[Solved] Not correct rendering time estimation

[Bormax]

Hi
Corona 6 doesn't estimate remaining rendering time correctly. Here is my example.
Pretty regular interior scene, image size 1200x900, no caustics, adaptivity ON, no lightmix, the only extra element is CShading_albedo, all Performance settings set to default except 4k cache (actually it's the same is with UHD cache), noise limit 4%, no any other CPU load during the rendering.

- 1st rendering time estimation (elapsed + left = total)
0:01:17 + 0:11:34 = 0:12:51

- 2nd rendering time estimation (elapsed + left = total)
0:11:34 + 0:07:09 = 0:18:43

- 3rd rendering time estimation (elapsed + left = total)
0:22:00 + 0:00:00 = 0:22:00

- final result is 0:22:50, where denoising is only 6 sec

So real total rendering time is almost 2 times longer than estimated rendering time from the beginning. Larger picture - bigger this difference in minutes. This is really bad for time planning. I don't remember such a big difference between estimated and real rendering time in previous versions of Corona

Max 2014

[Frood]

Looks pretty standard to me. A typical Corona ETA graph based on noise looks like in the attached image ("ETA deviation" in orange). It shows how close the estimated remaining time has been at a given time, compared to the actual time the rendering took. 90% means: Corona thought that the image will be ready in 90% of the time it actually needed.

Your first ETA value is from pass 5. It's the first time noise was calculated. Corona is (and has been) always wrong here (way too optimistic). The second is mid-render and is in the usual range between 80 to 95%(<- at best) of actual render time, this is where Coronas ETA usually resides. Well and the last looks like this probably because you use the standard recalc interval of 5, also standard.


Good Luck

[Bormax]

Thank you vey much Frood for this clarification! Nice to know these details.

Of course I've noticed that 1st remaining time estimation was always optimistic, but for me it was not that much, if I remember right. I counted that the actual time will be about 40-50% longer than estimated, and it was more or less right. But now it takes about 2 times longer than estimated time to get ready picture

[Frood]

I only have a few graphs from v6 (dailies), but v6 had a better prediction after the first noise calculation as far as I can see compared to v5 (using the same scene). But all in all it's very scene dependent, so only re-rendering that specific scene with v5 would show if the prediction got worse.


Good Luck

[Bormax]

Actually I did some tests today with the same scene. The full scene which I rendered earlier and posted results above is lit by HDRi and some artificial lights inside the rooms. If I switch off all artificial lights and render this scene keeping only HDRi as a light source, then I get 14 min render time and 1st estimation is just a 1.5 min more optimistic than the actual render time. So it's about 10% mismatch, pretty close. Seems that more lights involved - more difference between estimated and actual render time

[Frood]

About the lights, that's an interesting point, will test betimes a similar scene v5 vs. v7 daily.


Good Luck


BTW: Please check your forum private messages occasionally ;]