50 or 100 times
“We're now driving 50 or 100 times more experiments versus what we were doing before on the Mk1 IPU systems”
As published on wandb.ai, October 2023. Captured by usedby on Oct 8, 2026.
What happened
Graphcore uses Weights & Biases to track and compare machine learning experiments run across multiple IPU-POD systems and deployment locations, replacing internal tools and spreadsheets. It serves as a central source of truth for diagnosing failures and comparing past and current runs while training large models like BERT.
Summary written by usedby from the source page, in English. The figures are those of Weights & Biases and Graphcore, not ours.
Weights & Biases allows them to scale up their experiments as they scale up the systems they're building. It lets them more easily discover what went wrong faster so that it can be fixed sooner.
We're now driving 50 or 100 times more experiments versus what we were doing before on the Mk1 IPU systems
What the story claims, and what we checked
We compared the story with its live page on Oct 8, 2026.
- The figure: 50 or 100 timesCheckedPrinted word for word on the page, near the name of Graphcore.
- The passage quoted aboveCheckedCopied word for word from the page, near the name of Graphcore.
- The publication dateCheckedRead from the page’s own metadata, never guessed.
- Graphcore uses Weights & BiasesCheckedConfirmed line. Latest check across sources: Oct 8, 2026.
- The result itselfNot checkedWe quote it; we did not measure it.




