I will post first chance I get. In my initial test I built a simple cornel box and had it render to a 100 passes. With HT turned off using 1.6 it took 1:18 With HT turned on it took 1:57...
But with the 1.3 Benchmark with HT turned on it renders in :40 and with HT turned off I was getting :57...
I'll post some screenshots ASAP. I have a suspicion that there is some performance loss with many softwares once they see over 64 cores... Terragen 4 for example renders fast with HT off...
- Shawn
There is a truth to this, because after 64 'threads', Windows will create groups (so 80 threads will becomes two groups), and softwares not optimized for groups, will only see a single one. HT off is band-aid for this, because in this exact case, 40 physical cores, is more than 40 virtual threads.
But Corona sees groups perfectly and I myself use few 80 thread nodes ( E5 2698v4).
With benchmarks it's good to remember not everything is perfectly multi-threaded, so the longer they are the better. With renders going for 1 hour+ I see almost perfectly linear scaling.
