Cold Start Benchmark: 244 GiB Model Loads in 154 Seconds

QuixiAI · x · 2026-07-30

QuixiAI shared performance metrics for cold starting a massive 244 GiB model. The test reveals a load time of approximately 154 seconds on specific hardware setups, noted as being even faster than llama.cpp's loading times.

Original post →

More from Infra

Infra channel →