What this render log message does mean?

[TomGore]

Often, I have this message among other render log lines:
"Warning num samples per thread reduced to 103003 rendering might be slower"

I render on GPU (next) with dual 2080 RTX sli/NVlink and with multi core processor Intel I9-7940X which is 14 cores and 28 threads. I use CPU+GPU1+GPU2 in performance tab.

How to get rid this message or rather how slower it can be? It appears both on Low gpu priority on or off, doesn't matter.

[a7az0th]

Hi ,

This message is intended for us developers only and you should not be concerned by it. We will remove it from future V-Ray versions as it only confuses users. It appears on heavier scenes that tightly occupy GPU memory and means we would've liked to have a bit extra memory. Feel free to ignore it.

[TomGore]

By the way - are the most recent builds supporting RTX / NVlink capabilities already? Octane did some tests and released first build 2018.2 where you can test RTX gains. These are amazing in case of heavier scenes - on average 3x faster, which in terms of money is like 3 cards less to purchase. I have dual RTX 2080 and right now I don't know if I already use RTX gain in Vray GPU or not? I've read on Vlado blog post that NVlink should be already working, but I would feel good to hear it from devs as well. Perhaps some line in render log like "Warning: RTX / Nvlink in use" could be nice way to tell this.

[a7az0th]

If you check the V-Ray log you will se info about NVLink usage. V-Ray has been supporting NVLink for a long time now and RTX is no exception. However for GeForce cards, NVidia requires you to enable SLI, or the NVLink is not recognized, so if you need the NVLink you might need to enable SLI.
On the subject of RT cores, currently there is no official V-Ray GPU build that supports RT Cores. We have an experimental one, but it still does not support all the features of V-Ray GPU. Out goal is to achieve full feature parity before releasing it, so that users don't have to worry what features work and where, but simply render
and get that nice speedup.

[TomGore]

Thanks for clarification - any rough ETA for RTX goodness?