Mac Studio M5 Ultra 80-core GPU vs NVIDIA DGX Spark
Prices, stock and measured speed side by side. Last checked 2026-10-16.
Verdict
NVIDIA DGX Spark is cheaper in USD by $6,100 (130%). In EUR, NVIDIA DGX Spark is cheaper by €8,353 (at least one EUR price is converted). On gpt-oss-120b, Mac Studio M5 Ultra 80-core GPU is faster (130 vs 58.72 tok/s; sources and engines differ, see below). Per dollar, NVIDIA DGX Spark gives more large-model speed (12.5 vs 12.0 tok/s per $1k). Mac Studio M5 Ultra 80-core GPU has more memory (256GB), which matters for larger models and long contexts.
Generated from the numbers below; no hands-on testing.
| Mac Studio M5 Ultra 80-core GPU, 256GB / 1TB | NVIDIA DGX Spark (Founders Edition) | |
|---|---|---|
| Price (USD) | $10,799 | $4,699 |
| Price (EUR) | €12,529 | ≈€4,176 |
| Memory | 256GB | 128GB |
| Stock | not tracked shop | not tracked shop |
| gpt-oss-120b tok/s | 130 | 58.72 |
| gpt-oss-120b tok/s per $1k | 12.0 | 12.5 |
| gpt-oss-120b tok/s per €1k | 10.4 | 14.1 |
| Llama-2 7B tok/s | 179.1 | 57.21 |
| Llama-2 7B tok/s per $1k | 16.6 | 12.2 |
| Llama-2 7B tok/s per €1k | 14.3 | 13.7 |
| Stores | US store · EU (Ireland) | US store · EU store |
Benchmark sources
- Mac Studio M5 Ultra 80-core GPU: large-model source (LM Studio, spoken in video (not llama-bench)) · small-model source
- NVIDIA DGX Spark: large-model source (llama.cpp CUDA, llama-bench tg32, b7946) · small-model source
Large: gpt-oss-120b MXFP4 (59 GiB weights), text generation tok/s at empty context, single request. Small: llama.cpp llama-bench, Llama-2 7B Q4_0, tg128 (text generation tok/s); best of FA on/off as posted.