This PCIe AI Accelerator Card Can Run 700B LLMs Locally With 384 GB Memory at Just 240W, Less Than Half the Power of RTX PRO 6000 Blackwell

Skymizer unveiled the HTX301, a PCIe AI accelerator based on its HyperThought platform for on-prem LLM inference.

The company says the card can run models up to 700B parameters using standard DRAM, with as much as 384 GB of memory and a 240W TDP.

If the claims hold up, the HTX301 could let enterprises run very large models on a single local server instead of relying on GPU clusters or cloud services.

Skymizer positioned the device at Computex alongside rivals such as NVIDIA RTX PRO 6000 Blackwell and AMD Instinct MI350P.