DGX Spark (GH200 480GB) vs Gaudi 3 128GB for Local LLMs

The Gaudi 3 128GB has about 555% more memory bandwidth (3,276 vs 500 GB/s). Batch-1 decode tracks bandwidth closely, so a model that fits both cards should decode roughly 555% faster on the Gaudi 3 128GB. The DGX Spark (GH200 480GB) fits more of the catalog at Q4/8K — 66 vs 53 of 69 models (480 vs 128 GB VRAM). Largest model that fits the DGX Spark (GH200 480GB) but not the Gaudi 3 128GB: DeepSeek-V4-Pro (861.6B). MSRP is $3,000 for the DGX Spark (GH200 480GB) vs $12,500 for the Gaudi 3 128GB. MSRP is a launch list price, not live retail — Amazon prices move; click through for the current price. Not retail-shoppable: Gaudi 3 128GB (datacenter-class; cloud rental is the realistic way to use it). In the Q4/8K head-to-head below (66 shared models), the Gaudi 3 128GB posts the higher tok/s in 66 rows vs 0. Methodology.

Spec comparison

	DGX Spark (GH200 480GB)	Gaudi 3 128GB
Vendor	NVIDIA	Intel
VRAM	480 GB	128 GB
Memory type	LPDDR5X	HBM2e
Memory bandwidth	500 GB/s	3,276 GB/s
TDP	170 W	600 W
MSRP (list)	$3,000	$12,500
Buy	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 66 shared models

Model	Params (B)	DGX Spark (GH200 480GB) tok/s	DGX Spark (GH200 480GB) fits	Gaudi 3 128GB tok/s	Gaudi 3 128GB fits	Winner
MiniCPM5-1B	1.08	294.6 est.	Yes	1930.3 est.	Yes	Gaudi 3 128GB
Qwen3.5-2B	2.0	159.1 est.	Yes	1042.4 est.	Yes	Gaudi 3 128GB
DeepSeek-R1-0528-Qwen3-8B-layer-mix-bpw-3.8-mlx	2.3	138.3 est.	Yes	906.4 est.	Yes	Gaudi 3 128GB
Qwen3.6-35B-A3B	36.0	106.1 est.	Yes	694.9 est.	Yes	Gaudi 3 128GB
Qwen3.5-35B-A3B	34.7	106.1 est.	Yes	694.9 est.	Yes	Gaudi 3 128GB
GLM-4.7-Flash	31.2	106.1 est.	Yes	694.9 est.	Yes	Gaudi 3 128GB
Qwen3-30B-A3B-Instruct-2507	30.5	106.1 est.	Yes	694.9 est.	Yes	Gaudi 3 128GB
Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16	30.0	106.1 est.	Yes	694.9 est.	Yes	Gaudi 3 128GB
Qwen3-Coder-Next	79.7	106.1 est.	Yes	694.9 est.	Yes	Gaudi 3 128GB
Ornith-1.0-35B	35.0	106.1 est.	Yes	694.9 est.	Yes	Gaudi 3 128GB
Qwen3-Next-80B-A3B-Instruct	81.3	106.1 est.	Yes	694.9 est.	Yes	Gaudi 3 128GB
qwen35b-a3b-fable-sft-abliterated	34.7	106.1 est.	Yes	694.9 est.	Yes	Gaudi 3 128GB
Llama-3.2-3B-Instruct	3.2	99.4 est.	Yes	651.5 est.	Yes	Gaudi 3 128GB
MiniMax-M2.7	228.7	31.8 est.	Yes	588.1 est.	No	Gaudi 3 128GB
gpt-oss-20b	21.5	88.4 est.	Yes	579.1 est.	Yes	Gaudi 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

	DGX Spark (GH200 480GB)	Gaudi 3 128GB
Catalog models that fit	66 of 69	53 of 69
Largest exclusive fit	DeepSeek-V4-Pro (861.6B)	—

DGX Spark (GH200 480GB) details → · Gaudi 3 128GB details →