Back to GeForce RTX 2060

• Card will have 3GB, 4GB and 6GB versions
• Also will have older GDD5X and new GDDR6 versions

Total mess.

Price of RTX 2060 is rumored to be $400

To be short - go and buy ex mining RX 580 8GB.

More info on decoders/encoders

https://developer.nvidia.com/nvidia-video-codec-sdk

Davinci by design (node based, tend to add constant features with slow code) is worst possible candidate for such tests.

True. However Node based compositors can (and some do) use GPU acceleration. Fusion however does not, using mainly CPU code (or openCL which mostly falls back to CPU).

Most image transforms should be calculated with shaders. Shaders can handle high bit depth color, and are super simple to implement.

Up until now, main issue with shaders was that raw shader code was not compiled, hence any user could easily be discovered. Vulkan allows compiled shaders, so I suspect that more proprietary softwares will embrace this technology now.

Well, guys focus on 4K with artificial setups, like adding 3 slow plugins and 4 tracking areas.

And they directly are interested in certain tests outcome (with different outcomes they'll go burst), hence all the parameter choosing.

Davinci by design (node based, tend to add constant features with slow code) is worst possible candidate for such tests.

In reality if you write NLE proper 1050 Ti works perfect with 4K at 60fps with one tracking area and color grading+sharpness and other stuff.

On DLSS thingy

All results in one place

Well, bad

Ray tracing is amazing.

Funny numbers: Those INT32 numbers are total BS! (*)

Looks like they just feed 32bit integers into the 32bit floatingpoint unit, which is no problem (FP is more complicated), just to get a new entry into this "feature list" and "show" that the new cards a sooo much better. I bet the old ones can do that as well.

(* If the new cards can do INT32 and FP32 in parallel, each running at 13-16 TFLOPS at the same time, than I'm sorry and my post is BS... but I highly doubt it.)