Theres still a lot of info missing. Performance at FP32, FP16, FP8, FP4 etc are not going to be the same. 160GB is great, but its compute levels are closer to a Jetson Orin AGX with only 64GB 275 TOPS compute, which wont keep up with the 1200GB/s at higher FP rates.
System Price Memory Memory Bandwidth Published AI Compute Power Large-LLM Position
Xiaomi AI Cube Prototype TBA Up to 160GB 1,220 GB/s O100 near-memory O3: 200 TOPS NPU; O100 compute undisclosed 150W sustained 🔥 Potentially extremely strong
Jetson AGX Thor 128GB $5,499 128GB unified 273 GB/s 2.07 PFLOPS FP4 sparse 130W 🥇 Highest known compute
ASUS Ascent GX10 $3,999 128GB unified 273 GB/s 1.0 PFLOP FP4 sparse ~140W GB10 🥇 Best value
NVIDIA DGX Spark $4,699 128GB unified 273 GB/s 1.0 PFLOP FP4 sparse 140W GB10 Excellent turnkey option