⏱ Hardware benchmarks · Experiential playback · Free

Test-drive local LLM
hardware before you buy

Benchmark tables answer "what is faster?" LapTime answers "what does faster feel like?" — prompt ingest time, TTFT, and decode speed across hardware configs and models.

Open LapTime → GitHub
Live at
laptime.run
Free to use · No account required
Open app →
Features
What it does
🎬
Experiential playback
Watch prompt ingest, TTFT, and decode play out in real time — not just a number in a table. You feel the difference between hardware tiers.
⚖️
Head-to-head comparison
Pick two hardware configs and a model. See both run side by side so the tradeoff is concrete, not abstract.
🤖
Model-aware
Benchmarks are per-model, not just per-hardware. A 70B model on a GB10 isn't the same as a 7B model on the same hardware.
🔗
Affiliate links
Hardware cards link to purchase options. Benchmarks inform the buy decision at the moment it's being made.
Why it exists
The problem with benchmark tables
✗ Benchmark tables
·Abstract numbers — 147 tok/s means nothing without context
·No model-specific data for the model you actually want to run
·Can't compare the end-to-end experience — just peak throughput
·TTFT and ingest speed are usually missing entirely
✓ LapTime
·Watch it happen — ingest, TTFT, and decode play out in real time
·Per-model benchmarks for the workload you're actually running
·Side-by-side comparison so the tradeoff is concrete
·Full time-to-first-token visibility — often the bottleneck