GPU Training Journey: From Small Experiments to Cogni
At Titans Lab, we develop advanced synthetic intelligence engines optimized for local and global applications. Over the past few months, we have scaled our GPU compute from small single-GPU experiments to multi-GPU clusters for larger SI engines. This ongoing investment in GPUs allows us to push the boundaries of engine performance, reliability, and autonomy while still thinking like a startup. The underlying training spend in South African Rand (R) is already very high, even at these scales.
The visual below shows approximate GPU usage bands for Titans Lab core systems and Cogni engines, from small single-GPU runs through to larger GPU clusters, with very high rand costs behind the scenes.
Scale: Bars are drawn on an approximate 0–64+ GPU range.
Titans Lab bands: From focused single-GPU experiments through to multi-GPU synthetic intelligence stacks used in production research.
Cogni band: Represents large-scale assistant engines and long-running fine-tuning cycles for Cogni deployments, typically running on roughly 32–64 GPUs for several weeks on startup-scale clusters.
Cape Town, South Africa
Reg. No 2025/312191/07