Company intelligence
RidgeScope
Turn more GPU hours into productive training
Sunnyvale, US
4 employees
4 public signals
About RidgeScope
RidgeScope finds where allocated GPUs stop producing useful work — idle jobs, hangs, stragglers, data starvation, checkpoint I/O, fabric bottlenecks, and runs that finish without learning. Then it shows what the evidence supports and the next action.
Verified activity
Signals from RidgeScope
4 published signals
Research & Knowledge
RidgeScope reports that GPT-6 Astra completed pretraining in roughly 90–110 days using more than 100,000 NVIDIA Grace Blackwell GPUs.
Reported by Evgeny Potapov
Research & Knowledge
RidgeScope reports that GPT-5.5, codenamed Spud, completed pretraining on March 24 at the Abilene campus.
Reported by Evgeny Potapov
Research & Knowledge
RidgeScope is experimenting with a pattern where larger models teach fleets of smaller models to perform individual tasks faster and more efficiently.
Reported by Evgeny Potapov
Products & Services
RidgeScope is building a GPU and ML training observability platform at the intersection of low-level hardware, distributed systems, performance engineering, machine learning, and foundation-model infrastructure.
Reported by Evgeny Potapov