← AI Hashrate

Gaudi 3 128GB vs RTX 4090D 24GB (China) for Local LLMs

The Gaudi 3 128GB has about 225% more memory bandwidth (3,276 vs 1,008 GB/s). Batch-1 decode tracks bandwidth closely, so a model that fits both cards should decode roughly 225% faster on the Gaudi 3 128GB. The Gaudi 3 128GB fits more of the catalog at Q4/8K — 47 vs 33 of 61 models (128 vs 24 GB VRAM). Largest model that fits the Gaudi 3 128GB but not the RTX 4090D 24GB (China): DeepSeek-V4-Flash (158.1B). MSRP is $12,500 for the Gaudi 3 128GB vs $1,599 for the RTX 4090D 24GB (China). MSRP is a launch list price, not live retail — Amazon prices move; click through for the current price. Not retail-shoppable: Gaudi 3 128GB, RTX 4090D 24GB (China) (datacenter-class; cloud rental is the realistic way to use it). In the Q4/8K head-to-head below (47 shared models), the Gaudi 3 128GB posts the higher tok/s in 47 rows vs 0. Methodology.

Spec comparison

Gaudi 3 128GBRTX 4090D 24GB (China)
VendorIntelNVIDIA
VRAM128 GB24 GB
Memory typeHBM2eGDDR6X
Memory bandwidth3,276 GB/s1,008 GB/s
TDP600 W425 W
MSRP (list)$12,500$1,599
Buy——

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)Gaudi 3 128GB tok/sGaudi 3 128GB fitsRTX 4090D 24GB (China) tok/sRTX 4090D 24GB (China) fitsWinner
Qwen3-0.6B0.82605.9 est.Yes801.8 est.YesGaudi 3 128GB
gemma-3-1b-it1.02084.7 est.Yes641.5 est.YesGaudi 3 128GB
Llama-3.2-1B-Instruct1.21737.3 est.Yes534.5 est.YesGaudi 3 128GB
Qwen3-1.7B2.01042.4 est.Yes320.7 est.YesGaudi 3 128GB
Qwen3-30B-A3B30.5694.9 est.Yes213.8 est.YesGaudi 3 128GB
Qwen3-Coder-30B-A3B-Instruct30.5694.9 est.Yes213.8 est.YesGaudi 3 128GB
GLM-4.7-Flash31.2694.9 est.Yes213.8 est.YesGaudi 3 128GB
Qwen3-30B-A3B-Instruct-250730.5694.9 est.Yes213.8 est.YesGaudi 3 128GB
Qwen3-Next-80B-A3B-Instruct81.3694.9 est.Yes164.0 est.NoGaudi 3 128GB
Qwen3-Coder-Next79.7694.9 est.Yes170.4 est.NoGaudi 3 128GB
Llama-3.2-3B-Instruct3.2651.5 est.Yes200.5 est.YesGaudi 3 128GB
gpt-oss-20b21.5579.1 est.Yes178.2 est.YesGaudi 3 128GB
Phi-4-mini-instruct3.8548.6 est.Yes168.8 est.YesGaudi 3 128GB
Phi-4-mini-reasoning3.8548.6 est.Yes168.8 est.YesGaudi 3 128GB
gemma-3-4b-it4.3534.5 est.Yes164.5 est.YesGaudi 3 128GB

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

Gaudi 3 128GBRTX 4090D 24GB (China)
Catalog models that fit47 of 6133 of 61
Largest exclusive fitDeepSeek-V4-Flash (158.1B)—

Gaudi 3 128GB details → · RTX 4090D 24GB (China) details →