For comparison, the RTX 5090 has a memory bandwidth of 1,792 GB/s. The GX10 will likely be quite disappointing in terms of tokens per second and therefore not well suited for real-time interaction with a state-of-the-art large language model or as a coding assistant.