Rays weekly update

Tuesday, July 30th 2019

Software engineering

  • Generally, everything works. Task price prediction works, payments work, receipts work, and actual rendering works too! Service is running on closed Beta.
  • Price prediction is not accurate so far and it is delaying the service launch. We've spent the last few weeks tuning this mechanism. It required us to generate massive educational datasets (actual render time vs lower quality evaluation render) a couple of times since we changed the criteria and fixed bugs. Previously we used to run evaluation render with lower sample rate and resolution, letting the user change resolution and sample rate after the evaluation is done. Now we only use a lower sample rate, all other settings are the same as final render. It is still not very accurate, but we work on it.
  • This week we plan to update price prediction (frontend and backend).
  • There is a bottleneck on the prediction phase. Because we use only one GPU for prediction we do not want to waste other GPUs time, so we look for specific machines with exactly 1 GPU onboard, which are rare on Sonm. This will be solved by adding more machines of this kind and increasing rental fare for them.
  • We expect to have a reliable prediction by the end of this week, so once we have it, we will have a final overall look and set a launch date with a countdown timer. This may give us a launch date in the mid August.

Marketing & PR

  • We launched a test ads campaign on facebook. There are very surprising results, which we will not make public in details, but here are some insights:
    • we managed to get some leads below $1 cost. Actually, our business plan allows for an orders of magnitude higher lead cost based on expected ARPU. Our practice results so far are probably inaccurate, so we will keep experimenting. Anyway, we now have more information to update our unit economy - there were many discussions in this field recently.
    • there was an issue with an accusation, that Rays is related to cryptocurrency, which has no basis. Rays is a traditional business, making payments with credit cards and providing a competitive render service. Who could even think and predict this?
    • different countries have predictably different conditions, costs and reactions.
  • Waiting for the render comparison video author to publish an English version so we can share it.
  • Preparing a post from our workshop at CGevent.


  • Writing letters to different entities.
  • Sent some merchandise by mail.


  • Preparing a frame agreement with our hardware providers to compensate for rented equipment.


  • Two new software developers joined and started working.
  • There is a total of 5 pure software developers working fulltime on Rays by now (not counting other technical folks, which are traditionally mistakenly all referred as "devs" - architects, analysts, designers, devops, etc).