← AI Hashrate 中文

DGX Spark (GH200 480GB) vs RTX 3060 12GB for Local LLMs

The DGX Spark (GH200 480GB) has about 39% more memory bandwidth (500 vs 360 GB/s). Batch-1 decode tracks bandwidth closely, so a model that fits both cards should decode roughly 39% faster on the DGX Spark (GH200 480GB). The DGX Spark (GH200 480GB) fits more of the catalog at Q4/8K — 66 vs 18 of 69 models (480 vs 12 GB VRAM). Largest model that fits the DGX Spark (GH200 480GB) but not the RTX 3060 12GB: DeepSeek-V4-Pro (861.6B). MSRP is $3,000 for the DGX Spark (GH200 480GB) vs $329 for the RTX 3060 12GB. 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 (41 shared models), the DGX Spark (GH200 480GB) posts the higher tok/s in 32 rows vs 9. Methodology.

Spec comparison

DGX Spark (GH200 480GB)RTX 3060 12GB
VendorNVIDIANVIDIA
VRAM480 GB12 GB
Memory typeLPDDR5XGDDR6
Memory bandwidth500 GB/s360 GB/s
TDP170 W170 W
MSRP (list)$3,000$329
BuySearch Amazon (affiliate)MSI GeForce RTX 3060 Ventus 2X 12G OC (affiliate) · More listings

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 41 shared models

ModelParams (B)DGX Spark (GH200 480GB) tok/sDGX Spark (GH200 480GB) fitsRTX 3060 12GB tok/sRTX 3060 12GB fitsWinner
MiniCPM5-1B1.08294.6 est.Yes212.1 est.YesDGX Spark (GH200 480GB)
Qwen3.5-2B2.0159.1 est.Yes114.5 est.YesDGX Spark (GH200 480GB)
DeepSeek-R1-0528-Qwen3-8B-layer-mix-bpw-3.8-mlx2.3138.3 est.Yes99.6 est.YesDGX Spark (GH200 480GB)
Llama-3.2-3B-Instruct3.299.4 est.Yes122.9 measuredYesRTX 3060 12GB
Qwen3.6-35B-A3B36.0106.1 est.Yes71.8 est.NoDGX Spark (GH200 480GB)
Qwen3.5-35B-A3B34.7106.1 est.Yes77.0 est.NoDGX Spark (GH200 480GB)
GLM-4.7-Flash31.2106.1 est.Yes88.9 est.NoDGX Spark (GH200 480GB)
Qwen3-30B-A3B-Instruct-250730.5106.1 est.Yes92.6 est.NoDGX Spark (GH200 480GB)
Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF1630.0106.1 est.Yes101.8 est.NoDGX Spark (GH200 480GB)
Qwen3-Coder-Next79.7106.1 est.Yes15.5 est.NoDGX Spark (GH200 480GB)
Ornith-1.0-35B35.0106.1 est.Yes72.0 est.NoDGX Spark (GH200 480GB)
Qwen3-Next-80B-A3B-Instruct81.3106.1 est.Yes14.9 est.NoDGX Spark (GH200 480GB)
qwen35b-a3b-fable-sft-abliterated34.7106.1 est.Yes73.1 est.NoDGX Spark (GH200 480GB)
gemma-4-26B-A4B-it25.283.7 est.Yes101.2 est.NoRTX 3060 12GB
gpt-oss-20b21.588.4 est.Yes84.0 measuredNoDGX Spark (GH200 480GB)

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

DGX Spark (GH200 480GB)RTX 3060 12GB
Catalog models that fit66 of 6918 of 69
Largest exclusive fitDeepSeek-V4-Pro (861.6B)—

DGX Spark (GH200 480GB) details → · RTX 3060 12GB details →