← AI Hashrate 中文

RTX 3090 24GB vs V100 32GB SXM2 for Local LLMs

The RTX 3090 24GB has about 4% more memory bandwidth (936 vs 900 GB/s). Batch-1 decode tracks bandwidth closely, so a model that fits both cards should decode roughly 4% faster on the RTX 3090 24GB. The V100 32GB SXM2 fits more of the catalog at Q4/8K — 43 vs 41 of 73 models (32 vs 24 GB VRAM). Largest model that fits the V100 32GB SXM2 but not the RTX 3090 24GB: Qwen3.6-35B-A3B-FP8 (36.0B). MSRP is $1,499 for the RTX 3090 24GB vs $10,000 for the V100 32GB SXM2. MSRP is a launch list price, not live retail — Amazon prices move; click through for the current price. Not retail-shoppable: V100 32GB SXM2 (datacenter-class; cloud rental is the realistic way to use it). In the Q4/8K head-to-head below (56 shared models), the RTX 3090 24GB posts the higher tok/s in 38 rows vs 18. Methodology.

Spec comparison

RTX 3090 24GBV100 32GB SXM2
VendorNVIDIANVIDIA
VRAM24 GB32 GB
Memory typeGDDR6XHBM2
Memory bandwidth936 GB/s900 GB/s
TDP350 W300 W
MSRP (list)$1,499$10,000
BuyMSI GeForce RTX 3090 Ventus 3X 24G OC · More listings · eBay · Newegg · B&H Photo

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 56 shared models

ModelParams (B)RTX 3090 24GB tok/sRTX 3090 24GB fitsV100 32GB SXM2 tok/sV100 32GB SXM2 fitsWinner
MiniCPM5-1B1.08931.4 est.Yes895.6 est.YesRTX 3090 24GB
Qwen3.5-2B2.0503.0 est.Yes483.6 est.YesRTX 3090 24GB
DeepSeek-R1-0528-Qwen3-8B-layer-mix-bpw-3.8-mlx2.3437.4 est.Yes420.5 est.YesRTX 3090 24GB
Qwen3.6-35B-A3B36.0335.3 est.Yes322.4 est.YesRTX 3090 24GB
Qwen3.5-35B-A3B34.7335.3 est.Yes322.4 est.YesRTX 3090 24GB
GLM-4.7-Flash31.2335.3 est.Yes322.4 est.YesRTX 3090 24GB
Qwen3-30B-A3B-Instruct-250730.5335.3 est.Yes322.4 est.YesRTX 3090 24GB
Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF1630.0335.3 est.Yes322.4 est.YesRTX 3090 24GB
Ornith-1.0-35B35.0335.3 est.Yes322.4 est.YesRTX 3090 24GB
qwen35b-a3b-fable-sft-abliterated34.7335.3 est.Yes322.4 est.YesRTX 3090 24GB
Llama-3.2-3B-Instruct3.2314.4 est.Yes302.3 est.YesRTX 3090 24GB
gpt-oss-20b21.5161.8 measuredYes268.7 est.YesV100 32GB SXM2
Qwen3-Coder-Next79.7156.9 est.No268.3 est.NoV100 32GB SXM2
gemma-4-26B-A4B-it25.2264.7 est.Yes254.5 est.YesRTX 3090 24GB
Qwen3-Next-80B-A3B-Instruct81.3151.0 est.No258.2 est.NoV100 32GB SXM2

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 3090 24GBV100 32GB SXM2
Catalog models that fit41 of 7343 of 73
Largest exclusive fitQwen3.6-35B-A3B-FP8 (36.0B)

RTX 3090 24GB details → · V100 32GB SXM2 details →