The performance gains from Nvidia's Hopper to Blackwell GPUs come from increased size and power, not efficiency. This signals a potential scaling limit, creating an opportunity for radically new hardware primitives and neural network architectures beyond today's matrix-multiplication-centric models.

New AI models are designed to perform well on available, dominant hardware like NVIDIA's GPUs. This creates a self-reinforcing cycle where the incumbent hardware dictates which model architectures succeed, making it difficult for superior but incompatible chip designs to gain traction.

The MI300X's superior memory bandwidth and 192GB VRAM make it faster than H100s for non-FP8 dense transformers or MoE models. Quentin Anthony from Zyphra notes AMD's software has caught up, creating an under-appreciated arbitrage opportunity for teams willing to build on their stack.

Top-tier kernels like FlashAttention are co-designed with specific hardware (e.g., H100). This tight coupling makes waiting for future GPUs an impractical strategy. The competitive edge comes from maximizing the performance of available hardware now, even if it means rewriting kernels for each new generation.

Model architecture decisions directly impact inference performance. AI company Zyphra pre-selects target hardware and then chooses model parameters—such as a hidden dimension with many powers of two—to align with how GPUs split up workloads, maximizing efficiency from day one.

While many focus on compute metrics like FLOPS, the primary bottleneck for large AI models is memory bandwidth—the speed of loading weights into the GPU. This single metric is a better indicator of real-world performance from one GPU generation to the next than raw compute power.

GPUs were designed for graphics, not AI. It was a "twist of fate" that their massively parallel architecture suited AI workloads. Chips designed from scratch for AI would be much more efficient, opening the door for new startups to build better, more specialized hardware and challenge incumbents.

Instead of using high-level compilers like Triton, elite programmers design algorithms based on specific hardware properties (e.g., AMD's MI300X). This bottom-up approach ensures the code fully exploits the hardware's strengths, a level of control often lost through abstractions like Triton.

The popular PyTorch Profiler only shows the 'tip of the iceberg.' To achieve meaningful performance gains, engineers must move beyond it and analyze 50-60 low-level GPU metrics related to streaming multiprocessors, instruction pipelines, and specialized function units. Most of the PyTorch community stops too early.