Скачать или смотреть DevStandup #16 - Bottlenecks: Where Performance Really Breaks.

🎙 Episode #16 of DevStandup features John Boero — a global consultant with two decades of experience across information security, infrastructure performance, DevOps, and solutions architecture. Today, John lives as a traveling nomad, offering contract services while actively contributing to open-source communities like Fedora and Llama.cpp.

This time, John brings a session that hits close to home for anyone working with modern AI systems:

“Bottlenecks: Where Performance Really Breaks.”

As AI workloads scale and hardware limits are pushed harder than ever, teams often underestimate the physical and architectural choke points hidden beneath their models. John draws from years of field experience to break down the common — and sometimes surprising — performance killers that engineers encounter.

He’ll explore:

⚙️ Storage bottlenecks and the I/O traps
Why high-throughput compute collapses under slow disks, and how to spot early warning signs.

🌐 Network overhead and packetization pitfalls
How latency, MTU mismatches, and microbursts quietly destroy distributed AI performance.

🌡️ Thermal limits and the realities of Ohm’s law
Why your hardware may be throttling long before you think it is — and how to mitigate it.

🧮 The hidden penalty of AI quantization
What teams overlook when compressing models, and how accuracy vs. speed trade-offs really play out.

🧠 Memory bandwidth vs. compute for training and inference
Why memory, not GPUs, is often the true limiting factor in modern AI pipelines.

John grounds the entire talk in practical experience — offering tactical insights rather than theory, and giving listeners a clearer picture of how to diagnose and avoid real-world bottlenecks.

Perfect for DevOps engineers, SREs, ML practitioners, and performance-minded developers who want a clearer roadmap to high-efficiency AI infrastructure.

📣 Interested in being our next speaker?
Share your journey and become part of the DevStandup community!
👉 https://forms.gle/kMoNXyX4k8f3gbgT7

🔔 Like, comment, and subscribe for more deep dives into AI systems, DevOps practices, and performance engineering.