DGX Spark vs Mac Studio M4 Max 128 GB for local AI
Today the Mac Studio M4 Max 128 GB is cheaper: $4,899 against $7,799.98 for the DGX Spark. Per gigabyte of memory the Mac Studio M4 Max 128 GB wins, $38 against $60. For Qwen3.8 27B · Q4_K_M the Mac Studio M4 Max 128 GB is faster, about 20 against 11 tokens per second.
| NVIDIA DGX Spark | Apple Mac Studio M4 Max, 128 GB | |
|---|---|---|
| Lowest price today, delivered | $7,799.98 on eBay | $4,899 on eBay |
| Price per GB | $60/GB | $38/GB |
| Memory | 128 GB LPDDR5x unified | 128 GB LPDDR5x unified |
| Memory bandwidth | 273 GB/s | 546 GB/s |
| FP16 compute, dense | 125 TFLOPS | 18.4 TFLOPS |
| Reference models that fit at Q4_K_M | 18 of 28 | 18 of 28 |
| Power | 240 W | n/a |
| Form factor | Desktop box, 150 x 150 x 50 mm | Desktop box, 197 x 197 x 95 mm |
| Interface | unified memory | unified memory |
| Software platform | cuda | metal |
| Launched | 2025 | 2025 |
| Launch price | $3,999 | $3,499 |
Qwen3.8 27B · Q4_K_M by quantization
| Quant | DGX Spark | Mac Studio M4 Max 128 GB |
|---|---|---|
| FP16 | ~3.3 tok/s | ~6.1 tok/s |
| Q8_0 | ~6.2 tok/s | ~11 tok/s |
| Q6_K | ~8.0 tok/s | ~15 tok/s |
| Q5_K_M | ~9.3 tok/s | ~17 tok/s |
| Q4_K_M | ~11 tok/s | ~20 tok/s |
| Q3_K_M | ~13 tok/s | ~25 tok/s |
| IQ2_M | ~19 tok/s | ~36 tok/s |
Fit means the weights plus an 8K context cache fit in memory; speed is the estimated decode rate. Prices are the cheapest live eBay listing that passed our filters, item plus shipping. How we estimate.