ASUS announces the UGen Intel Arc Pro B70 with 32GB of GDDR6 ECC for local inference, and cites 43.4 tokens a second for a dual-card setup. It gives no price or ship date.
NVIDIA reports its Nemotron 3.5 Transcribe ranks first among open models on Hugging Face's English speech-recognition board, with 4B of 30B parameters active. Quantized RTX Spark builds are coming.
Weights that fit in 32GB still left the RTX 5090 waiting half an hour for a first token, and the fix was a different inference engine rather than a different card.
NVIDIA set October for RTX Spark, split the N1X into a 32GB and a 128GB tier, and shipped a router that spreads agent work across the PCs already on a home network.
