Charles Frye

What every AI engineer needs to know about GPUs — Charles Frye, Modal

Jul 20, 2025 · 19:52

Charles Frye of Modal explains that AI engineers need to understand GPU hardware constraints to optimize inference, arguing GPUs embrace bandwidth over latency and that Tensor Cores for low-precision matrix-matrix multiplication are the key resource. He describes how GPUs achieve 16,000+ parallel threads per cycle on H100, and notes Patterson’s Law: bandwidth improves at the square of latency. The main insight: arithmetic intensity favors N² operations per N memory loads, so matrix-matrix operations are efficient while matrix-vector is wasteful. Frye demonstrates that running a small 8B model 1,000 times on the same prompt matches GPT-4 quality, and that multi-token prediction and multi-sample query become nearly free because Tensor Cores handle expanded batches as matrix-matrix multiplications. He recommends using smaller models that fit on a single GPU and scaling via multiple generations.

How fast are LLM inference engines anyway? — Charles Frye, Modal

Jun 27, 2025 · 16:07

Charles Frye presents benchmarks from hundreds of runs on Modal comparing open-source inference engines VLM, SGLang, and TensorRTLM across models like Qwen 3 and Gemma 27B, arguing open weights models have caught up to proprietary ones, making self-hosting viable. He shows, for example, that Qwen 3 (MoE) on VLM achieves ~1 request/sec with 128 input tokens and 1024 output tokens, while switching to a RAG-like workload (1024 in, 128 out) yields a 4x throughput improvement. Frye warns that optimizing for context over reasoning can improve latency without sacrificing quality, and notes that the engines' out-of-the-box performance varies by model—e.g., SGLang underperforms VLM on Gemma due to less optimization. He also highlights the gap between prefill (parallel) and decode (autoregressive) speeds, which a rationalist would expect from transformer architecture. The benchmarks, available at modal.com/llmalmanac, aim to help engineers choose hardware and engines, with contributions welcome for optimized configs like TensorRTLM's knobs.

AI Engineering 201: The Rest of the Owl

Nov 8, 2023 · 56:57

Charles Frye, instructor of the Full Stack LLM Bootcamp, presents essential patterns for building language user interfaces (LUIs), arguing that while RAG chatbots are the 'to-do list app' of AI engineering, structured outputs via function calling (e.g., OpenAI's JSON schema, Instructor library) improve robustness, and agents with memory (like generative agents or Voyager in Minecraft) represent the true AI frontier. He emphasizes the need for hybrid search combining vector and keyword retrieval (citing Vespa, Postgres, Redis), and warns that monitoring and evaluation are the hardest engineering challenges: monitoring user behavior, latency quantiles (especially 99th percentile), and costs must be paired with observability tools like Honeycomb or Gantry, while evaluation often requires iterated decomposition or using LLMs as evaluators (GPT-4 as 90th percentile crowd worker). The episode concludes that shipping to learn — starting with production data to generate tests — is the dominant engineering mindset, and that the field is still filling in the gaps between inference and full product value.

[Workshop] AI Engineering 201: Inference

Nov 7, 2023 · 1:43:16

Charles Frye, instructor of the Full Stack LLM Bootcamp, leads a workshop on AI engineering inference, focusing on the build-versus-buy decision between proprietary and open models. He argues that proprietary models like OpenAI's GPT-4 and Anthropic's Claude are currently more capable but expensive, while open models like LLaMA 2 are less capable but offer hackability, though they may catch up if capabilities requirements saturate. Frye covers inference on end-user devices, noting that running models locally avoids network latency but faces tight memory and power constraints—e.g., a 7B parameter model requires 14 GB, too large for phone RAM. He explains inference-as-a-service (e.g., OpenAI, Replicate) versus self-serving on cloud GPUs or serverless platforms like Modal, highlighting that memory bandwidth is the bottleneck: GPUs have 1.5 TB/s memory bandwidth vs 312 TFLOPS compute, so batching is crucial for throughput. Frye also discusses inference arithmetic, custom silicon like TPUs which offer ~30% better efficiency but not drastic gains, and containerization challenges for GPU workloads.