Home › GPUs & Macs › NVIDIA DGX Spark (GB10) vs Apple M4 Max (40-core GPU)
NVIDIA DGX Spark (GB10) vs Apple M4 Max (40-core GPU) for local LLMs
Across 4 tests measured on both, the Apple M4 Max (40-core GPU) is faster in 4 and generates a median 1.46× the tokens/s of the NVIDIA DGX Spark (GB10).
On Llama 2 7B (Q4_0, llama-bench) the NVIDIA DGX Spark (GB10) generates 60.4 tokens/s and the Apple M4 Max (40-core GPU) 83.1 tokens/s (measured).
Same model, same quant, same test
| Model | Quant / test | DGX Spark (GB10) tok/s | M4 Max (40-core GPU) tok/s | Faster | Sources and run details |
|---|---|---|---|---|---|
| Llama 2 7B | Q4_0 llama-bench | 60.4 Vulkan | 83.1 Metal | M4 Max (40-core GPU) +38% | llama.cpp #10879 DGX Spark (GB10) · llama.cpp #4167 M4 Max (40-core GPU) |
| Llama 3.1 8B Instruct | Q4_K_M LocalScore | 34.2 llamafile | 52.9 llamafile | M4 Max (40-core GPU) +55% | LocalScore DGX Spark (GB10) · LocalScore M4 Max (40-core GPU) |
| Llama 3.2 1B Instruct | Q4_K_M LocalScore | 132 llamafile | 180 llamafile | M4 Max (40-core GPU) +36% | LocalScore DGX Spark (GB10) · LocalScore M4 Max (40-core GPU) |
| Qwen2.5 14B Instruct | Q4_K_M LocalScore | 18.3 llamafile | 28.9 llamafile | M4 Max (40-core GPU) +58% | LocalScore DGX Spark (GB10) · LocalScore M4 Max (40-core GPU) |
Generation speed in tokens/s, best measured backend for each chip (shown under the number). Backends can differ between the two chips; check them before drawing conclusions. Measured results only, nothing estimated.
Specs
| NVIDIA DGX Spark (GB10) | Apple M4 Max (40-core GPU) | |
|---|---|---|
| Memory | 128 GB | Unified memory |
| Memory bandwidth | 273 GB/s | 546 GB/s |
All results for the NVIDIA DGX Spark (GB10) · All results for the Apple M4 Max (40-core GPU) · More comparisons