NVIDIA's local GLM checkpoint needs two GPUs to run

NVIDIA's FP4-Local checkpoint of Z.ai's GLM-5.3-Flash, released October 5th, requires two RTX PRO 6000 Blackwell GPUs or two DGX Spark systems in every configuration NVIDIA has validated, using about 166 GiB of VRAM. NVIDIA RTX Spark promoted the checkpoint on October 20th alongside a Qwen counterpart that occupies about 96 GiB on disk and is validated on a single DGX Spark or RTX PRO 6000 Blackwell GPU with offloading. The gap means running GLM-5.3-Flash locally still demands a multi-GPU or multi-system setup.

NVIDIA's local GLM checkpoint needs two GPUs to run Released October 5th, the FP4 version of Z.ai's GLM-5.3-Flash uses about 166 GiB of VRAM; Qwen's counterpart occupies about 96 GiB on disk. By Ryan Merket https://runtimewire.com/author/ryan-merket · Published Primary source: X https://x.com/NVIDIARTXSpark/status/2112635102069248023 Why it matters NVIDIA's Qwen checkpoint is validated on a single DGX Spark or RTX PRO 6000 Blackwell GPU with offloading, while GLM's listed configurations require two GPUs or two DGX Spark systems. Local inference can still require a multi-GPU or multi-system setup. NVIDIA's FP4-Local version of Z.ai's GLM-5.3-Flash https://runtimewire.com/models/z-ai/glm-5.3-flash:batch requires two RTX PRO 6000 Blackwell GPUs or two DGX Spark systems in the configurations NVIDIA has validated. The checkpoint went up on October 5th; NVIDIA RTX Spark's October 20th post https://x.com/NVIDIARTXSpark/status/2112635102069248023 promoted it alongside a smaller Qwen checkpoint as a way to run large models locally.… When you purchase through links in our articles, we may earn a small commission https://runtimewire.com/affiliate-monetization-standards . This doesn’t affect our editorial independence.