We put 2× NVIDIA RTX PRO 6000 Blackwell GPUs through 435 benchmark runs across 12 model families, ranging from 9B to 122B parameters, in more than 85 configurations. Token throughput, concurrency, KV cache limitations, cost per million tokens... We put everything into one
High-bandwith dedicated servers with ultra-low latencies & zero packet loss in 67 locations globally.

