Apple M5 Ultra vs NVIDIA DGX Spark

The M5 is the newer part: Apple M5, launched in 2026, against the DGX Spark's Grace Blackwell from 2025. Newer architectures typically add lower-precision formats and better throughput per watt, so check the precision rows your workload actually uses.

Memory is 512 GB against 128 GB, fed at 1.2 TB/s versus 273 GB/s. More memory per GPU means larger models fit before you have to shard across cards, which often matters more than raw TFLOPS for inference.

For LLM training and inference, weight the FP8 and FP16 rows and memory capacity most heavily. For scientific and HPC workloads, the FP64 row is the one to read.