Head to head

RTX 4090 vs NVIDIA DGX Spark (GB10)

The NVIDIA DGX Spark (GB10) holds more — 128 GB against 24 GB — which decides what you can load at all. The RTX 4090 has more bandwidth at 1008 GB/s, which decides how fast tokens come out once a model fits.

Model (Q4_K_M, 8K)RTX 4090tok/sNVIDIA DGX Spark (GB10)tok/s
Llama 3.1 8B Runs comfortably 116 Runs comfortably 35.1
Llama 3.1 70B Won't fit 1.60 Runs comfortably 4.37
Llama 3.2 3B Runs comfortably 254 Runs comfortably 78.1
Llama 3.2 1B Runs comfortably 669 Runs comfortably 209
Llama 4 Scout 109B-A17B Won't fit 5.18 Runs comfortably 14.8
Qwen3 8B Runs comfortably 112 Runs comfortably 33.9
Qwen3 14B Runs comfortably 65.5 Runs comfortably 19.8
Qwen3 32B Fits, but tight 30.5 Runs comfortably 9.17
Qwen3 4B Runs comfortably 202 Runs comfortably 62.2
Qwen3 30B-A3B Runs comfortably 118 Runs comfortably 53.5
Qwen3 235B-A22B Won't fit 3.23 Won't fit 6.20
Qwen2.5-Coder 32B Fits, but tight 30.5 Runs comfortably 9.17
Qwen2.5 7B Runs comfortably 128 Runs comfortably 38.9
Qwen2.5 72B Won't fit 1.53 Runs comfortably 4.24
Gemma 3 4B Runs comfortably 215 Runs comfortably 66.4
Gemma 3 12B Runs comfortably 78.5 Runs comfortably 23.8

Specifications

RTX 4090NVIDIA DGX Spark (GB10)
Memory24 GB128 GB
Bandwidth1008 GB/s273 GB/s
FP16 compute165 TFLOPS125 TFLOPS
Launch price$1,599$3,999
ArchitectureAdaBlackwell