Rays weekly update
16 jul 2019
- Orchestrator. Released new version with many bugs fixed.
- Render time prediction. We tried to figure out a reasonable formula to predict full work render based on low sample rate low-resolution test runs. Human formula lacked precision and the results were bad. But hey! We used Jupyter notebook to train a model on 4 thousand of render jobs and the first results we had with machine learning has beat our human prediction formulae! The ML-based prediction has the precision of render price (based on render time prediction and fixed prices) within 5 cents (standard deviation). Humans are losing again.
- There will be a minimum payment of roughly 50 cents based on payment system constraints. The change will be probably compensated with bonus credits.
- Backend work:
- Include new prediction model in working code.
- Finish fair resource sharing.
- Use only machines with 4+ GPUs.
- Collect worker logs. Required for correct failures resolution and fair bans.
- If any frame being calculated twice longer than predicted - terminate the work and restart. We now have precise enough prediction to do this.
- New flow finished and released on internal staging.
- Not finished: pay with bonus and mixed payment with bonus and cash.
- Receipt machine will be called from frontend as part of payment API calls.
- A couple of patches have to be applied.
- Full workflow cycle not yet passed. Currently, it fails on the test payment scheme with test payment cards. Debugging, after fixed will be open for internal testing.
- Keep working to release an updated closed beta on JulyВ 24th.
- Keep working to release publicly on July 31.
Marketing & PR
- Preparing to publish Rays CGevent presentation in English. Translation done, polishing things us. Manin.
- The launch countdown timer will be initiated once both conditions met: 10 days before launch and after we will feel confident (after full workflow cycle will be passed).
- Rays website gallery (portfolio) is being finished, to be released this week. Polygalov, Chedrik.
- Product hunt preparation started. Ivanova.
- Test facebook ad campaign warming up. Ivanova.
- Artem to discuss the possible future tariff plans.
- Preparing a template agreement for Rays-Suppliers pair: (1) legal computing power usage and (2) NDA framework. A shortlist of suppliers for priority signature is prepared.
- A long and painful upgrade is finished. A Rays typical machine is now equipped with 32 Gb RAM and has at least 4 GPUs. A partial upgrade to Intel i5 is already done as well.
- New offer issued, the developer is expected to join by late summer.