B-A3B-Instruct (81.3B). MSRP is $999 for the Arc Pro B70 32GB vs $8,000 for the L40S 48GB. 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 (53 shared models), the L40S 48GB posts the higher tok/s in 52 rows vs 1. Methodology.

Spec comparison

Arc Pro B70 32GBL40S 48GB
VendorIntelNVIDIA
VRAM32 GB48 GB
Memory typeGDDR6GDDR6
Memory bandwidth608 GB/s864 GB/s
TDP225 W350 W
MSRP (list)$999$8,000
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 53 shared models

ModelParams (B)Arc Pro B70 32GB tok/sArc Pro B70 32GB fitsL40S 48GB tok/sL40S 48GB fitsWinner
MiniCPM5-1B1.08358.2 est.Yes509.1 est.YesL40S 48GB
Qwen3.5-2B2.0193.5 est.Yes274.9 est.YesL40S 48GB
DeepSeek-R1-0528-Qwen3-8B-layer-mix-bpw-3.8-mlx2.3168.2 est.Yes239.1 est.YesL40S 48GB
Qwen3-Coder-Next79.7186.8 est.No183.3 est.YesArc Pro B70 32GB
Qwen3.6-35B-A3B36.0129.0 est.Yes183.3 est.YesL40S 48GB
Qwen3.5-35B-A3B34.7129.0 est.Yes183.3 est.YesL40S 48GB
GLM-4.7-Flash31.2129.0 est.Yes183.3 est.YesL40S 48GB
Qwen3-30B-A3B-Instruct-250730.5129.0 est.Yes183.3 est.YesL40S 48GB
Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF1630.0129.0 est.Yes183.3 est.YesL40S 48GB
Ornith-1.0-35B35.0129.0 est.Yes183.3 est.YesL40S 48GB
Qwen3-Next-80B-A3B-Instruct81.3179.7 est.No183.3 est.YesL40S 48GB
qwen35b-a3b-fable-sft-abliterated34.7129.0 est.Yes183.3 est.YesL40S 48GB
Llama-3.2-3B-Instruct3.2120.9 est.Yes171.8 est.YesL40S 48GB
gpt-oss-20b21.5107.5 est.Yes152.7 est.YesL40S 48GB
gpt-oss-120b120.447.6 est.No152.2 est.NoL40S 48GB

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

Arc Pro B70 32GBL40S 48GB
Catalog models that fit39 of 6941 of 69
Largest exclusive fitQwen3-Next-80B-A3B-Instruct (81.3B)

Arc Pro B70 32GB details → · L40S 48GB details →