Inference & Serving · advanced
WebGPU Compute
WGSL, compute kernels, matmul, CPU parity.
Mental model
WebGPU lets JavaScript run compute shaders on the GPU, which can speed up matrix math 10-100x in the browser. The real bottlenecks are moving data to and from the GPU and compiling the shaders — not the math itself. Scope: this card owns one backend — compute shaders, WGSL, and the transfer and compilation costs that dominate matmul in practice. The surrounding browser runtime is `ml-browser-runtime`; running outside the browser is `local-on-device-inference`.
How to study WebGPU Compute
Begin by restating the mental model in your own words, then connect it to a concrete system you have built or operated. Name the mechanism, the constraint it addresses, and the trade-off it introduces. Use Stanford CS336 — L5 GPUs & memory hierarchy (Spring 2025), WebGPU API (MDN) to check details, but close the source before writing your explanation. Retrieval is the learning step; rereading is only preparation.
Next, compare WebGPU Compute with the neighboring concepts in its roadmap. Ask what changes in correctness, latency, resource use, operability, and failure recovery. Complete WebGPU buffer upload size and preserve the command, input, output, and one failed attempt as evidence. Finish by explaining the idea without jargon to someone who has not studied the track.
Proof of understanding
- Explain the mechanism from first principles and identify the state it reads or changes.
- Give one situation where the concept is the right choice and one where it is not.
- Predict a realistic failure mode before running the drill, then compare the prediction with evidence.
- Connect the result to a roadmap or build artifact instead of treating the concept as isolated trivia.
Learn from primary sources
Practice and explain it back
WebGPU buffer upload size
Tensor 1024×1024 float32. Byte size? If maxStorageBufferBindingSize is 128MB, does one buffer hold the tensor?
Expected evidence: 4MB — fits easily; watch alignment and copy queue staging.
Open the interactive drill →Review prompts
- Your WebGPU matmul is far slower than expected. What are the two usual causes, and neither is the arithmetic?
Build evidence
Synthesize: Inference & Serving
Build a production mental model for inference engines, memory, kernels, routing, hardware, and serving economics. Produce one working system, benchmark, or evidence-backed design that integrates the path.
- Implements or precisely models the core mechanisms from all three milestones
- Includes at least one injected failure or adversarial case and demonstrates recovery
- Reports quality, latency, resource, reliability, or usability measurements relevant to the domain
- Ships a concise architecture note explaining decisions, trade-offs, and remaining risks
Synthesize: Multimodal & Spatial Computing
Connect vision, audio, generation, on-device intelligence, robotics, spatial interfaces, and HCI. Produce one working system, benchmark, or evidence-backed design that integrates the path.
- Implements or precisely models the core mechanisms from all three milestones
- Includes at least one injected failure or adversarial case and demonstrates recovery
- Reports quality, latency, resource, reliability, or usability measurements relevant to the domain
- Ships a concise architecture note explaining decisions, trade-offs, and remaining risks
Prerequisites
Related concepts
None assigned yet.